perf(vk): the renderer's threads rewritten: one render thread, the presenter on every PC, parts that end at the game's own pass ends #1763

Merged
HumanBean merged 5 commits from HumanBean/Longshot:perf/thread-redesign into main 2026-11-20 05:47:20 +00:00
Owner

Summary

The renderer's threads, rewritten as one design for every PC, with no old path kept:

  • One render thread records every command buffer. The recorder thread (which handed the game's thread's calls over 20-25 times a frame, each a sleep and a wake-up) and the parallel replay (segment workers) are gone, about 3,100 lines. The parallel replay lost on every PC it was measured on with its workers on (RTX 3090 / Ryzen 9 7900X -10% to -25%, Windows RTX 3070 -10% to -35% with an access violation, Linux RX 7900 XTX -3% to -34%): a Tracy capture showed the frame is the game's own draw building on its job workers, which two segment workers made 28% slower by sharing its cores. Threaded rendering's Auto is the render thread from 4 cores.
  • The presenter thread on every PC. Each present is made on a thread of its own once its copy is done, on the presents' own queue (the replay's, taking turns with its submits, or the Present queue row's), with NVIDIA Reflex's present id. On Linux with NVIDIA's driver a present blocks ~4.5 ms for the copy, which no longer holds the render thread. (From #1702, now closed.)
  • A submission's part ends where the game has ended its render pass, from 120 draws, at 512 at the latest, instead of every 240 draws wherever that fell: no render pass is cut in two, so its targets are not written out and read back. In the Market this beat the fixed count in all 9 paired rounds (walking +15%).
  • Readbacks: the culling camera block (0x3980) is read without its wait on every PC; on the Steam Deck and handhelds the newest submission's results too, #238's +20% there (from #1725, closed with this).
  • Barriers: a draw pass after only draw passes waits for them in its fragment stages; full barriers with nothing recorded since the last are left out.
  • Diagnostics: RX_VK_GPU_TIMELINE=1 (each recording's submit and GPU span, the jobs, the game's fence waits, the presents), RX_VK_PASS_TIMING=3 (each command outside a named span timed as glue), and the log's new textures by kind.

Read first: gpu/vk/vk_replay.cpp (maybe_split, presenter_loop, present_thread_on, readback_late_wanted), gpu/vk/threaded_rendering.h, gpu/vk/vk_texture.cpp (pass_start_barrier, full_barrier), docs/dev/reference/threads.md.

Testing

Tested: CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X: the previous head (a2cc7b5, the same code before this rebase onto main's docs and tools revamp) in the game through Kokiri Forest, Hyrule Field and Castle Town Market, 3 interleaved rounds against main plus a 15 s Market walk, no crash, no Vulkan errors, pictures as main's (the differences are fireflies, Navi and the upscaler's noise). The lab (RX 580 / i7-9700K, RADV, 3 rounds): even with main where the GPU is 85-94% busy, the same draws per frame. The rebased head (21b606e) builds on Linux (dev.py vk) and passes a Windows llvm-mingw syntax check of every changed renderer file; it has not run in the game yet here (main's MKWVK_API_VERSION 35 needs a new base). mise run test: not run on this PC (the maintainer's rule); CI runs it.

Still to test:

  • Every platform (windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, steamdeck, amd+igpu): the three areas against main, expecting fps at least main's and no crash (@Beanstalk).
  • linux+amd and windows+amd: the Market standing, where the previous head measured -24.5% (RX 7900 XTX) and -14.5% (RX 9070 XT), with each piece turned back in turn (RX_VK_PART_MIN=240 RX_VK_PART_MAX=240, RX_VK_PASS_BARRIER=full, RX_VK_READBACK_LATE=0), expecting one of them to recover it or none, which points at the presenter thread (@Beanstalk).
  • steamdeck: the frame log's flash count with the newest submission's readbacks late (the exposure meter is among them), expecting none (@Beanstalk).
  • Linux, RTX 3090: the rebased head in the game, expecting the same as a2cc7b5 (by hand, once the base is built).

Measurements

Verdict: faster on NVIDIA and in most scenes on fast AMD cards, even where the GPU is the limit, and slower in the Market standing on fast AMD cards (under investigation, above).

Setup: CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X; 2560x1440, DLSS Quality, uncapped, Reflex on; each area 15 s standing, 5 s walking, 20 s turning; the renderer built alone on one base (main at 9ea65c7 and this PR at a2cc7b5), a warm-up each, 3 rounds interleaved on the PC's timing queue; fps from the renderer's present counts, the median of the rounds.

fps Before: main @ 9ea65c7 After: this PR @ a2cc7b5
Kokiri Forest still / walking / turning 69.4 / 73.2 / 75.1 73.5 / 79.6 / 79.5 (+6% / +9% / +6%)
Hyrule Field still / walking / turning 94.3 / 92.9 / 93.8 99.5 / 100.8 / 100.2 (+6% / +8% / +7%)
Castle Town Market still / walking (15 s) / turning 79.7 / 76.0 / 68.6 82.6 / 78.2 / 70.6 (+4% / +3% / +3%)
Frames over 50 ms, all areas 27 24

The Test Farm on a2cc7b5 (one report each, main against this PR): RX 7900 XTX (Linux) Kokiri +14% / +17%, Hyrule Field +14% / +30%, Market +8% turning and -24.5% standing; RX 9070 XT (Windows) Kokiri +11% / +14%, Hyrule Field -2% / -5%, Market +5% turning and -14.5% standing; Radeon 760M (Linux) within its noise. The switches this adds or changes ship at the defaults these runs chose (RX_VK_PART_MIN 120, RX_VK_PART_MAX 512: a fixed 30, 60, 120, 960 or no parts were each slower than these). HOS_PROFILE_DIR and RX_VK_ASYNC are not new: their rows changed (RX_VK_GPU_TIMELINE writes into HOS_PROFILE_DIR; RX_VK_ASYNC's default says the one 4-core tier), and their defaults are as before. The numbers were taken on main 9ea65c7; this head is on main 068fb8eb, whose changes to the renderer since are comments and the version files, nothing the frame runs, so they still hold (the Test Farm measures this head against main again).

Risks

What could break: the Market standing on fast AMD cards (measured, above); a present problem on a driver not yet run with the presenter thread on (Intel); flashes on the Deck from the newest submission's readbacks read late. Turning it off: RX_VK_PART_MIN=240 RX_VK_PART_MAX=240 (the old parts), RX_VK_PASS_BARRIER=full, RX_VK_READBACK_LATE=0; the presenter thread and the removed threads have no switch, so a revert is the way back. What a player notices: nothing in their files; RX_VK_RECORDER*, RX_VK_PARALLEL*, RX_VK_PRESENT_THREAD and RX_VK_CHUNK in a longshot_env.txt do nothing now (their rows are GONE).

Checklist

Ready for review

  • One topic, on current main; the title is a Conventional Commit.
  • Tested on my own PC, in the game for a change the game runs, as Testing says. (The previous head, a2cc7b5, in the game; the rebased head not yet: it needs a new base for main's MKWVK_API_VERSION 35.)
  • mise run test passes (any failure that also fails on main is named). (Not run on this PC; CI runs it.)
  • Measured against main, or Measurements says why there is nothing to measure.
  • No game files, build output or hand-edited generated files; third_party/ untouched; nothing GPL-incompatible.

When it applies

  • A switch added or changed: it ships at the default the measurements chose, with its row in gpu/vk/switches.def.
  • A check under tests/ that reads changed source is updated, not weakened.
  • Host code builds on Windows (llvm-mingw) and Linux, with UTF-8 paths. (Windows: a syntax check of every changed renderer file.)
  • Docs updated.
  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.

Ready to merge

  • Everything in Still to test has run, its results in Tested.
  • Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers").
  • Approved in review, every comment answered.
  • CI passes on the head; on current main, measured again or a comment says why the numbers still hold.

🤖 Generated with Claude Code

https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP

## Summary The renderer's threads, rewritten as one design for every PC, with no old path kept: - **One render thread records every command buffer.** The recorder thread (which handed the game's thread's calls over 20-25 times a frame, each a sleep and a wake-up) and the parallel replay (segment workers) are gone, about 3,100 lines. The parallel replay lost on every PC it was measured on with its workers on (RTX 3090 / Ryzen 9 7900X -10% to -25%, Windows RTX 3070 -10% to -35% with an access violation, Linux RX 7900 XTX -3% to -34%): a Tracy capture showed the frame is the game's own draw building on its job workers, which two segment workers made 28% slower by sharing its cores. Threaded rendering's Auto is the render thread from 4 cores. - **The presenter thread on every PC.** Each present is made on a thread of its own once its copy is done, on the presents' own queue (the replay's, taking turns with its submits, or the Present queue row's), with NVIDIA Reflex's present id. On Linux with NVIDIA's driver a present blocks ~4.5 ms for the copy, which no longer holds the render thread. (From #1702, now closed.) - **A submission's part ends where the game has ended its render pass**, from 120 draws, at 512 at the latest, instead of every 240 draws wherever that fell: no render pass is cut in two, so its targets are not written out and read back. In the Market this beat the fixed count in all 9 paired rounds (walking +15%). - **Readbacks:** the culling camera block (0x3980) is read without its wait on every PC; on the Steam Deck and handhelds the newest submission's results too, #238's +20% there (from #1725, closed with this). - **Barriers:** a draw pass after only draw passes waits for them in its fragment stages; full barriers with nothing recorded since the last are left out. - **Diagnostics:** `RX_VK_GPU_TIMELINE=1` (each recording's submit and GPU span, the jobs, the game's fence waits, the presents), `RX_VK_PASS_TIMING=3` (each command outside a named span timed as glue), and the log's new textures by kind. Read first: `gpu/vk/vk_replay.cpp` (`maybe_split`, `presenter_loop`, `present_thread_on`, `readback_late_wanted`), `gpu/vk/threaded_rendering.h`, `gpu/vk/vk_texture.cpp` (`pass_start_barrier`, `full_barrier`), `docs/dev/reference/threads.md`. ## Testing **Tested:** CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X: the previous head (a2cc7b5, the same code before this rebase onto main's docs and tools revamp) in the game through Kokiri Forest, Hyrule Field and Castle Town Market, 3 interleaved rounds against main plus a 15 s Market walk, no crash, no Vulkan errors, pictures as main's (the differences are fireflies, Navi and the upscaler's noise). The lab (RX 580 / i7-9700K, RADV, 3 rounds): even with main where the GPU is 85-94% busy, the same draws per frame. The rebased head (21b606e) builds on Linux (`dev.py vk`) and passes a Windows llvm-mingw syntax check of every changed renderer file; it has not run in the game yet here (main's MKWVK_API_VERSION 35 needs a new base). `mise run test`: not run on this PC (the maintainer's rule); CI runs it. **Still to test:** - Every platform (windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, steamdeck, amd+igpu): the three areas against main, expecting fps at least main's and no crash (@Beanstalk). - linux+amd and windows+amd: the Market standing, where the previous head measured -24.5% (RX 7900 XTX) and -14.5% (RX 9070 XT), with each piece turned back in turn (`RX_VK_PART_MIN=240 RX_VK_PART_MAX=240`, `RX_VK_PASS_BARRIER=full`, `RX_VK_READBACK_LATE=0`), expecting one of them to recover it or none, which points at the presenter thread (@Beanstalk). - steamdeck: the frame log's flash count with the newest submission's readbacks late (the exposure meter is among them), expecting none (@Beanstalk). - Linux, RTX 3090: the rebased head in the game, expecting the same as a2cc7b5 (by hand, once the base is built). ## Measurements **Verdict:** faster on NVIDIA and in most scenes on fast AMD cards, even where the GPU is the limit, and slower in the Market standing on fast AMD cards (under investigation, above). Setup: CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X; 2560x1440, DLSS Quality, uncapped, Reflex on; each area 15 s standing, 5 s walking, 20 s turning; the renderer built alone on one base (main at 9ea65c7 and this PR at a2cc7b5), a warm-up each, 3 rounds interleaved on the PC's timing queue; fps from the renderer's present counts, the median of the rounds. | fps | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | Kokiri Forest still / walking / turning | 69.4 / 73.2 / 75.1 | 73.5 / 79.6 / 79.5 (+6% / +9% / +6%) | | Hyrule Field still / walking / turning | 94.3 / 92.9 / 93.8 | 99.5 / 100.8 / 100.2 (+6% / +8% / +7%) | | Castle Town Market still / walking (15 s) / turning | 79.7 / 76.0 / 68.6 | 82.6 / 78.2 / 70.6 (+4% / +3% / +3%) | | Frames over 50 ms, all areas | 27 | 24 | The Test Farm on a2cc7b5 (one report each, main against this PR): RX 7900 XTX (Linux) Kokiri +14% / +17%, Hyrule Field +14% / +30%, Market +8% turning and -24.5% standing; RX 9070 XT (Windows) Kokiri +11% / +14%, Hyrule Field -2% / -5%, Market +5% turning and -14.5% standing; Radeon 760M (Linux) within its noise. The switches this adds or changes ship at the defaults these runs chose (`RX_VK_PART_MIN` 120, `RX_VK_PART_MAX` 512: a fixed 30, 60, 120, 960 or no parts were each slower than these). `HOS_PROFILE_DIR` and `RX_VK_ASYNC` are not new: their rows changed (`RX_VK_GPU_TIMELINE` writes into `HOS_PROFILE_DIR`; `RX_VK_ASYNC`'s default says the one 4-core tier), and their defaults are as before. The numbers were taken on main 9ea65c7; this head is on main 068fb8eb, whose changes to the renderer since are comments and the version files, nothing the frame runs, so they still hold (the Test Farm measures this head against main again). ## Risks What could break: the Market standing on fast AMD cards (measured, above); a present problem on a driver not yet run with the presenter thread on (Intel); flashes on the Deck from the newest submission's readbacks read late. Turning it off: `RX_VK_PART_MIN=240 RX_VK_PART_MAX=240` (the old parts), `RX_VK_PASS_BARRIER=full`, `RX_VK_READBACK_LATE=0`; the presenter thread and the removed threads have no switch, so a revert is the way back. What a player notices: nothing in their files; `RX_VK_RECORDER*`, `RX_VK_PARALLEL*`, `RX_VK_PRESENT_THREAD` and `RX_VK_CHUNK` in a `longshot_env.txt` do nothing now (their rows are GONE). ## Checklist ### Ready for review - [x] One topic, on current `main`; the title is a Conventional Commit. - [ ] Tested on my own PC, in the game for a change the game runs, as Testing says. (The previous head, a2cc7b5, in the game; the rebased head not yet: it needs a new base for main's MKWVK_API_VERSION 35.) - [ ] `mise run test` passes (any failure that also fails on `main` is named). (Not run on this PC; CI runs it.) - [x] Measured against `main`, or Measurements says why there is nothing to measure. - [x] No game files, build output or hand-edited generated files; `third_party/` untouched; nothing GPL-incompatible. ### When it applies - [x] A switch added or changed: it ships at the default the measurements chose, with its row in `gpu/vk/switches.def`. - [x] A check under `tests/` that reads changed source is updated, not weakened. - [x] Host code builds on Windows (llvm-mingw) and Linux, with UTF-8 paths. (Windows: a syntax check of every changed renderer file.) - [x] Docs updated. - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. ### Ready to merge - [ ] Everything in Still to test has run, its results in Tested. - [ ] Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers"). - [ ] Approved in review, every comment answered. - [ ] CI passes on the head; on current `main`, measured again or a comment says why the numbers still hold. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP
(From #1702.) present_out hands a present carrying Reflex's present id to the presenter thread too: present_hand_off
takes the id, presenter_loop presents with it and marks PRESENT_END (vk_reflex.cpp reflex_present_end), and the game's
thread clears its presenting state at the hand-off (reflex_present_handed_off). Before, Reflex on kept every present on the
game's thread, so DLSS players, who have Reflex on, never had the presenter thread.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The presenter thread no longer has a queue of its own. A second queue of the replay's family time-shared the GPU with
the replay (presents 2-4 ms later, docs/design.md section 13 rule 5), and a compute family's needed the swapchain
shared between families, the setup the Present queue row lost the GPU with on an RTX 5080 and 4090. It now waits for
the frame's copy, then presents on the presents' own queue (the replay's, or the Present queue row's), holding that
queue's lock for the call alone: the copy is done, so the call does not wait for the GPU (NVIDIA's driver on Linux
waited ~4.5 ms in it). With that, it is on everywhere; RX_VK_PRESENT_THREAD is gone, VK_KHR_present_wait is enabled
wherever the driver has it, and presenter_queue.h and the device policy's presenter_queue go.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The recorder thread made the replay's Vulkan calls one at a time, handed over call by call: it woke 20-25 times a frame,
and on an i7-9700K those short sleeps let the CPU's clock control slow every core (#1745), while kept spinning it cost a
Ryzen 9 7900X 3-11%. It is gone, with vk_recorder.h and its switches. The parallel replay takes its place by default:
a big command buffer's segments (160 draws and dispatches or more) are recorded by segment workers, each into a command
buffer of its own, and submitted in order (replay_handle, as RX_VK_PARALLEL did). A worker wakes a few times a frame.

- How many: the Threaded rendering row (threaded_rendering.h workers_for: the cores left once the game's main thread and
  the render thread have one each, 2-6, from 6 cores); RX_VK_PARALLEL over it.
- Calibrated in play (par_calibrate_present): once big command buffers come, the count alternates between the ceiling,
  half of it and none, 120 presents each, six rounds; the lowest median frame time is kept and logged.
  RX_VK_PAR_CALIBRATE=0: the ceiling.
- The black frames that kept the parallel replay off: each worker now notes its own recording part
  (memory_recording_part_here); before, every worker noted the last segment's, so a CPU read could wait for a part before
  the one that read the page in place.
- The work count is the recording thread's own (vk_work.h), so full_barrier's dedupe stays on with the workers.
- What still turns off where segments may run (par_ceiling >= 2), as before with RX_VK_PARALLEL: the capture's late fix,
  the early readbacks, DLAA before the bloom and the faded meshes' rule.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
A worker on the core the host keeps for the render thread took turns with it, while the render thread waits for the
workers and submits each segment as it is done; and a segment on an efficiency core holds the segments after it, which
are submitted in order. Windows: the fastest EfficiencyClass's logical processors (fastest_kind_cpus, as the render thread
may use); elsewhere the process's set; minus the reserved core where one is kept.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
Every draw pass began with a full barrier (ALL_COMMANDS to ALL_COMMANDS). Now a pass records it only where something
else came since the last full barrier on the command buffer: a dispatch, a copy, a fill or update, a blit or resolve, a
whole image's clear, an upscaler library's dispatches (vk_work.h note_other_work), any of which a pass's vertex fetch or
shaders may read. Where only draw passes came, every earlier write is made visible to the pass's fragment-side stages (the
source stays every stage) and its vertex, task and mesh work overlaps the passes before it, as on the console, which orders
a pass's fragments after the pass before, not its geometry; a pass whose geometry reads an earlier pass's output has the
game's own barrier between (full). The old opt-in rule (RX_VK_PASS_BARRIER=game) relaxed 21 of ~400 barriers a frame in
the Market, as it needed nothing at all between the passes; RX_VK_PASS_BARRIER=full is the old behaviour, a test.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The game reads its culling camera block (0x3980 bytes) right after submitting the work that fills it, so on the console
it cannot count on that work being done; the reads of it were 40 to 45% of the readback wait on a Steam Deck (2.4 ms on a
heavy view, 4.5 to 4.9 in Hyrule Field). It is now read as guest memory holds it, the newest write-back already made,
on every PC (readback_late.h default_sizes). Every other read waits as before: the exposure meter, whose late read is the
white flashes, and Hyrule Field's culling tables, not shown safe to read late. RX_VK_READBACK_LATE=0: every read waits.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The recorder thread's checks go with it (test_recorder_sync, test_recorder_own_idle and their fixtures); its source
checks of vk_work.h live on in the new test_pass_barrier.py, which also compiles full_barrier and pass_start_barrier over a
spy driver (first pass full, pass after pass fragment-side, other work between full, another command buffer full, a
worker's work never another thread's, RX_VK_PASS_BARRIER=full). test_threaded_rendering checks workers_for and the tiers
at every core count and the readers of the row and RX_VK_PARALLEL; test_present_thread the presenter on everywhere on the
presents' own queue with Reflex's id; test_readback_late the 0x3980 default and =0; test_readback_early a worker's own
recording part. Fixtures that stubbed the recorder now stub vk_work.h's counts. None of them run on this PC (no test
suites here); the Test Farm's and the reviewers' runs do.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
docs(vk): the renderer's threads (docs/development/threads.md), and the switches and pages the redesign changed
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
PR description / Description follows the template (pull_request_target) Successful in 10s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
CI / Workflow lint (actionlint) (pull_request) Successful in 12s
Third-party parts / As their notes say (pull_request) Successful in 12s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/amd+igpu 0/1 pass, 0 running, 2 queued
beanstalk/windows+nvidia 0/1 pass, 1 running, 0 queued
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 0/1 pass, 2 running, 0 queued
beanstalk/linux+nvidia 0/1 pass, 0 running, 2 queued
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+amd 0/1 pass, 0 running, 2 queued
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
9259257365
A page of its own for the new layout: each thread's job, the hand-offs (segments, not calls), the ordering kept, the
segment workers' ceiling and calibration, where they run, what stays off where they may record, and how to measure.
docs/development.md's switch rows (the recorder's and RX_VK_PRESENT_THREAD gone; RX_VK_PARALLEL, RX_VK_PAR_CALIBRATE,
RX_VK_PASS_BARRIER, RX_VK_READBACK_LATE, Threaded rendering), the frame pipeline, design.md's rules (section 4 rule 7,
section 13 rule 5), tracy.md, device-policy.md, the Steam Deck playbook and the README's Threaded rendering.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
Author
Owner

Tested in game: CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X: Kokiri Forest, Hyrule Field and the Market with Auto's render thread and segment workers and the presenter thread; no error

Tested in game: CachyOS Linux, RTX 3090 (NVIDIA 620.70.06), Ryzen 9 7900X: Kokiri Forest, Hyrule Field and the Market with Auto's render thread and segment workers and the presenter thread; no error
Author
Owner

@Beanstalk please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, and amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), defaults otherwise (Threaded rendering Auto). From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. The segment workers' count is measured in play during the first ~30 s of play (logged as vk: segment workers: N kept), so please run each area long enough after loading. Measure fps, p99 frame time and stutters, and compare pictures with main. Also run one extra cell on the PR side with RX_VK_PAR_CALIBRATE=0 (the ceiling, no calibration) to separate the calibration from the rest. Expect fps at least main's everywhere, no crash, and no new picture differences beyond main's own noise. Logs to look for: vk: Threaded rendering:, vk: segment workers run on logical processors, vk: presenter thread:.

@Beanstalk please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, and amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), defaults otherwise (Threaded rendering Auto). From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. The segment workers' count is measured in play during the first ~30 s of play (logged as `vk: segment workers: N kept`), so please run each area long enough after loading. Measure fps, p99 frame time and stutters, and compare pictures with main. Also run one extra cell on the PR side with `RX_VK_PAR_CALIBRATE=0` (the ceiling, no calibration) to separate the calibration from the rest. Expect fps at least main's everywhere, no crash, and no new picture differences beyond main's own noise. Logs to look for: `vk: Threaded rendering:`, `vk: segment workers run on logical processors`, `vk: presenter thread:`.
Member
Beanstalk test farm · an earlier run: commit 9259257 (the newer results are in a comment below)

Commit 9259257 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

Two starts per side will visit Kokiri Forest, Hyrule Field and the Market, with V-Sync off and the frame-rate cap uncapped. Each area gets a 20-second standing measurement, a camera turn and a 15-second post-turn measurement; FPS, p99 frame time, stutters, worker-count logs and pictures are compared with main. An extra PR-only calibration-off cell separates the worker ceiling from the default calibration; the plan takes roughly 40 minutes per tester. Crashes, lower FPS, increased p99 or stutters, and picture differences above the comparison tolerance fail the plan. Windows and Linux Intel targets currently have no online tester and will wait for one. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • Matrix: 1 cell, each on main and the PR: RX_VK_PAR_CALIBRATE 0 (PR only).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 0 s warm-up), take the picture forest-standing to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), take the picture forest-after-turn to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 0 s warm-up), take the picture field-standing to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-after-turn to compare with main, teleport to Market, measure market-standing for 20 s (after 0 s warm-up), take the picture market-standing to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-after-turn to compare with main; 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate, the game's log, numbers from the game's log.
  • The PR fails if: it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or stutters is more than 0% above main; or frame time (p99) is more than 0% above main; or forbidden log lines is above 0; or expected log lines missing is above 0.
  • Time: about 27 min of game runs per tester (4 game starts), plus the two builds.

Started at 17:35 UTC.

Estimate: done in about 2 h 30 min, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd 0/1 — — — queued
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 1/1 1 none — — running
windows+intel 0/1 — — — queued
linux+amd 0/1 — — — running
linux+nvidia 0/1 — — — queued
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — queued
amd+igpu 0/1 — — — queued

Measurements · windows+nvidia

Verdict: running on windows+nvidia (1/1 report).

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 0 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 0 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 0 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Nothing was measured: no metric came back.

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_PAR_CALIBRATE switch the PR adds (set on the PR side only) 0 not in main the PR's default three-areas, main without it → the PR with 0: not measured untested
Markdown for the PR description
## Measurements

**Verdict:** running on windows+nvidia (1/1 report).

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 0 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 0 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 0 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Nothing was measured: no metric came back.

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: not measured | untested |
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>9259257</code> (the newer results are in a comment below)</summary> Commit `9259257` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) Two starts per side will visit Kokiri Forest, Hyrule Field and the Market, with V-Sync off and the frame-rate cap uncapped. Each area gets a 20-second standing measurement, a camera turn and a 15-second post-turn measurement; FPS, p99 frame time, stutters, worker-count logs and pictures are compared with main. An extra PR-only calibration-off cell separates the worker ceiling from the default calibration; the plan takes roughly 40 minutes per tester. Crashes, lower FPS, increased p99 or stutters, and picture differences above the comparison tolerance fail the plan. Windows and Linux Intel targets currently have no online tester and will wait for one. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **Matrix:** 1 cell, each on main and the PR: RX_VK_PAR_CALIBRATE `0` (PR only). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 0 s warm-up), take the picture forest-standing to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), take the picture forest-after-turn to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 0 s warm-up), take the picture field-standing to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-after-turn to compare with main, teleport to Market, measure market-standing for 20 s (after 0 s warm-up), take the picture market-standing to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-after-turn to compare with main; 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate, the game's log, numbers from the game's log. - **The PR fails if:** it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or stutters is more than 0% above main; or frame time (p99) is more than 0% above main; or forbidden log lines is above 0; or expected log lines missing is above 0. - **Time:** about 27 min of game runs per tester (4 game starts), plus the two builds. Started at 17:35 UTC. **Estimate:** done in about 2 h 30 min, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd | 0/1 | — | — | — | queued | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_7d8dbc2686d159db) | none | — | — | running | | windows+intel | 0/1 | — | — | — | queued | | linux+amd | 0/1 | — | — | — | running | | linux+nvidia | 0/1 | — | — | — | queued | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | queued | | amd+igpu | 0/1 | — | — | — | queued | #### Measurements · windows+nvidia **Verdict:** running on windows+nvidia (1/1 report). Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 0 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 0 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 0 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: not measured | untested | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** running on windows+nvidia (1/1 report). Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 0 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 0 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 0 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: not measured | untested | ``` </details> </details>
On 4 cores the recorder thread beside the render thread was the slowest choice (60.9 / 62.1 fps against 67.6 / 67.9 with
the render thread alone, a third of the frames long), which is why Auto's second thread started at 6. That cost was the
recorder's call-by-call hand-off; the segment workers wake a few times a frame, and the calibration keeps none where two are
slower. So Auto is the render thread and segment workers from 4 cores (2 workers on 4, 3 on 5, 4 on 6, up to 6), Off below.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The late fix (a capture's late pages merged on the GPU, with no wait: the wait cost ~4.8 ms of a 17.8 ms frame, and on
an 8 GB card took Hyrule Field from 55 to 35-45 fps) was off with the parallel replay, its workers recording copies out of
the captures in any order. Every capture a submission's segments copy from is taken before its replay, so its fixes are all
known when the segments start: memory_sync_mirrors_before_segments now emits every pending fix in the recording's uploads
ahead of them, and a worker's copy out of a capture comes after them in the queue. So it stays on with the workers.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
perf(vk): the segment workers' calibration settles in seconds and is kept for the CPU and GPU
Some checks failed
beanstalk/windows+amd 0/1 pass, 0 running, 2 queued
beanstalk/windows+nvidia 1/1 fail
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 2/1 fail
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/amd+igpu 0/1 pass, 2 running, 0 queued
beanstalk/linux+nvidia 0/1 pass, 1 running, 1 queued
beanstalk/approval approved by HumanBean (reviewed)
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
Third-party parts / As their notes say (pull_request) Successful in 13s
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 5s
85c0d23d0d
The first calibration (120-present windows, 6 rounds over the ceiling, half and none) never finished within a session of
play: the A/B's sessions alternated their worker count throughout. Now 60-present windows, 4 rounds, over the ceiling, half
of it, two and none (on a Ryzen 9 7900X, 6 workers fixed was 10-17% under main in Kokiri Forest: the best can be well
under the ceiling). The choice is written to cache/longshot_segment_workers.txt with the CPU and GPU; the next start on the
same pair takes it at once. A new CPU, GPU or ceiling measures again.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
Member
Beanstalk test farm · an earlier run: commit 85c0d23 (the newer results are in a comment below)

Commit 85c0d23 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

Two starts per side visit Kokiri Forest, Hyrule Field and Castle Town Market with V-Sync off and the frame rate uncapped. Each area is measured standing and after a camera turn; the turning input itself is not measured. An extra PR cell disables calibration to expose the worker ceiling, and the retained worker count is recorded when logged. The plan takes about 45 minutes per tester and fails on crashes, lower FPS, worse p99 or stutter counts, or picture differences above the allowed tolerance. Intel targets have no testers listed online and will wait for matching PCs; other combinations may also wait if their tags do not coincide on one PC. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • Matrix: 1 cell, each on main and the PR: RX_VK_PAR_CALIBRATE 0 (PR only).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure kokiri-standing for 20 s (after 15 s warm-up), take the picture kokiri-standing to compare with main, move the stick, measure kokiri-turned for 15 s (after 0 s warm-up), take the picture kokiri-turned to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 15 s warm-up), take the picture field-standing to compare with main, measure field-turned for 15 s (after 0 s warm-up), take the picture field-turned to compare with main, teleport to Market, measure market-standing for 20 s (after 15 s warm-up), take the picture market-standing to compare with main, measure market-turned for 15 s (after 0 s warm-up), take the picture market-turned to compare with main; 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate, numbers from the game's log, the game's log.
  • The PR fails if: it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 3%; or stutters is more than 0% above main; or frame time (p99) is more than 0% above main; or forbidden log lines is above 0.
  • Time: about 30 min of game runs per tester (4 game starts), plus the two builds.

Started at 17:48 UTC.

Estimate: done in about 2 h, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd 0/1 — — — queued
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3070, 621.24.0 2/1 1 2 1 PR run 14.5 vs 11.2 (+29%) — fail
windows+intel 0/1 — — — queued
linux+amd · 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7; 2: linux+amd+radv+dgpu AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6 2/1 1 2 none 14.4 vs 14.1 (+2%) — fail ⚠ outlier
linux+nvidia 0/1 — — — running
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — queued
amd+igpu 0/1 — — — running

Measurements · windows+nvidia

Verdict: fail on windows+nvidia (2/1 report): three-areas crash; three-areas kokiri-standing.fps_mean.min_delta_pct (-31.6%); three-areas kokiri-turned.fps_mean.min_delta_pct (-35.3%) failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3070, 621.24.0 AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ 85c0d23
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · frame rate 115.0 84.1 (−27%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · frame rate 87.1 56.4 (−35%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · frame rate 98.8 65.3 (−34%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · frame rate 104.0 83.6 (−20%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · frame rate 94.5 64.6 (−32%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · frame rate 69.5 62.5 (−10%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · stutters 0.0 0.0
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · stutters 0.0 0.0
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · stutters 3.0 0.0 (−100%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · stutters 0.0 0.0
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · stutters 0.0 0.5
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · stutters 0.0 3.0
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · frame time (p99) (ms) 11.2 14.5 (+29%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · frame time (p99) (ms) 15.6 21.0 (+35%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · frame time (p99) (ms) 14.7 18.7 (+27%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · frame time (p99) (ms) 12.1 15.1 (+25%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · frame time (p99) (ms) 13.3 19.1 (+44%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · frame time (p99) (ms) 19.7 20.6 (+5%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_PAR_CALIBRATE switch the PR adds (set on the PR side only) 0 not in main the PR's default three-areas, main without it → the PR with 0: 115.0 → 84.1 (−27%) doesn't help
Markdown for the PR description
## Measurements

**Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.6%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-35.3%)` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 |
|---|---|---|
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 115.0 | 84.1 (−27%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 87.1 | 56.4 (−35%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 98.8 | 65.3 (−34%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 104.0 | 83.6 (−20%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 94.5 | 64.6 (−32%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 69.5 | 62.5 (−10%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.0 | 0.0 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.0 | 0.0 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 3.0 | 0.0 (−100%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.0 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.0 | 0.5 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 0.0 | 3.0 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 11.2 | 14.5 (+29%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.6 | 21.0 (+35%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 14.7 | 18.7 (+27%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 12.1 | 15.1 (+25%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 13.3 | 19.1 (+44%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 19.7 | 20.6 (+5%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 |

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 115.0 → 84.1 (−27%) | doesn't help |

Measurements · linux+amd

Verdict: fail on linux+amd (2/1 report): three-areas kokiri-standing.fps_mean.min_delta_pct (-31.8%); three-areas kokiri-turned.fps_mean.min_delta_pct (-34.1%); three-areas field-turned.fps_mean.min_delta_pct (-3.5%) failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM; 2: linux+amd+radv+dgpu AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6 11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ 85c0d23
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · frame rate 95.3 84.1 (−12%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · frame rate 93.7 75.3 (−20%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · frame rate 81.1 78.5 (−3%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · frame rate 88.8 93.1 (+5%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · frame rate 91.4 70.4 (−23%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · frame rate 72.5 64.2 (−11%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · stutters 0.2 0.2 (+0%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · stutters 0.5 0.2 (−50%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · stutters 0.2 2.0 (+700%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · stutters 0.0 0.2
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · stutters 0.8 0.5 (−33%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · stutters 19.2 3.0 (−84%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-turned · frame time (p99) (ms) 14.1 14.4 (+2%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-turned · frame time (p99) (ms) 15.0 15.2 (+1%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-turned · frame time (p99) (ms) 15.7 17.1 (+9%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · field-standing · frame time (p99) (ms) 14.6 13.9 (−5%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · kokiri-standing · frame time (p99) (ms) 14.7 16.5 (+12%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · market-standing · frame time (p99) (ms) 39.6 21.0 (−47%)
three-areas · RX_VK_PAR_CALIBRATE=0 (PR only) · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_PAR_CALIBRATE switch the PR adds (set on the PR side only) 0 not in main the PR's default three-areas, main without it → the PR with 0: 95.3 → 84.1 (−12%) doesn't help
Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd (2/1 report): `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.8%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-34.1%)`; `three-areas field-turned.fps_mean.min_delta_pct (-3.5%)` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 |
|---|---|---|
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 95.3 | 84.1 (−12%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 93.7 | 75.3 (−20%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 81.1 | 78.5 (−3%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 88.8 | 93.1 (+5%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 91.4 | 70.4 (−23%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 72.5 | 64.2 (−11%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.2 | 0.2 (+0%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.5 | 0.2 (−50%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 0.2 | 2.0 (+700%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.2 |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.8 | 0.5 (−33%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 19.2 | 3.0 (−84%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 14.1 | 14.4 (+2%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.0 | 15.2 (+1%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 15.7 | 17.1 (+9%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 14.6 | 13.9 (−5%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 14.7 | 16.5 (+12%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 39.6 | 21.0 (−47%) |
| three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 |

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 95.3 → 84.1 (−12%) | doesn't help |
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>85c0d23</code> (the newer results are in a comment below)</summary> Commit `85c0d23` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) Two starts per side visit Kokiri Forest, Hyrule Field and Castle Town Market with V-Sync off and the frame rate uncapped. Each area is measured standing and after a camera turn; the turning input itself is not measured. An extra PR cell disables calibration to expose the worker ceiling, and the retained worker count is recorded when logged. The plan takes about 45 minutes per tester and fails on crashes, lower FPS, worse p99 or stutter counts, or picture differences above the allowed tolerance. Intel targets have no testers listed online and will wait for matching PCs; other combinations may also wait if their tags do not coincide on one PC. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **Matrix:** 1 cell, each on main and the PR: RX_VK_PAR_CALIBRATE `0` (PR only). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure kokiri-standing for 20 s (after 15 s warm-up), take the picture kokiri-standing to compare with main, move the stick, measure kokiri-turned for 15 s (after 0 s warm-up), take the picture kokiri-turned to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 15 s warm-up), take the picture field-standing to compare with main, measure field-turned for 15 s (after 0 s warm-up), take the picture field-turned to compare with main, teleport to Market, measure market-standing for 20 s (after 15 s warm-up), take the picture market-standing to compare with main, measure market-turned for 15 s (after 0 s warm-up), take the picture market-turned to compare with main; 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate, numbers from the game's log, the game's log. - **The PR fails if:** it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 3%; or stutters is more than 0% above main; or frame time (p99) is more than 0% above main; or forbidden log lines is above 0. - **Time:** about 30 min of game runs per tester (4 game starts), plus the two builds. Started at 17:48 UTC. **Estimate:** done in about 2 h, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd | 0/1 | — | — | — | queued | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_0954dc7f3eee1317) [2](https://beanstalk.magicbean.dev/reports/rp_b051cb093387ac27) | 1 PR run | 14.5 vs 11.2 (+29%) | — | fail | | windows+intel | 0/1 | — | — | — | queued | | linux+amd · 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_62a0802c7ceda675) [2](https://beanstalk.magicbean.dev/reports/rp_eedfff236d4fddc3) | none | 14.4 vs 14.1 (+2%) | — | fail ⚠ outlier | | linux+nvidia | 0/1 | — | — | — | running | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | queued | | amd+igpu | 0/1 | — | — | — | running | #### Measurements · windows+nvidia **Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.6%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-35.3%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 | |---|---|---| | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 115.0 | 84.1 (−27%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 87.1 | 56.4 (−35%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 98.8 | 65.3 (−34%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 104.0 | 83.6 (−20%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 94.5 | 64.6 (−32%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 69.5 | 62.5 (−10%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 3.0 | 0.0 (−100%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.0 | 0.5 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 0.0 | 3.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 11.2 | 14.5 (+29%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.6 | 21.0 (+35%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 14.7 | 18.7 (+27%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 12.1 | 15.1 (+25%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 13.3 | 19.1 (+44%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 19.7 | 20.6 (+5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 115.0 → 84.1 (−27%) | doesn't help | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.6%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-35.3%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 | |---|---|---| | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 115.0 | 84.1 (−27%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 87.1 | 56.4 (−35%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 98.8 | 65.3 (−34%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 104.0 | 83.6 (−20%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 94.5 | 64.6 (−32%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 69.5 | 62.5 (−10%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 3.0 | 0.0 (−100%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.0 | 0.5 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 0.0 | 3.0 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 11.2 | 14.5 (+29%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.6 | 21.0 (+35%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 14.7 | 18.7 (+27%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 12.1 | 15.1 (+25%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 13.3 | 19.1 (+44%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 19.7 | 20.6 (+5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 115.0 → 84.1 (−27%) | doesn't help | ``` </details> #### Measurements · linux+amd **Verdict:** fail on linux+amd (2/1 report): `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.8%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-34.1%)`; `three-areas field-turned.fps_mean.min_delta_pct (-3.5%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 | |---|---|---| | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 95.3 | 84.1 (−12%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 93.7 | 75.3 (−20%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 81.1 | 78.5 (−3%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 88.8 | 93.1 (+5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 91.4 | 70.4 (−23%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 72.5 | 64.2 (−11%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.2 | 0.2 (+0%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.5 | 0.2 (−50%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 0.2 | 2.0 (+700%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.2 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.8 | 0.5 (−33%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 19.2 | 3.0 (−84%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 14.1 | 14.4 (+2%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.0 | 15.2 (+1%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 15.7 | 17.1 (+9%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 14.6 | 13.9 (−5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 14.7 | 16.5 (+12%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 39.6 | 21.0 (−47%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 95.3 → 84.1 (−12%) | doesn't help | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd (2/1 report): `three-areas kokiri-standing.fps_mean.min_delta_pct (-31.8%)`; `three-areas kokiri-turned.fps_mean.min_delta_pct (-34.1%)`; `three-areas field-turned.fps_mean.min_delta_pct (-3.5%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 15 s warm-up, kokiri-turned 15 s after 0 s warm-up, field-standing 20 s after 15 s warm-up, field-turned 15 s after 0 s warm-up, market-standing 20 s after 15 s warm-up, market-turned 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ 85c0d23 | |---|---|---| | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame rate | 95.3 | 84.1 (−12%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame rate | 93.7 | 75.3 (−20%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame rate | 81.1 | 78.5 (−3%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame rate | 88.8 | 93.1 (+5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame rate | 91.4 | 70.4 (−23%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame rate | 72.5 | 64.2 (−11%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · stutters | 0.2 | 0.2 (+0%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · stutters | 0.5 | 0.2 (−50%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · stutters | 0.2 | 2.0 (+700%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · stutters | 0.0 | 0.2 | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · stutters | 0.8 | 0.5 (−33%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · stutters | 19.2 | 3.0 (−84%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-turned · frame time (p99) (ms) | 14.1 | 14.4 (+2%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-turned · frame time (p99) (ms) | 15.0 | 15.2 (+1%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-turned · frame time (p99) (ms) | 15.7 | 17.1 (+9%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · field-standing · frame time (p99) (ms) | 14.6 | 13.9 (−5%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · kokiri-standing · frame time (p99) (ms) | 14.7 | 16.5 (+12%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · market-standing · frame time (p99) (ms) | 39.6 | 21.0 (−47%) | | three-areas · `RX_VK_PAR_CALIBRATE=0 (PR only)` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PAR_CALIBRATE` | switch the PR adds (set on the PR side only) | `0` | not in main | the PR's default | three-areas, main without it → the PR with 0: 95.3 → 84.1 (−12%) | doesn't help | ``` </details> </details>
Author
Owner

@Beanstalk (new head 85c0d23: segment workers from 4 cores, the late fix kept with them, a faster calibration saved per CPU and GPU) please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, and amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), defaults otherwise (Threaded rendering Auto). From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. The segment workers' count is measured in play during the first ~10-15 s of play and then saved (logged as vk: segment workers: N kept), so please run each area long enough after loading. Measure fps, p99 frame time and stutters, and compare pictures with main. Also run one extra cell on the PR side with RX_VK_PAR_CALIBRATE=0 (the ceiling, no calibration) to separate the calibration from the rest. Expect fps at least main's everywhere, no crash, and no new picture differences beyond main's own noise. Logs to look for: vk: Threaded rendering:, vk: segment workers run on logical processors, vk: presenter thread:.

@Beanstalk (new head 85c0d23: segment workers from 4 cores, the late fix kept with them, a faster calibration saved per CPU and GPU) please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, and amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), defaults otherwise (Threaded rendering Auto). From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. The segment workers' count is measured in play during the first ~10-15 s of play and then saved (logged as `vk: segment workers: N kept`), so please run each area long enough after loading. Measure fps, p99 frame time and stutters, and compare pictures with main. Also run one extra cell on the PR side with `RX_VK_PAR_CALIBRATE=0` (the ceiling, no calibration) to separate the calibration from the rest. Expect fps at least main's everywhere, no crash, and no new picture differences beyond main's own noise. Logs to look for: `vk: Threaded rendering:`, `vk: segment workers run on logical processors`, `vk: presenter thread:`.
perf(vk): segment workers off by default: they lost on every PC measured, and crashed on one
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 24s
PR description / Description follows the template (pull_request_target) Successful in 27s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 34s
CI / Workflow lint (actionlint) (pull_request) Successful in 35s
Third-party parts / As their notes say (pull_request) Successful in 38s
beanstalk/approval not approved yet
59286c3201
Measured with every core's worth of segment workers (the calibration off), the parallel replay lost on every PC tried: an
RTX 3090 / Ryzen 9 7900X 10-25% (the calibration kept none there anyway), an RTX 3070 on Windows 10-35% with an access
violation in both runs, an RX 7900 XTX on Linux 3-34% (the Test Farm on #1763). A Tracy capture on the 7900X shows why: the
frame is the game's own draw building on its job workers, the main thread waiting on it; two segment workers made that 28%
slower (they share its cores) while the render thread, busy 22% of the time, gained nothing. So Auto and On give the render
thread alone (threaded_rendering.h kWorkersEnabled); RX_VK_PARALLEL=<n> still records segments, for testing. With none,
what turned off where segments may run is back on: DLAA before the bloom, the faded meshes' rule, the occlusion copy, the
early readbacks.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
Author
Owner

@Beanstalk (new head: segment workers are now off by default; the last run's RX_VK_PAR_CALIBRATE=0 cell, every core's worth of workers, lost 3-35% and crashed on the RTX 3070, so that path is opt-in only.) Please test this PR with its defaults, no switches, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's and no crash; the log's vk: Threaded rendering: line should say the render thread only.

@Beanstalk (new head: segment workers are now off by default; the last run's RX_VK_PAR_CALIBRATE=0 cell, every core's worth of workers, lost 3-35% and crashed on the RTX 3070, so that path is opt-in only.) Please test this PR **with its defaults, no switches**, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's and no crash; the log's `vk: Threaded rendering:` line should say the render thread only.
Member
Beanstalk test farm · an earlier run: commit a2cc7b5 (the newer results are in a comment below)

Commit a2cc7b5 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

On eight requested hardware targets, this compares the PR defaults with main across Kokiri Forest, Hyrule Field and Castle Town Market, using two starts per side and taking roughly 23 minutes per tester. Each area is confirmed before measurement; FPS, p99 frame time, stutters, crashes and pictures are compared with V-Sync off and the frame cap uncapped. Camera turns are exercised between measurements, but the plan cannot measure while holding the camera stick, so the 15-second follow-up measurements occur after turning. A crash, lower FPS in any segment, a p99 increase over 15%, or a picture difference above the threshold fails the PR; stutters are reported. Intel targets have no currently listed online tester and will wait for one. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-still for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), take the picture forest-turned to compare with main, teleport to Hyrule Field, measure field-still for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-turned to compare with main, teleport to Market, measure market-still for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-turned to compare with main; 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate.
  • The PR fails if: it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or frame time (p99) is more than 15% above main.
  • Time: about 28 min of game runs per tester (4 game starts), plus the two builds.

Started at 2026-11-19 21:35 UTC.

Estimate: done in about 40 min, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd · 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9070 XT, 2.0.416 1/1 1 none 11.5 vs 11.4 (+1%) — fail
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3090, 621.24.0 2/1 1 2 none 16.6 vs 16.5 (+1%) — fail
windows+intel 0/1 — — — queued
linux+amd · 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7; 2: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8 2/1 1 2 none 46.0 vs 50.5 (−9%) — fail ⚠ outlier
linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 1/1 1 none — — queued
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — running
amd+igpu · 1: windows+amd+amd-proprietary+igpu AMD Radeon(TM) 880M Graphics, 2.0.416; 2: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8 2/1 1 2 none 71.8 vs 81.2 (−12%) — fail

Run by hand on PCs that don't match their target (not counted toward it):

Target Ran on Result
linux+nvidia off target (ran on windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4080, 621.14.0) fail

Measurements · windows+amd

Verdict: fail on windows+amd (1/1 report): three-areas field-still.fps_mean.min_delta_pct (-2.1%); three-areas field-after-turn.fps_mean.min_delta_pct (-4.6%); three-areas market-still.fps_mean.min_delta_pct (-14.5%) failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9070 XT, 2.0.416 AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ a2cc7b5
three-areas · field-still · frame rate 111.8 109.5 (−2%)
three-areas · forest-still · frame rate 99.8 110.8 (+11%)
three-areas · market-still · frame rate 68.7 58.8 (−14%)
three-areas · field-after-turn · frame rate 119.6 114.0 (−5%)
three-areas · forest-after-turn · frame rate 107.8 123.0 (+14%)
three-areas · market-after-turn · frame rate 94.8 99.3 (+5%)
three-areas · field-still · frame time (p99) (ms) 11.4 11.5 (+1%)
three-areas · forest-still · frame time (p99) (ms) 15.2 12.7 (−17%)
three-areas · market-still · frame time (p99) (ms) 19.5 21.8 (+12%)
three-areas · field-after-turn · frame time (p99) (ms) 10.7 11.4 (+6%)
three-areas · forest-after-turn · frame time (p99) (ms) 11.7 10.8 (−8%)
three-areas · market-after-turn · frame time (p99) (ms) 14.3 14.7 (+3%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+amd (1/1 report): `three-areas field-still.fps_mean.min_delta_pct (-2.1%)`; `three-areas field-after-turn.fps_mean.min_delta_pct (-4.6%)`; `three-areas market-still.fps_mean.min_delta_pct (-14.5%)` failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9070 XT, 2.0.416` `AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 |
|---|---|---|
| three-areas · field-still · frame rate | 111.8 | 109.5 (−2%) |
| three-areas · forest-still · frame rate | 99.8 | 110.8 (+11%) |
| three-areas · market-still · frame rate | 68.7 | 58.8 (−14%) |
| three-areas · field-after-turn · frame rate | 119.6 | 114.0 (−5%) |
| three-areas · forest-after-turn · frame rate | 107.8 | 123.0 (+14%) |
| three-areas · market-after-turn · frame rate | 94.8 | 99.3 (+5%) |
| three-areas · field-still · frame time (p99) (ms) | 11.4 | 11.5 (+1%) |
| three-areas · forest-still · frame time (p99) (ms) | 15.2 | 12.7 (−17%) |
| three-areas · market-still · frame time (p99) (ms) | 19.5 | 21.8 (+12%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 10.7 | 11.4 (+6%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 11.7 | 10.8 (−8%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 14.3 | 14.7 (+3%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · windows+nvidia

Verdict: fail on windows+nvidia (2/1 report): three-areas field-still.fps_mean.min_delta_pct (-1.7%); three-areas market-still.fps_mean.min_delta_pct (-6.2%); three-areas screenshot.max_diff failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3090, 621.24.0 AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ a2cc7b5
three-areas · field-still · frame rate 78.5 77.1 (−2%)
three-areas · forest-still · frame rate 69.2 73.5 (+6%)
three-areas · market-still · frame rate 60.7 56.9 (−6%)
three-areas · field-after-turn · frame rate 81.7 85.2 (+4%)
three-areas · forest-after-turn · frame rate 61.2 76.3 (+25%)
three-areas · market-after-turn · frame rate 69.5 73.6 (+6%)
three-areas · field-still · frame time (p99) (ms) 16.5 16.6 (+1%)
three-areas · forest-still · frame time (p99) (ms) 22.5 19.5 (−13%)
three-areas · market-still · frame time (p99) (ms) 22.2 40.6 (+83%)
three-areas · field-after-turn · frame time (p99) (ms) 16.5 15.6 (−5%)
three-areas · forest-after-turn · frame time (p99) (ms) 22.0 18.1 (−18%)
three-areas · market-after-turn · frame time (p99) (ms) 19.6 19.8 (+1%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-still.fps_mean.min_delta_pct (-1.7%)`; `three-areas market-still.fps_mean.min_delta_pct (-6.2%)`; `three-areas screenshot.max_diff` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 |
|---|---|---|
| three-areas · field-still · frame rate | 78.5 | 77.1 (−2%) |
| three-areas · forest-still · frame rate | 69.2 | 73.5 (+6%) |
| three-areas · market-still · frame rate | 60.7 | 56.9 (−6%) |
| three-areas · field-after-turn · frame rate | 81.7 | 85.2 (+4%) |
| three-areas · forest-after-turn · frame rate | 61.2 | 76.3 (+25%) |
| three-areas · market-after-turn · frame rate | 69.5 | 73.6 (+6%) |
| three-areas · field-still · frame time (p99) (ms) | 16.5 | 16.6 (+1%) |
| three-areas · forest-still · frame time (p99) (ms) | 22.5 | 19.5 (−13%) |
| three-areas · market-still · frame time (p99) (ms) | 22.2 | 40.6 (+83%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 16.5 | 15.6 (−5%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 22.0 | 18.1 (−18%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 19.6 | 19.8 (+1%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+amd

Verdict: fail on linux+amd (2/1 report): three-areas market-still.fps_mean.min_delta_pct (-24.5%); three-areas screenshot.max_diff; three-areas forest-still.fps_mean.min_delta_pct (-17.7%) failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM; 2: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8 AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ a2cc7b5
three-areas · field-still · frame rate 61.1 68.5 (+12%)
three-areas · forest-still · frame rate 61.1 66.2 (+8%)
three-areas · market-still · frame rate 43.5 35.2 (−19%)
three-areas · field-after-turn · frame rate 57.1 72.9 (+28%)
three-areas · forest-after-turn · frame rate 61.6 71.0 (+15%)
three-areas · market-after-turn · frame rate 55.5 58.4 (+5%)
three-areas · field-still · frame time (p99) (ms) 50.5 46.0 (−9%)
three-areas · forest-still · frame time (p99) (ms) 71.2 44.4 (−38%)
three-areas · market-still · frame time (p99) (ms) 81.7 82.1 (+1%)
three-areas · field-after-turn · frame time (p99) (ms) 47.2 38.1 (−19%)
three-areas · forest-after-turn · frame time (p99) (ms) 45.3 42.4 (−6%)
three-areas · market-after-turn · frame time (p99) (ms) 54.5 53.7 (−2%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd (2/1 report): `three-areas market-still.fps_mean.min_delta_pct (-24.5%)`; `three-areas screenshot.max_diff`; `three-areas forest-still.fps_mean.min_delta_pct (-17.7%)` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 |
|---|---|---|
| three-areas · field-still · frame rate | 61.1 | 68.5 (+12%) |
| three-areas · forest-still · frame rate | 61.1 | 66.2 (+8%) |
| three-areas · market-still · frame rate | 43.5 | 35.2 (−19%) |
| three-areas · field-after-turn · frame rate | 57.1 | 72.9 (+28%) |
| three-areas · forest-after-turn · frame rate | 61.6 | 71.0 (+15%) |
| three-areas · market-after-turn · frame rate | 55.5 | 58.4 (+5%) |
| three-areas · field-still · frame time (p99) (ms) | 50.5 | 46.0 (−9%) |
| three-areas · forest-still · frame time (p99) (ms) | 71.2 | 44.4 (−38%) |
| three-areas · market-still · frame time (p99) (ms) | 81.7 | 82.1 (+1%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 47.2 | 38.1 (−19%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 45.3 | 42.4 (−6%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 54.5 | 53.7 (−2%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+nvidia

Verdict: queued on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Nothing was measured: no metric came back.

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** queued on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Nothing was measured: no metric came back.

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · amd+igpu

Verdict: fail on amd+igpu (2/1 report): three-areas market-after-turn.fps_mean.min_delta_pct (-8.6%); three-areas screenshot.max_diff failed.

Hardware: 1: windows+amd+amd-proprietary+igpu AMD Radeon(TM) 880M Graphics, 2.0.416 AMD Ryzen AI 9 365 w/ Radeon 880M, 32 GB RAM; 2: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8 AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 9ea65c7 After: this PR @ a2cc7b5
three-areas · field-still · frame rate 19.4 19.7 (+2%)
three-areas · forest-still · frame rate 18.6 20.5 (+10%)
three-areas · market-still · frame rate 16.2 17.3 (+7%)
three-areas · field-after-turn · frame rate 21.3 21.5 (+1%)
three-areas · forest-after-turn · frame rate 21.9 22.1 (+1%)
three-areas · market-after-turn · frame rate 21.9 20.1 (−9%)
three-areas · field-still · frame time (p99) (ms) 81.2 71.8 (−12%)
three-areas · forest-still · frame time (p99) (ms) 80.1 70.6 (−12%)
three-areas · market-still · frame time (p99) (ms) 97.1 81.0 (−17%)
three-areas · field-after-turn · frame time (p99) (ms) 71.8 69.4 (−3%)
three-areas · forest-after-turn · frame time (p99) (ms) 75.4 75.8 (+1%)
three-areas · market-after-turn · frame time (p99) (ms) 72.4 78.7 (+9%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on amd+igpu (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-8.6%)`; `three-areas screenshot.max_diff` failed.

Hardware: 1: windows+amd+amd-proprietary+igpu `AMD Radeon(TM) 880M Graphics, 2.0.416` `AMD Ryzen AI 9 365 w/ Radeon 880M, 32 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 |
|---|---|---|
| three-areas · field-still · frame rate | 19.4 | 19.7 (+2%) |
| three-areas · forest-still · frame rate | 18.6 | 20.5 (+10%) |
| three-areas · market-still · frame rate | 16.2 | 17.3 (+7%) |
| three-areas · field-after-turn · frame rate | 21.3 | 21.5 (+1%) |
| three-areas · forest-after-turn · frame rate | 21.9 | 22.1 (+1%) |
| three-areas · market-after-turn · frame rate | 21.9 | 20.1 (−9%) |
| three-areas · field-still · frame time (p99) (ms) | 81.2 | 71.8 (−12%) |
| three-areas · forest-still · frame time (p99) (ms) | 80.1 | 70.6 (−12%) |
| three-areas · market-still · frame time (p99) (ms) | 97.1 | 81.0 (−17%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 71.8 | 69.4 (−3%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 75.4 | 75.8 (+1%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 72.4 | 78.7 (+9%) |

### Flags and settings

None: the plan varied no setting, switch or build option.
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>a2cc7b5</code> (the newer results are in a comment below)</summary> Commit `a2cc7b5` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) On eight requested hardware targets, this compares the PR defaults with main across Kokiri Forest, Hyrule Field and Castle Town Market, using two starts per side and taking roughly 23 minutes per tester. Each area is confirmed before measurement; FPS, p99 frame time, stutters, crashes and pictures are compared with V-Sync off and the frame cap uncapped. Camera turns are exercised between measurements, but the plan cannot measure while holding the camera stick, so the 15-second follow-up measurements occur after turning. A crash, lower FPS in any segment, a p99 increase over 15%, or a picture difference above the threshold fails the PR; stutters are reported. Intel targets have no currently listed online tester and will wait for one. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-still for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), take the picture forest-turned to compare with main, teleport to Hyrule Field, measure field-still for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-turned to compare with main, teleport to Market, measure market-still for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-turned to compare with main; 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate. - **The PR fails if:** it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or frame time (p99) is more than 15% above main. - **Time:** about 28 min of game runs per tester (4 game starts), plus the two builds. Started at 2026-11-19 21:35 UTC. **Estimate:** done in about 40 min, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd · 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9070 XT, 2.0.416` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_4be1ca7c7b82e2fe) | none | 11.5 vs 11.4 (+1%) | — | fail | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_2aeded2141050710) [2](https://beanstalk.magicbean.dev/reports/rp_ebcdda49abed53ad) | none | 16.6 vs 16.5 (+1%) | — | fail | | windows+intel | 0/1 | — | — | — | queued | | linux+amd · 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_6ed6fe8ef91ac4f5) [2](https://beanstalk.magicbean.dev/reports/rp_8a65ac1d92af5f80) | none | 46.0 vs 50.5 (−9%) | — | fail ⚠ outlier | | linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_f364084243440858) | none | — | — | queued | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | running | | amd+igpu · 1: windows+amd+amd-proprietary+igpu `AMD Radeon(TM) 880M Graphics, 2.0.416`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_dc42e91c4f46f3ea) [2](https://beanstalk.magicbean.dev/reports/rp_43e0263c89565cf2) | none | 71.8 vs 81.2 (−12%) | — | fail | Run by hand on PCs that don't match their target (not counted toward it): | Target | Ran on | Result | | --- | --- | --- | | linux+nvidia | off target (ran on windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4080, 621.14.0`) | [fail](https://beanstalk.magicbean.dev/reports/rp_e46ae7e4ee04c30a) | #### Measurements · windows+amd **Verdict:** fail on windows+amd (1/1 report): `three-areas field-still.fps_mean.min_delta_pct (-2.1%)`; `three-areas field-after-turn.fps_mean.min_delta_pct (-4.6%)`; `three-areas market-still.fps_mean.min_delta_pct (-14.5%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9070 XT, 2.0.416` `AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 111.8 | 109.5 (−2%) | | three-areas · forest-still · frame rate | 99.8 | 110.8 (+11%) | | three-areas · market-still · frame rate | 68.7 | 58.8 (−14%) | | three-areas · field-after-turn · frame rate | 119.6 | 114.0 (−5%) | | three-areas · forest-after-turn · frame rate | 107.8 | 123.0 (+14%) | | three-areas · market-after-turn · frame rate | 94.8 | 99.3 (+5%) | | three-areas · field-still · frame time (p99) (ms) | 11.4 | 11.5 (+1%) | | three-areas · forest-still · frame time (p99) (ms) | 15.2 | 12.7 (−17%) | | three-areas · market-still · frame time (p99) (ms) | 19.5 | 21.8 (+12%) | | three-areas · field-after-turn · frame time (p99) (ms) | 10.7 | 11.4 (+6%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 11.7 | 10.8 (−8%) | | three-areas · market-after-turn · frame time (p99) (ms) | 14.3 | 14.7 (+3%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+amd (1/1 report): `three-areas field-still.fps_mean.min_delta_pct (-2.1%)`; `three-areas field-after-turn.fps_mean.min_delta_pct (-4.6%)`; `three-areas market-still.fps_mean.min_delta_pct (-14.5%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9070 XT, 2.0.416` `AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 111.8 | 109.5 (−2%) | | three-areas · forest-still · frame rate | 99.8 | 110.8 (+11%) | | three-areas · market-still · frame rate | 68.7 | 58.8 (−14%) | | three-areas · field-after-turn · frame rate | 119.6 | 114.0 (−5%) | | three-areas · forest-after-turn · frame rate | 107.8 | 123.0 (+14%) | | three-areas · market-after-turn · frame rate | 94.8 | 99.3 (+5%) | | three-areas · field-still · frame time (p99) (ms) | 11.4 | 11.5 (+1%) | | three-areas · forest-still · frame time (p99) (ms) | 15.2 | 12.7 (−17%) | | three-areas · market-still · frame time (p99) (ms) | 19.5 | 21.8 (+12%) | | three-areas · field-after-turn · frame time (p99) (ms) | 10.7 | 11.4 (+6%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 11.7 | 10.8 (−8%) | | three-areas · market-after-turn · frame time (p99) (ms) | 14.3 | 14.7 (+3%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · windows+nvidia **Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-still.fps_mean.min_delta_pct (-1.7%)`; `three-areas market-still.fps_mean.min_delta_pct (-6.2%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 78.5 | 77.1 (−2%) | | three-areas · forest-still · frame rate | 69.2 | 73.5 (+6%) | | three-areas · market-still · frame rate | 60.7 | 56.9 (−6%) | | three-areas · field-after-turn · frame rate | 81.7 | 85.2 (+4%) | | three-areas · forest-after-turn · frame rate | 61.2 | 76.3 (+25%) | | three-areas · market-after-turn · frame rate | 69.5 | 73.6 (+6%) | | three-areas · field-still · frame time (p99) (ms) | 16.5 | 16.6 (+1%) | | three-areas · forest-still · frame time (p99) (ms) | 22.5 | 19.5 (−13%) | | three-areas · market-still · frame time (p99) (ms) | 22.2 | 40.6 (+83%) | | three-areas · field-after-turn · frame time (p99) (ms) | 16.5 | 15.6 (−5%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 22.0 | 18.1 (−18%) | | three-areas · market-after-turn · frame time (p99) (ms) | 19.6 | 19.8 (+1%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-still.fps_mean.min_delta_pct (-1.7%)`; `three-areas market-still.fps_mean.min_delta_pct (-6.2%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 78.5 | 77.1 (−2%) | | three-areas · forest-still · frame rate | 69.2 | 73.5 (+6%) | | three-areas · market-still · frame rate | 60.7 | 56.9 (−6%) | | three-areas · field-after-turn · frame rate | 81.7 | 85.2 (+4%) | | three-areas · forest-after-turn · frame rate | 61.2 | 76.3 (+25%) | | three-areas · market-after-turn · frame rate | 69.5 | 73.6 (+6%) | | three-areas · field-still · frame time (p99) (ms) | 16.5 | 16.6 (+1%) | | three-areas · forest-still · frame time (p99) (ms) | 22.5 | 19.5 (−13%) | | three-areas · market-still · frame time (p99) (ms) | 22.2 | 40.6 (+83%) | | three-areas · field-after-turn · frame time (p99) (ms) | 16.5 | 15.6 (−5%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 22.0 | 18.1 (−18%) | | three-areas · market-after-turn · frame time (p99) (ms) | 19.6 | 19.8 (+1%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+amd **Verdict:** fail on linux+amd (2/1 report): `three-areas market-still.fps_mean.min_delta_pct (-24.5%)`; `three-areas screenshot.max_diff`; `three-areas forest-still.fps_mean.min_delta_pct (-17.7%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 61.1 | 68.5 (+12%) | | three-areas · forest-still · frame rate | 61.1 | 66.2 (+8%) | | three-areas · market-still · frame rate | 43.5 | 35.2 (−19%) | | three-areas · field-after-turn · frame rate | 57.1 | 72.9 (+28%) | | three-areas · forest-after-turn · frame rate | 61.6 | 71.0 (+15%) | | three-areas · market-after-turn · frame rate | 55.5 | 58.4 (+5%) | | three-areas · field-still · frame time (p99) (ms) | 50.5 | 46.0 (−9%) | | three-areas · forest-still · frame time (p99) (ms) | 71.2 | 44.4 (−38%) | | three-areas · market-still · frame time (p99) (ms) | 81.7 | 82.1 (+1%) | | three-areas · field-after-turn · frame time (p99) (ms) | 47.2 | 38.1 (−19%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 45.3 | 42.4 (−6%) | | three-areas · market-after-turn · frame time (p99) (ms) | 54.5 | 53.7 (−2%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd (2/1 report): `three-areas market-still.fps_mean.min_delta_pct (-24.5%)`; `three-areas screenshot.max_diff`; `three-areas forest-still.fps_mean.min_delta_pct (-17.7%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 61.1 | 68.5 (+12%) | | three-areas · forest-still · frame rate | 61.1 | 66.2 (+8%) | | three-areas · market-still · frame rate | 43.5 | 35.2 (−19%) | | three-areas · field-after-turn · frame rate | 57.1 | 72.9 (+28%) | | three-areas · forest-after-turn · frame rate | 61.6 | 71.0 (+15%) | | three-areas · market-after-turn · frame rate | 55.5 | 58.4 (+5%) | | three-areas · field-still · frame time (p99) (ms) | 50.5 | 46.0 (−9%) | | three-areas · forest-still · frame time (p99) (ms) | 71.2 | 44.4 (−38%) | | three-areas · market-still · frame time (p99) (ms) | 81.7 | 82.1 (+1%) | | three-areas · field-after-turn · frame time (p99) (ms) | 47.2 | 38.1 (−19%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 45.3 | 42.4 (−6%) | | three-areas · market-after-turn · frame time (p99) (ms) | 54.5 | 53.7 (−2%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+nvidia **Verdict:** queued on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** queued on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · amd+igpu **Verdict:** fail on amd+igpu (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-8.6%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+igpu `AMD Radeon(TM) 880M Graphics, 2.0.416` `AMD Ryzen AI 9 365 w/ Radeon 880M, 32 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 19.4 | 19.7 (+2%) | | three-areas · forest-still · frame rate | 18.6 | 20.5 (+10%) | | three-areas · market-still · frame rate | 16.2 | 17.3 (+7%) | | three-areas · field-after-turn · frame rate | 21.3 | 21.5 (+1%) | | three-areas · forest-after-turn · frame rate | 21.9 | 22.1 (+1%) | | three-areas · market-after-turn · frame rate | 21.9 | 20.1 (−9%) | | three-areas · field-still · frame time (p99) (ms) | 81.2 | 71.8 (−12%) | | three-areas · forest-still · frame time (p99) (ms) | 80.1 | 70.6 (−12%) | | three-areas · market-still · frame time (p99) (ms) | 97.1 | 81.0 (−17%) | | three-areas · field-after-turn · frame time (p99) (ms) | 71.8 | 69.4 (−3%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 75.4 | 75.8 (+1%) | | three-areas · market-after-turn · frame time (p99) (ms) | 72.4 | 78.7 (+9%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on amd+igpu (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-8.6%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+igpu `AMD Radeon(TM) 880M Graphics, 2.0.416` `AMD Ryzen AI 9 365 w/ Radeon 880M, 32 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-still 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-still 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-still 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 9ea65c7 | After: this PR @ a2cc7b5 | |---|---|---| | three-areas · field-still · frame rate | 19.4 | 19.7 (+2%) | | three-areas · forest-still · frame rate | 18.6 | 20.5 (+10%) | | three-areas · market-still · frame rate | 16.2 | 17.3 (+7%) | | three-areas · field-after-turn · frame rate | 21.3 | 21.5 (+1%) | | three-areas · forest-after-turn · frame rate | 21.9 | 22.1 (+1%) | | three-areas · market-after-turn · frame rate | 21.9 | 20.1 (−9%) | | three-areas · field-still · frame time (p99) (ms) | 81.2 | 71.8 (−12%) | | three-areas · forest-still · frame time (p99) (ms) | 80.1 | 70.6 (−12%) | | three-areas · market-still · frame time (p99) (ms) | 97.1 | 81.0 (−17%) | | three-areas · field-after-turn · frame time (p99) (ms) | 71.8 | 69.4 (−3%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 75.4 | 75.8 (+1%) | | three-areas · market-after-turn · frame time (p99) (ms) | 72.4 | 78.7 (+9%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> </details>
Member

@HumanBean the PR got a new push (59286c3) after your comment. Review it and approve again.

@HumanBean the PR got a new push (`59286c3`) after your comment. Review it and approve again.
Contributor

Early findings from Windows 11 + RTX 4070 SUPER: it runs cleanly here, frames are noticeably more even, and the segment size looks like a useful tuning knob.

Just notes from trying the WIP, in case they help. PC: Ryzen 9 9900X (12 cores, 24 threads), RTX 4070 SUPER, Windows 11. 3840x2160 output, DLSS Quality, 90 fps cap, V-Sync on. A walking route: the Market, Hyrule Castle's grounds, the Market again (60 s). One build for every row: main 3653c9c5 with this PR merged (85c0d23d) and a few local test changes that are the same in every row. Only switches differ, one run each unless said.

  • No flash or dark frame seen in the runs watched (by eye, not the frame log). No errors or device loss.
  • Smoother frame to frame in the Market: the median step between frames is 0.67 ms, against 1.12-1.16 ms on our previous build (v3.0.0 with the old recorder; a different base, so only roughly comparable). The render thread's own CPU time falls from 9.9 to 4.6 ms a frame, with the rest on the workers.
  • The calibration chose 6 workers ("6: 14.17 ms, 3: 14.86, 2: 15.30, 0: 14.87").

The segment size and the scene interact. Medians of the frame time, same build:

Switches castle Market (steady 55 s) Market: frames 1.5x their neighbours and 6 ms over, a minute
as it comes 13.81 ms 13.69 ms 15
RX_VK_PASS_BARRIER=full 13.82 13.69 12
RX_VK_PARALLEL=0 RX_VK_PASS_BARRIER=full 12.74 14.95 8
RX_VK_PARALLEL_SEG=80 (2 runs) 12.27 / 12.33 13.60 / 13.63 25 / 22
RX_VK_PARALLEL_SEG=640 18.61 21.20
  • In the GPU-bound castle, smaller pieces help. With 80 it runs 12.3 ms, our old build's speed, and the Market's median is the lowest we measured. Bigger pieces go the other way: with 640, the game waits about 8.5 ms at the fence. So how early the GPU gets each piece seems to matter.
  • On the CPU side, more pieces cost something in the Market's heavy frames. With 80, the extra stutters are mostly render-thread and game-side frames. The worker count and the segment size might be worth calibrating together, or per scene: few big pieces where the CPU is the limit, small ones where the GPU is.
  • The narrower barriers neither helped nor hurt measurably here.
  • Side note: RX_VK_PASS_TIMING records serially (gprof_on() in the parallel check in vk_replay.cpp), so a timed run doesn't show the segments.
**Early findings from Windows 11 + RTX 4070 SUPER: it runs cleanly here, frames are noticeably more even, and the segment size looks like a useful tuning knob.** Just notes from trying the WIP, in case they help. PC: Ryzen 9 9900X (12 cores, 24 threads), RTX 4070 SUPER, Windows 11. 3840x2160 output, DLSS Quality, 90 fps cap, V-Sync on. A walking route: the Market, Hyrule Castle's grounds, the Market again (60 s). One build for every row: main 3653c9c5 with this PR merged (85c0d23d) and a few local test changes that are the same in every row. Only switches differ, one run each unless said. - **No flash or dark frame seen** in the runs watched (by eye, not the frame log). No errors or device loss. - **Smoother frame to frame in the Market:** the median step between frames is 0.67 ms, against 1.12-1.16 ms on our previous build (v3.0.0 with the old recorder; a different base, so only roughly comparable). The render thread's own CPU time falls from 9.9 to 4.6 ms a frame, with the rest on the workers. - **The calibration chose 6 workers** ("6: 14.17 ms, 3: 14.86, 2: 15.30, 0: 14.87"). The segment size and the scene interact. Medians of the frame time, same build: | Switches | castle | Market (steady 55 s) | Market: frames 1.5x their neighbours and 6 ms over, a minute | |---|---|---|---| | as it comes | 13.81 ms | 13.69 ms | 15 | | `RX_VK_PASS_BARRIER=full` | 13.82 | 13.69 | 12 | | `RX_VK_PARALLEL=0 RX_VK_PASS_BARRIER=full` | 12.74 | 14.95 | 8 | | `RX_VK_PARALLEL_SEG=80` (2 runs) | 12.27 / 12.33 | 13.60 / 13.63 | 25 / 22 | | `RX_VK_PARALLEL_SEG=640` | 18.61 | 21.20 | | - **In the GPU-bound castle, smaller pieces help.** With 80 it runs 12.3 ms, our old build's speed, and the Market's median is the lowest we measured. Bigger pieces go the other way: with 640, the game waits about 8.5 ms at the fence. So how early the GPU gets each piece seems to matter. - **On the CPU side, more pieces cost something in the Market's heavy frames.** With 80, the extra stutters are mostly render-thread and game-side frames. The worker count and the segment size might be worth calibrating together, or per scene: few big pieces where the CPU is the limit, small ones where the GPU is. - **The narrower barriers neither helped nor hurt** measurably here. - **Side note:** `RX_VK_PASS_TIMING` records serially (`gprof_on()` in the parallel check in `vk_replay.cpp`), so a timed run doesn't show the segments.
WIP: under measurement. A part ends once 120 draws long, with no render pass open (or anywhere past 240) while the GPU has
at most one part left, and at 512 at the latest; RX_VK_CHUNK=<n> keeps a fixed count for A/B. RX_VK_GPU_TIMELINE=1 writes
each recording's submit and GPU start and end, the jobs, the game's fence waits and the presents to
longshot_gpu_timeline.csv.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
The segment workers, the dry pass, the calibration and the lock they shared are gone, with no switch back
(RX_VK_PARALLEL, RX_VK_PARALLEL_SEG and RX_VK_PAR_CALIBRATE are GONE rows). Measured with every core's worth of
workers they lost on every PC tried: an RTX 3090 / Ryzen 9 7900X 10-25% (the calibration kept none there), an RTX
3070 on Windows 10-35% with an access violation in both runs, an RX 7900 XTX on Linux 3-34% (the Test Farm on
#1763). A Tracy capture on the 7900X shows why: the game's own draw building on its job threads is the frame, and
two workers made it 28% slower (they shared its cores) while the render thread, busy 22% of the time, gained
nothing.

What went: par_ceiling, par_threads and the calibration (cache/longshot_segment_workers.txt), the worker pool and
its affinity, the dry pass (Mode::Dry, par_points, dry_mark_written), segment recording and its init command
buffers (InitRedirect, texture_segment, the per-thread recording part), the pre-segment mirror sync, the read-back
refresh's segment order and GPU write hazards, ParLock and par_active everywhere, the FPS counter's workers lane,
Threaded rendering's worker tier (the row now decides the render thread alone: Auto from 4 cores). What they had
turned off is on by its own policy again: the capture's late fix, the early readbacks and late pages left to
their reads, DLAA before the bloom, the faded-mesh rule, the occlusion copy, the frame overlap lines.

Tests that read the source follow; the ones that checked only the parallel replay are deleted. Docs say one
recording thread (docs/development/threads.md).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
perf(vk): a submission's part ends where the game has ended its render pass, from 120 draws, at 512 at the latest
Some checks failed
beanstalk/windows+amd 1/1 fail
beanstalk/windows+nvidia 1/1 fail
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 1 running, 1 queued
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 2/1 fail
beanstalk/linux+nvidia 0/1 pass, 0 running, 1 queued
beanstalk/amd+igpu 1/1 fail
Third-party parts / The same as upstream (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 14s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 20s
Website / Website on Cloudflare (pull_request) Has been skipped
Third-party parts / As their notes say (pull_request) Successful in 24s
CI / Workflow lint (actionlint) (pull_request) Successful in 23s
PR description / Description follows the template (pull_request_target) Successful in 20s
Website / Website on Cloudflare (pull_request_target) Has been skipped
a2cc7b53ec
A part ends so that the GPU starts on what is recorded while the rest is replayed (no parts: 30-42% slower in every area),
but a part that ends inside a render pass ends the pass early: its targets written out and read in again. A fixed count
(RX_VK_CHUNK, 240) cut wherever it fell; cutting whenever the GPU ran out of work fed itself (18% slower). Now a part ends
at the first place the game has ended its pass (~150 a frame: its barriers, dispatches, copies) once it is RX_VK_PART_MIN
(120) long, and inside a pass only at RX_VK_PART_MAX (512). Market, RTX 3090 / Ryzen 9 7900X, 3 rounds interleaved against
every 240: standing 71.2 against 68.2 fps, walking 73.2 against 63.5, turning 90.3 against 88.1, faster in all 9 pairs;
walking's 99th percentile 26.8 ms against 36.7; frames over 50 ms 20 against 44. RX_VK_CHUNK is GONE.
RX_VK_GPU_TIMELINE=1 (diag): each recording's submit and GPU start and end, the jobs, the game's fence waits and the
presents, to longshot_gpu_timeline.csv.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016o1k1PazAdTJxQQWQBksW2
Author
Owner

@Beanstalk (new head a2cc7b5: the parallel replay is removed entirely, one render thread records every command buffer; and a submission's parts now end where the game has already ended its render pass, from 120 draws, at 512 at the latest, instead of every 240 wherever it fell.) Please test this PR with its defaults, no switches, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Make sure each side has reached the area before measuring it (an earlier run measured one side still at the start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's everywhere and no crash; GPU-limited machines (the Deck, integrated GPUs) may gain the most from the new part ends.

@Beanstalk (new head a2cc7b5: the parallel replay is removed entirely, one render thread records every command buffer; and a submission's parts now end where the game has already ended its render pass, from 120 draws, at 512 at the latest, instead of every 240 wherever it fell.) Please test this PR **with its defaults, no switches**, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Make sure each side has reached the area before measuring it (an earlier run measured one side still at the start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's everywhere and no crash; GPU-limited machines (the Deck, integrated GPUs) may gain the most from the new part ends.
Member

@HumanBean the PR got a new push (a2cc7b5) after your comment. Review it and approve again.

@HumanBean the PR got a new push (`a2cc7b5`) after your comment. Review it and approve again.
Author
Owner

@Beanstalk (new head a2cc7b5: the parallel replay is removed entirely, one render thread records every command buffer; and a submission's parts now end where the game has already ended its render pass, from 120 draws, at 512 at the latest, instead of every 240 wherever it fell.) Please test this PR with its defaults, no switches, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Make sure each side has reached the area before measuring it (an earlier run measured one side still at the start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's everywhere and no crash; GPU-limited machines (the Deck, integrated GPUs) may gain the most from the new part ends.

@Beanstalk (new head a2cc7b5: the parallel replay is removed entirely, one render thread records every command buffer; and a submission's parts now end where the game has already ended its render pass, from 120 draws, at 512 at the latest, instead of every 240 wherever it fell.) Please test this PR **with its defaults, no switches**, against main, on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side. Make sure each side has reached the area before measuring it (an earlier run measured one side still at the start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. No extra cells. Expect fps at least main's everywhere and no crash; GPU-limited machines (the Deck, integrated GPUs) may gain the most from the new part ends.
Contributor

A first Windows run of this head a2cc7b53 against main 868f6be8 with #1729: Windows 11, RTX 4070, NVIDIA 616.56, Ryzen 9 7900X (the CPU from your measurements). The two builds differ only in the PR (packs on one base).

The route: a save, Debug tab → Teleport to Hyrule Field, Switch age to child, Set time 21:00, then about 2 minutes running around the field fighting the Stalchildren that spawn, with camera turns (it gives fights on demand: #1720, #1493 (comment)). Settings: 2560x1440, DLSS Quality, cap 90, frame generation off, Reflex and V-Sync on. Switches: RX_VK_PASS_TIMING=1 RX_VK_SLOW_LOG=50 HOS_THREAD_TIMES=5 HOS_FPS_LOG=1 on every run. A person played; numbers come from 120-present windows from the field's arrival, and a fight window has 200 or more dispatches a frame.

main this PR, run 1 this PR, run 2
fight fps (median, lowest) 69.4, 55.3 (11 windows) 65.2, 50.7 (8) 64.5, 55.7 (17)
other play fps (median) 76.9 77.2 75.4
replay CPU work a frame (a frame's replay time, submit, median) 7.30 ms 9.72 ms 10.05 ms
per draw (that over the median draws, ~2,100) 3.4 µs 4.7 µs 4.8 µs
host time in GPU resource waits (median) 2.19 ms 5.99 ms 6.33 ms
GPU time a frame (median) 8.6 ms 8.7 ms 8.9 ms
full barriers a frame (median) 438 426 —
  • Slower in fights here: 4-5 fps lower, and calm play level with main.
  • Mostly on the render thread: its replay work is ~35% higher at the same draw count, and it waits ~4 ms more a frame on the GPU, while GPU time is unchanged.
  • The narrower pass-start barriers barely show on this route: 426 full barriers a frame against 438.
  • No flashes seen in either run (watched by the player, not counted with the frame log), no fatal or device loss, and both ended with a clean quit.

A guess, not checked: with 283fbde2 the one render thread records every command buffer, and on this Windows NVIDIA PC main's parallel recording may have been doing real work (your Linux calibration had already picked 0 workers). Main was run once on this route; both of the PR's runs agree with each other.

A first Windows run of this head `a2cc7b53` against main `868f6be8` with #1729: Windows 11, RTX 4070, NVIDIA 616.56, Ryzen 9 7900X (the CPU from your measurements). The two builds differ only in the PR (packs on one base). **The route:** a save, Debug tab → Teleport to Hyrule Field, Switch age to child, Set time 21:00, then about 2 minutes running around the field fighting the Stalchildren that spawn, with camera turns (it gives fights on demand: #1720, https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1493#issuecomment-56003). Settings: 2560x1440, DLSS Quality, cap 90, frame generation off, Reflex and V-Sync on. Switches: `RX_VK_PASS_TIMING=1` `RX_VK_SLOW_LOG=50` `HOS_THREAD_TIMES=5` `HOS_FPS_LOG=1` on every run. A person played; numbers come from 120-present windows from the field's arrival, and a fight window has 200 or more dispatches a frame. | | main | this PR, run 1 | this PR, run 2 | |---|---|---|---| | fight fps (median, lowest) | 69.4, 55.3 (11 windows) | 65.2, 50.7 (8) | 64.5, 55.7 (17) | | other play fps (median) | 76.9 | 77.2 | 75.4 | | replay CPU work a frame (`a frame's replay time`, submit, median) | 7.30 ms | 9.72 ms | 10.05 ms | | per draw (that over the median draws, ~2,100) | 3.4 µs | 4.7 µs | 4.8 µs | | host time in GPU resource waits (median) | 2.19 ms | 5.99 ms | 6.33 ms | | GPU time a frame (median) | 8.6 ms | 8.7 ms | 8.9 ms | | full barriers a frame (median) | 438 | 426 | — | - **Slower in fights here:** 4-5 fps lower, and calm play level with main. - **Mostly on the render thread:** its replay work is ~35% higher at the same draw count, and it waits ~4 ms more a frame on the GPU, while GPU time is unchanged. - **The narrower pass-start barriers barely show on this route:** 426 full barriers a frame against 438. - **No flashes seen** in either run (watched by the player, not counted with the frame log), no fatal or device loss, and both ended with a clean quit. A guess, not checked: with 283fbde2 the one render thread records every command buffer, and on this Windows NVIDIA PC main's parallel recording may have been doing real work (your Linux calibration had already picked 0 workers). Main was run once on this route; both of the PR's runs agree with each other.
Contributor

A correction to my guess above: on main the work wasn't parallel segments but the recorder thread. In main's run its vk: recorder: lines show it busy 0.5-1.4 s in every 10 s (about 1-2 ms a frame of Vulkan calls taken off the replay), with the replay waiting for it only 50-100 ms in 10 s. This PR removes the recorder, which would put those calls back on the render thread, about the +2.4 ms of replay work measured here. So on this Windows NVIDIA PC the recorder was a net gain, not a loss as on the i7-9700K in #1745.

A correction to my guess above: on main the work wasn't parallel segments but the recorder thread. In main's run its `vk: recorder:` lines show it busy 0.5-1.4 s in every 10 s (about 1-2 ms a frame of Vulkan calls taken off the replay), with the replay waiting for it only 50-100 ms in 10 s. This PR removes the recorder, which would put those calls back on the render thread, about the +2.4 ms of replay work measured here. So on this Windows NVIDIA PC the recorder was a net gain, not a loss as on the i7-9700K in #1745.
HumanBean force-pushed perf/thread-redesign from a2cc7b53ec
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 14s
PR description / Description follows the template (pull_request_target) Successful in 20s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 20s
CI / Workflow lint (actionlint) (pull_request) Successful in 23s
Third-party parts / As their notes say (pull_request) Successful in 24s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+amd 1/1 fail
beanstalk/windows+nvidia 1/1 fail
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 2/1 fail
beanstalk/linux+nvidia 0/1 pass, 0 running, 1 queued
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 1 running, 1 queued
beanstalk/amd+igpu 1/1 fail
to 21b606ef63
Some checks failed
beanstalk/windows+intel 0/1 pass, 0 running, 0 queued
beanstalk/linux+nvidia 0/1 pass, 0 running, 0 queued
beanstalk/steamdeck 0/1 pass, 0 running, 0 queued
beanstalk/windows+amd 0/1 pass, 0 running, 0 queued
beanstalk/linux+amd 0/1 pass, 0 running, 0 queued
beanstalk/linux+intel 0/1 pass, 0 running, 0 queued
beanstalk/amd+igpu 0/1 pass, 0 running, 0 queued
beanstalk/windows+nvidia 0/1 pass, 0 running, 0 queued
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
Third-party parts / As their notes say (pull_request) Successful in 34s
PR description / Description follows the template (pull_request_target) Successful in 16s
CI / Conventional Commit title (pull_request) Successful in 21s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 30s
Website / Website on Cloudflare (pull_request_target) Has been skipped
CI / Workflow lint (actionlint) (pull_request) Successful in 37s
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt 1/1 fail
beanstalk/windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt 1/1 fail
CI / Website build and checks (pull_request) Has been cancelled
2026-11-20 00:41:24 +00:00
Compare
Member
Beanstalk test farm · an earlier run: commit 21b606e (the newer results are in a comment below)

Commit 21b606e · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

Two starts per side compare the PR’s default renderer with main in Kokiri Forest, Hyrule Field and the Market, using uncapped frame rates and V-Sync off; the route takes about 23 minutes per tester. It measures standing FPS, p99 frame time and stutters and compares pictures; crashes, any FPS decrease, a p99 increase over 15%, a stutter increase over 50%, or excessive picture differences fail the plan. The requested Intel targets will wait for matching testers. Camera turning during measurement, the AMD-only diagnostic cells and Steam Deck flash counts cannot be expressed by this plan; the farm forbids the frame-log switch. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main; 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate.
  • The PR fails if: it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or stutters is more than 50% above main; or frame time (p99) is more than 15% above main.
  • Time: about 25 min of game runs per tester (4 game starts), plus the two builds.

Started at 00:43 UTC.

Estimate: done in a few minutes, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd · 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 1/1 1 none 12.8 vs 12.7 (+1%) — fail
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4080, 621.1.0; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3090, 621.24.0 2/1 1 2 none 16.4 vs 16.2 (+1%) — fail
windows+intel 0/1 — — — queued
linux+amd · 1: linux+amd+radv+dgpu AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6 1/1 1 none 10.9 vs 16.0 (−32%) — fail
linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 1/1 1 none — — queued
linux+intel 0/1 — — — queued
steamdeck · 1: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 1/1 1 none 45.7 vs 55.9 (−18%) — fail
amd+igpu · 1: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8; 2: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 2/1 1 2 none 61.8 vs 71.0 (−13%) — fail ⚠ outlier

Measurements · windows+amd

Verdict: fail on windows+amd (1/1 report): three-areas field-standing.fps_mean.min_delta_pct (-0.5%); three-areas market-standing.fps_mean.min_delta_pct (-10.9%); three-areas screenshot.max_diff failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
three-areas · field-standing · frame rate 92.9 92.4 (−1%)
three-areas · forest-standing · frame rate 84.1 92.5 (+10%)
three-areas · market-standing · frame rate 64.5 57.5 (−11%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 0.0 0.0
three-areas · market-standing · stutters 1.0 0.0 (−100%)
three-areas · field-standing · frame time (p99) (ms) 12.7 12.8 (+1%)
three-areas · forest-standing · frame time (p99) (ms) 14.7 13.2 (−10%)
three-areas · market-standing · frame time (p99) (ms) 21.4 21.3 (−1%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+amd (1/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.5%)`; `three-areas market-standing.fps_mean.min_delta_pct (-10.9%)`; `three-areas screenshot.max_diff` failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| three-areas · field-standing · frame rate | 92.9 | 92.4 (−1%) |
| three-areas · forest-standing · frame rate | 84.1 | 92.5 (+10%) |
| three-areas · market-standing · frame rate | 64.5 | 57.5 (−11%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 0.0 | 0.0 |
| three-areas · market-standing · stutters | 1.0 | 0.0 (−100%) |
| three-areas · field-standing · frame time (p99) (ms) | 12.7 | 12.8 (+1%) |
| three-areas · forest-standing · frame time (p99) (ms) | 14.7 | 13.2 (−10%) |
| three-areas · market-standing · frame time (p99) (ms) | 21.4 | 21.3 (−1%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · windows+nvidia

Verdict: fail on windows+nvidia (2/1 report): three-areas field-standing.fps_mean.min_delta_pct (-0.1%); three-areas market-standing.fps_mean.min_delta_pct (-0.2%) failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4080, 621.1.0 AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3090, 621.24.0 AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
three-areas · field-standing · frame rate 80.7 80.7 (−0%)
three-areas · forest-standing · frame rate 70.2 74.4 (+6%)
three-areas · market-standing · frame rate 62.3 62.2 (−0%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 1.0 0.0 (−100%)
three-areas · market-standing · stutters 1.5 0.0 (−100%)
three-areas · field-standing · frame time (p99) (ms) 16.2 16.4 (+1%)
three-areas · forest-standing · frame time (p99) (ms) 21.4 18.8 (−12%)
three-areas · market-standing · frame time (p99) (ms) 21.7 21.6 (−1%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.1%)`; `three-areas market-standing.fps_mean.min_delta_pct (-0.2%)` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4080, 621.1.0` `AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| three-areas · field-standing · frame rate | 80.7 | 80.7 (−0%) |
| three-areas · forest-standing · frame rate | 70.2 | 74.4 (+6%) |
| three-areas · market-standing · frame rate | 62.3 | 62.2 (−0%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 1.0 | 0.0 (−100%) |
| three-areas · market-standing · stutters | 1.5 | 0.0 (−100%) |
| three-areas · field-standing · frame time (p99) (ms) | 16.2 | 16.4 (+1%) |
| three-areas · forest-standing · frame time (p99) (ms) | 21.4 | 18.8 (−12%) |
| three-areas · market-standing · frame time (p99) (ms) | 21.7 | 21.6 (−1%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+amd

Verdict: fail on linux+amd (1/1 report): three-areas market-standing.fps_mean.min_delta_pct (-14.9%); three-areas forest-standing.stutter_count.max_delta_pct; three-areas field-standing.stutter_count.max_delta_pct failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6 11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
three-areas · field-standing · frame rate 76.0 104.6 (+38%)
three-areas · forest-standing · frame rate 87.7 110.5 (+26%)
three-areas · market-standing · frame rate 72.9 62.0 (−15%)
three-areas · field-standing · stutters 0.0 1.0
three-areas · forest-standing · stutters 0.0 0.5
three-areas · market-standing · stutters 3.5 28.0 (+700%)
three-areas · field-standing · frame time (p99) (ms) 16.0 10.9 (−32%)
three-areas · forest-standing · frame time (p99) (ms) 13.7 11.1 (−19%)
three-areas · market-standing · frame time (p99) (ms) 18.3 46.3 (+154%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-14.9%)`; `three-areas forest-standing.stutter_count.max_delta_pct`; `three-areas field-standing.stutter_count.max_delta_pct` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| three-areas · field-standing · frame rate | 76.0 | 104.6 (+38%) |
| three-areas · forest-standing · frame rate | 87.7 | 110.5 (+26%) |
| three-areas · market-standing · frame rate | 72.9 | 62.0 (−15%) |
| three-areas · field-standing · stutters | 0.0 | 1.0 |
| three-areas · forest-standing · stutters | 0.0 | 0.5 |
| three-areas · market-standing · stutters | 3.5 | 28.0 (+700%) |
| three-areas · field-standing · frame time (p99) (ms) | 16.0 | 10.9 (−32%) |
| three-areas · forest-standing · frame time (p99) (ms) | 13.7 | 11.1 (−19%) |
| three-areas · market-standing · frame time (p99) (ms) | 18.3 | 46.3 (+154%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+nvidia

Verdict: queued on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Nothing was measured: no metric came back.

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** queued on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Nothing was measured: no metric came back.

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · steamdeck

Verdict: fail on steamdeck (1/1 report): three-areas screenshot.max_diff; three-areas forest-standing.stutter_count.max_delta_pct (+200.0%) failed.

Hardware: 1: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 AMD Custom APU 0405, 12 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
three-areas · field-standing · frame rate 20.9 24.1 (+15%)
three-areas · forest-standing · frame rate 22.6 23.9 (+6%)
three-areas · market-standing · frame rate 16.5 17.7 (+7%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 0.5 1.5 (+200%)
three-areas · market-standing · stutters 6.0 4.0 (−33%)
three-areas · field-standing · frame time (p99) (ms) 55.9 45.7 (−18%)
three-areas · forest-standing · frame time (p99) (ms) 57.8 58.6 (+1%)
three-areas · market-standing · frame time (p99) (ms) 152.3 103.2 (−32%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on steamdeck (1/1 report): `three-areas screenshot.max_diff`; `three-areas forest-standing.stutter_count.max_delta_pct (+200.0%)` failed.

Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| three-areas · field-standing · frame rate | 20.9 | 24.1 (+15%) |
| three-areas · forest-standing · frame rate | 22.6 | 23.9 (+6%) |
| three-areas · market-standing · frame rate | 16.5 | 17.7 (+7%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 0.5 | 1.5 (+200%) |
| three-areas · market-standing · stutters | 6.0 | 4.0 (−33%) |
| three-areas · field-standing · frame time (p99) (ms) | 55.9 | 45.7 (−18%) |
| three-areas · forest-standing · frame time (p99) (ms) | 57.8 | 58.6 (+1%) |
| three-areas · market-standing · frame time (p99) (ms) | 152.3 | 103.2 (−32%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · amd+igpu

Verdict: fail on amd+igpu (2/1 report): three-areas screenshot.max_diff failed.

Hardware: 1: linux+amd+radv+igpu AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8 AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM; 2: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 AMD Custom APU 0405, 12 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
three-areas · field-standing · frame rate 20.0 22.0 (+10%)
three-areas · forest-standing · frame rate 20.0 23.0 (+15%)
three-areas · market-standing · frame rate 17.0 17.4 (+2%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 1.0 0.8 (−25%)
three-areas · market-standing · stutters 1.5 1.2 (−17%)
three-areas · field-standing · frame time (p99) (ms) 71.0 61.8 (−13%)
three-areas · forest-standing · frame time (p99) (ms) 72.8 67.7 (−7%)
three-areas · market-standing · frame time (p99) (ms) 101.1 95.2 (−6%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff` failed.

Hardware: 1: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM`; 2: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| three-areas · field-standing · frame rate | 20.0 | 22.0 (+10%) |
| three-areas · forest-standing · frame rate | 20.0 | 23.0 (+15%) |
| three-areas · market-standing · frame rate | 17.0 | 17.4 (+2%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 1.0 | 0.8 (−25%) |
| three-areas · market-standing · stutters | 1.5 | 1.2 (−17%) |
| three-areas · field-standing · frame time (p99) (ms) | 71.0 | 61.8 (−13%) |
| three-areas · forest-standing · frame time (p99) (ms) | 72.8 | 67.7 (−7%) |
| three-areas · market-standing · frame time (p99) (ms) | 101.1 | 95.2 (−6%) |

### Flags and settings

None: the plan varied no setting, switch or build option.
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>21b606e</code> (the newer results are in a comment below)</summary> Commit `21b606e` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) Two starts per side compare the PR’s default renderer with main in Kokiri Forest, Hyrule Field and the Market, using uncapped frame rates and V-Sync off; the route takes about 23 minutes per tester. It measures standing FPS, p99 frame time and stutters and compares pictures; crashes, any FPS decrease, a p99 increase over 15%, a stutter increase over 50%, or excessive picture differences fail the plan. The requested Intel targets will wait for matching testers. Camera turning during measurement, the AMD-only diagnostic cells and Steam Deck flash counts cannot be expressed by this plan; the farm forbids the frame-log switch. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main; 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate. - **The PR fails if:** it crashes; or frame rate is more than 0% below main; or a picture differs from main by more than 5%; or stutters is more than 50% above main; or frame time (p99) is more than 15% above main. - **Time:** about 25 min of game runs per tester (4 game starts), plus the two builds. Started at 00:43 UTC. **Estimate:** done in a few minutes, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd · 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_0b2c30e6fd8302f0) | none | 12.8 vs 12.7 (+1%) | — | fail | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4080, 621.1.0`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_ca2a478b5d338a8d) [2](https://beanstalk.magicbean.dev/reports/rp_18ae2dddc053534c) | none | 16.4 vs 16.2 (+1%) | — | fail | | windows+intel | 0/1 | — | — | — | queued | | linux+amd · 1: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_d6bffe70774fb1e0) | none | 10.9 vs 16.0 (−32%) | — | fail | | linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_1785830c5a9dffd9) | none | — | — | queued | | linux+intel | 0/1 | — | — | — | queued | | steamdeck · 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_b43bad86ac2dac8b) | none | 45.7 vs 55.9 (−18%) | — | fail | | amd+igpu · 1: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8`; 2: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_dea048ff0a3001bf) [2](https://beanstalk.magicbean.dev/reports/rp_9bec439f939ab519) | none | 61.8 vs 71.0 (−13%) | — | fail ⚠ outlier | #### Measurements · windows+amd **Verdict:** fail on windows+amd (1/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.5%)`; `three-areas market-standing.fps_mean.min_delta_pct (-10.9%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 92.9 | 92.4 (−1%) | | three-areas · forest-standing · frame rate | 84.1 | 92.5 (+10%) | | three-areas · market-standing · frame rate | 64.5 | 57.5 (−11%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.0 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 12.7 | 12.8 (+1%) | | three-areas · forest-standing · frame time (p99) (ms) | 14.7 | 13.2 (−10%) | | three-areas · market-standing · frame time (p99) (ms) | 21.4 | 21.3 (−1%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+amd (1/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.5%)`; `three-areas market-standing.fps_mean.min_delta_pct (-10.9%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 92.9 | 92.4 (−1%) | | three-areas · forest-standing · frame rate | 84.1 | 92.5 (+10%) | | three-areas · market-standing · frame rate | 64.5 | 57.5 (−11%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.0 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 12.7 | 12.8 (+1%) | | three-areas · forest-standing · frame time (p99) (ms) | 14.7 | 13.2 (−10%) | | three-areas · market-standing · frame time (p99) (ms) | 21.4 | 21.3 (−1%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · windows+nvidia **Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.1%)`; `three-areas market-standing.fps_mean.min_delta_pct (-0.2%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4080, 621.1.0` `AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 80.7 | 80.7 (−0%) | | three-areas · forest-standing · frame rate | 70.2 | 74.4 (+6%) | | three-areas · market-standing · frame rate | 62.3 | 62.2 (−0%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 1.0 | 0.0 (−100%) | | three-areas · market-standing · stutters | 1.5 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 16.2 | 16.4 (+1%) | | three-areas · forest-standing · frame time (p99) (ms) | 21.4 | 18.8 (−12%) | | three-areas · market-standing · frame time (p99) (ms) | 21.7 | 21.6 (−1%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+nvidia (2/1 report): `three-areas field-standing.fps_mean.min_delta_pct (-0.1%)`; `three-areas market-standing.fps_mean.min_delta_pct (-0.2%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4080, 621.1.0` `AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3090, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 128 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 80.7 | 80.7 (−0%) | | three-areas · forest-standing · frame rate | 70.2 | 74.4 (+6%) | | three-areas · market-standing · frame rate | 62.3 | 62.2 (−0%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 1.0 | 0.0 (−100%) | | three-areas · market-standing · stutters | 1.5 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 16.2 | 16.4 (+1%) | | three-areas · forest-standing · frame time (p99) (ms) | 21.4 | 18.8 (−12%) | | three-areas · market-standing · frame time (p99) (ms) | 21.7 | 21.6 (−1%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+amd **Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-14.9%)`; `three-areas forest-standing.stutter_count.max_delta_pct`; `three-areas field-standing.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 76.0 | 104.6 (+38%) | | three-areas · forest-standing · frame rate | 87.7 | 110.5 (+26%) | | three-areas · market-standing · frame rate | 72.9 | 62.0 (−15%) | | three-areas · field-standing · stutters | 0.0 | 1.0 | | three-areas · forest-standing · stutters | 0.0 | 0.5 | | three-areas · market-standing · stutters | 3.5 | 28.0 (+700%) | | three-areas · field-standing · frame time (p99) (ms) | 16.0 | 10.9 (−32%) | | three-areas · forest-standing · frame time (p99) (ms) | 13.7 | 11.1 (−19%) | | three-areas · market-standing · frame time (p99) (ms) | 18.3 | 46.3 (+154%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-14.9%)`; `three-areas forest-standing.stutter_count.max_delta_pct`; `three-areas field-standing.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 9070 XT (RADV GFX1201), 26.2.6` `11th Gen Intel(R) Core(TM) i9-11900KB @ 3.30GHz, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 76.0 | 104.6 (+38%) | | three-areas · forest-standing · frame rate | 87.7 | 110.5 (+26%) | | three-areas · market-standing · frame rate | 72.9 | 62.0 (−15%) | | three-areas · field-standing · stutters | 0.0 | 1.0 | | three-areas · forest-standing · stutters | 0.0 | 0.5 | | three-areas · market-standing · stutters | 3.5 | 28.0 (+700%) | | three-areas · field-standing · frame time (p99) (ms) | 16.0 | 10.9 (−32%) | | three-areas · forest-standing · frame time (p99) (ms) | 13.7 | 11.1 (−19%) | | three-areas · market-standing · frame time (p99) (ms) | 18.3 | 46.3 (+154%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+nvidia **Verdict:** queued on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** queued on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · steamdeck **Verdict:** fail on steamdeck (1/1 report): `three-areas screenshot.max_diff`; `three-areas forest-standing.stutter_count.max_delta_pct (+200.0%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 20.9 | 24.1 (+15%) | | three-areas · forest-standing · frame rate | 22.6 | 23.9 (+6%) | | three-areas · market-standing · frame rate | 16.5 | 17.7 (+7%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.5 | 1.5 (+200%) | | three-areas · market-standing · stutters | 6.0 | 4.0 (−33%) | | three-areas · field-standing · frame time (p99) (ms) | 55.9 | 45.7 (−18%) | | three-areas · forest-standing · frame time (p99) (ms) | 57.8 | 58.6 (+1%) | | three-areas · market-standing · frame time (p99) (ms) | 152.3 | 103.2 (−32%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on steamdeck (1/1 report): `three-areas screenshot.max_diff`; `three-areas forest-standing.stutter_count.max_delta_pct (+200.0%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 20.9 | 24.1 (+15%) | | three-areas · forest-standing · frame rate | 22.6 | 23.9 (+6%) | | three-areas · market-standing · frame rate | 16.5 | 17.7 (+7%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.5 | 1.5 (+200%) | | three-areas · market-standing · stutters | 6.0 | 4.0 (−33%) | | three-areas · field-standing · frame time (p99) (ms) | 55.9 | 45.7 (−18%) | | three-areas · forest-standing · frame time (p99) (ms) | 57.8 | 58.6 (+1%) | | three-areas · market-standing · frame time (p99) (ms) | 152.3 | 103.2 (−32%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · amd+igpu **Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff` failed. Hardware: 1: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM`; 2: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 20.0 | 22.0 (+10%) | | three-areas · forest-standing · frame rate | 20.0 | 23.0 (+15%) | | three-areas · market-standing · frame rate | 17.0 | 17.4 (+2%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 1.0 | 0.8 (−25%) | | three-areas · market-standing · stutters | 1.5 | 1.2 (−17%) | | three-areas · field-standing · frame time (p99) (ms) | 71.0 | 61.8 (−13%) | | three-areas · forest-standing · frame time (p99) (ms) | 72.8 | 67.7 (−7%) | | three-areas · market-standing · frame time (p99) (ms) | 101.1 | 95.2 (−6%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff` failed. Hardware: 1: linux+amd+radv+igpu `AMD Radeon 760M Graphics (RADV PHOENIX), 26.0.8` `AMD Ryzen 5 230 w/ Radeon 760M Graphics, 16 GB RAM`; 2: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | three-areas · field-standing · frame rate | 20.0 | 22.0 (+10%) | | three-areas · forest-standing · frame rate | 20.0 | 23.0 (+15%) | | three-areas · market-standing · frame rate | 17.0 | 17.4 (+2%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 1.0 | 0.8 (−25%) | | three-areas · market-standing · stutters | 1.5 | 1.2 (−17%) | | three-areas · field-standing · frame time (p99) (ms) | 71.0 | 61.8 (−13%) | | three-areas · forest-standing · frame time (p99) (ms) | 72.8 | 67.7 (−7%) | | three-areas · market-standing · frame time (p99) (ms) | 101.1 | 95.2 (−6%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> </details>
HumanBean changed title from WIP: perf(vk): the renderer's threads redesigned: segment workers measured in play, the presenter on every PC, narrower pass barriers to WIP: perf(vk): the renderer's threads rewritten: one render thread, the presenter on every PC, parts that end at the game's own pass ends 2026-11-20 00:42:05 +00:00
Author
Owner

@Beanstalk (new head 21b606e: rebased onto main 068fb8eb as one commit; it adds the Steam Deck's late readbacks from #1725.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it (twice the PR side was measured still at the save's start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main.

Cells: the PR's defaults on every platform. On linux+amd and windows+amd only, three more PR-only cells, to find what makes the Market standing slower there (-24.5% on an RX 7900 XTX, -14.5% on an RX 9070 XT last time): RX_VK_PART_MIN=240 RX_VK_PART_MAX=240; RX_VK_PASS_BARRIER=full; RX_VK_READBACK_LATE=0. On the steamdeck, count the frame log's flashes (RX_VK_FRAME_LOG) on both sides: the PR reads the newest submission's results late there, the exposure meter among them, and expects no more flashes than main.

@Beanstalk (new head 21b606e: rebased onto main 068fb8eb as one commit; it adds the Steam Deck's late readbacks from #1725.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it (twice the PR side was measured still at the save's start). Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. Cells: the PR's defaults on every platform. On linux+amd and windows+amd only, three more PR-only cells, to find what makes the Market standing slower there (-24.5% on an RX 7900 XTX, -14.5% on an RX 9070 XT last time): `RX_VK_PART_MIN=240 RX_VK_PART_MAX=240`; `RX_VK_PASS_BARRIER=full`; `RX_VK_READBACK_LATE=0`. On the steamdeck, count the frame log's flashes (RX_VK_FRAME_LOG) on both sides: the PR reads the newest submission's results late there, the exposure meter among them, and expects no more flashes than main.
Author
Owner

@Beanstalk plan only linux+amd and windows+amd (an RX 9070 XT, RX 9060 XT or RX 7900 XTX each), on this head 21b606e: the Market standing loses 11-15% against main there while every other platform gains, and the renderer's own timing points at the replay's CPU cost per draw on AMD's drivers. From the kokiri-start save, teleport to Castle Town Market and make sure the game has reached it, wait 15 s, then measure 40 s standing still; vsync=2 (Off), frame_rate_cap=6 (Uncapped). Cells, each on main and the PR: the defaults; RX_VK_PROF_FINE=1 (logs where each draw's replay time goes, every few seconds: keep the whole log of both sides); on the PR only, RX_VK_PART_MIN=240 RX_VK_PART_MAX=240, RX_VK_PASS_BARRIER=full and RX_VK_READBACK_LATE=0. Two game starts a side. Keep Windows Intel and Linux Intel queued as they are.

@Beanstalk plan only linux+amd and windows+amd (an RX 9070 XT, RX 9060 XT or RX 7900 XTX each), on this head 21b606e: the Market standing loses 11-15% against main there while every other platform gains, and the renderer's own timing points at the replay's CPU cost per draw on AMD's drivers. From the `kokiri-start` save, teleport to Castle Town Market and make sure the game has reached it, wait 15 s, then measure 40 s standing still; vsync=2 (Off), frame_rate_cap=6 (Uncapped). Cells, each on main and the PR: the defaults; `RX_VK_PROF_FINE=1` (logs where each draw's replay time goes, every few seconds: keep the whole log of both sides); on the PR only, `RX_VK_PART_MIN=240 RX_VK_PART_MAX=240`, `RX_VK_PASS_BARRIER=full` and `RX_VK_READBACK_LATE=0`. Two game starts a side. Keep Windows Intel and Linux Intel queued as they are.
Member
Beanstalk test farm · an earlier run: commit 21b606e, run 2 (the newer results are in a comment below)

Commit 21b606e · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

On Linux and Windows AMD PCs with one of the three requested GPUs, the plan compares standing in Castle Town Market on the PR and main, with V-Sync off and the frame rate uncapped. Two runs per side measure FPS and frame times after a 15-second wait, and compare a screenshot; the profiling axis also collects the renderer’s detailed replay logs. The run takes about 44 minutes per tester and fails on a crash, a large picture difference, a 5% FPS loss, or a 15% p99 frame-time increase. The three PR-only rollback experiments are not included, and existing Intel queues are not changed by this plan.

  • Where: linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (GPU "rx 7900 xtx" or "rx 9060 xt" or "rx 9070 xt", 1-2 reports), windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (GPU "rx 7900 xtx" or "rx 9060 xt" or "rx 9070 xt", 1-2 reports).
  • Matrix: 2 cells, each on main and the PR: RX_VK_PROF_FINE 0 or 1.
  • market-standing: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Market, take the picture market to compare with main, measure standing for 40 s (after 0 s warm-up); 2 runs on each side for each of 2 cells (2 game starts a run).
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate, the game's log.
  • The PR fails if: it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 10%; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0.
  • Time: about 48 min of game runs per tester (8 game starts), plus the two builds.

Started at 02:17 UTC. Run 2: re-planned by HumanBean.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt · 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 1/1 1 none 64.7 vs 18.7 (+247%) — fail
windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt · 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 1/1 1 none 21.4 vs 19.8 (+8%) — fail
Scenario · matrix cell Target Frame p99 (ms) vs main Result
market-standing · RX_VK_PROF_FINE=0 linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt 64.7 vs 18.7 (+247%) fail
market-standing · RX_VK_PROF_FINE=0 windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt 21.4 vs 19.8 (+8%) fail
market-standing · RX_VK_PROF_FINE=1 linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt 65.9 vs 18.4 (+258%) fail
market-standing · RX_VK_PROF_FINE=1 windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt — pass

Measurements · linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt

Verdict: fail on linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): market-standing standing.fps_mean.min_delta_pct (-18.9%); market-standing screenshot.max_diff; market-standing standing.frametime_p99_ms.max_delta_pct (+246.6%) failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM
Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
market-standing · RX_VK_PROF_FINE=0 · standing · frame rate 64.8 52.6 (−19%)
market-standing · RX_VK_PROF_FINE=0 · standing · frame time (p99) (ms) 18.7 64.7 (+247%)
market-standing · RX_VK_PROF_FINE=0 · forbidden log lines 0.0 0.0
market-standing · RX_VK_PROF_FINE=1 · standing · frame rate 65.0 46.2 (−29%)
market-standing · RX_VK_PROF_FINE=1 · standing · frame time (p99) (ms) 18.4 65.9 (+258%)
market-standing · RX_VK_PROF_FINE=1 · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_PROF_FINE switch 0, 1 main's default the PR's default market-standing, the PR at 0 → at 1: 52.6 → 46.2 (−12%) doesn't help
Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-18.9%)`; `market-standing screenshot.max_diff`; `market-standing standing.frametime_p99_ms.max_delta_pct (+246.6%)` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`  
Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 64.8 | 52.6 (−19%) |
| market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 18.7 | 64.7 (+247%) |
| market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 |
| market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 65.0 | 46.2 (−29%) |
| market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 18.4 | 65.9 (+258%) |
| market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 |

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: 52.6 → 46.2 (−12%) | doesn't help |

Measurements · windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt

Verdict: fail on windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): market-standing standing.fps_mean.min_delta_pct (-12.1%) failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM
Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 21b606e
market-standing · RX_VK_PROF_FINE=0 · standing · frame rate 65.4 57.4 (−12%)
market-standing · RX_VK_PROF_FINE=0 · standing · frame time (p99) (ms) 19.8 21.4 (+8%)
market-standing · RX_VK_PROF_FINE=0 · forbidden log lines 0.0 0.0
market-standing · RX_VK_PROF_FINE=1 · standing · frame rate 62.8 —
market-standing · RX_VK_PROF_FINE=1 · standing · frame time (p99) (ms) 20.2 —
market-standing · RX_VK_PROF_FINE=1 · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_PROF_FINE switch 0, 1 main's default the PR's default market-standing, the PR at 0 → at 1: not measured untested
Markdown for the PR description
## Measurements

**Verdict:** fail on windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-12.1%)` failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM`  
Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 21b606e |
|---|---|---|
| market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 65.4 | 57.4 (−12%) |
| market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 19.8 | 21.4 (+8%) |
| market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 |
| market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 62.8 | — |
| market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 20.2 | — |
| market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 |

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: not measured | untested |
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>21b606e</code>, run 2 (the newer results are in a comment below)</summary> Commit `21b606e` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) On Linux and Windows AMD PCs with one of the three requested GPUs, the plan compares standing in Castle Town Market on the PR and main, with V-Sync off and the frame rate uncapped. Two runs per side measure FPS and frame times after a 15-second wait, and compare a screenshot; the profiling axis also collects the renderer’s detailed replay logs. The run takes about 44 minutes per tester and fails on a crash, a large picture difference, a 5% FPS loss, or a 15% p99 frame-time increase. The three PR-only rollback experiments are not included, and existing Intel queues are not changed by this plan. - **Where:** linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (GPU "rx 7900 xtx" or "rx 9060 xt" or "rx 9070 xt", 1-2 reports), windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (GPU "rx 7900 xtx" or "rx 9060 xt" or "rx 9070 xt", 1-2 reports). - **Matrix:** 2 cells, each on main and the PR: RX_VK_PROF_FINE `0` or `1`. - **market-standing:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Market, take the picture market to compare with main, measure standing for 40 s (after 0 s warm-up); 2 runs on each side for each of 2 cells (2 game starts a run). - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate, the game's log. - **The PR fails if:** it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 10%; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0. - **Time:** about 48 min of game runs per tester (8 game starts), plus the two builds. Started at 02:17 UTC. Run 2: re-planned by HumanBean. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt · 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_ee51f502e4beaa1a) | none | 64.7 vs 18.7 (+247%) | — | fail | | windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt · 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_6a40d25c92db3423) | none | 21.4 vs 19.8 (+8%) | — | fail | | Scenario · matrix cell | Target | Frame p99 (ms) vs main | Result | | --- | --- | --- | --- | | market-standing · `RX_VK_PROF_FINE=0` | linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt | 64.7 vs 18.7 (+247%) | fail | | market-standing · `RX_VK_PROF_FINE=0` | windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt | 21.4 vs 19.8 (+8%) | fail | | market-standing · `RX_VK_PROF_FINE=1` | linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt | 65.9 vs 18.4 (+258%) | fail | | market-standing · `RX_VK_PROF_FINE=1` | windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt | — | pass | #### Measurements · linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt **Verdict:** fail on linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-18.9%)`; `market-standing screenshot.max_diff`; `market-standing standing.frametime_p99_ms.max_delta_pct (+246.6%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 64.8 | 52.6 (−19%) | | market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 18.7 | 64.7 (+247%) | | market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 65.0 | 46.2 (−29%) | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 18.4 | 65.9 (+258%) | | market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: 52.6 → 46.2 (−12%) | doesn't help | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-18.9%)`; `market-standing screenshot.max_diff`; `market-standing standing.frametime_p99_ms.max_delta_pct (+246.6%)` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 64.8 | 52.6 (−19%) | | market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 18.7 | 64.7 (+247%) | | market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 65.0 | 46.2 (−29%) | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 18.4 | 65.9 (+258%) | | market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: 52.6 → 46.2 (−12%) | doesn't help | ``` </details> #### Measurements · windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt **Verdict:** fail on windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-12.1%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 65.4 | 57.4 (−12%) | | market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 19.8 | 21.4 (+8%) | | market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 62.8 | — | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 20.2 | — | | market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: not measured | untested | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+amd+gpu-rx-7900-xtx_rx-9060-xt_rx-9070-xt (1/1 report): `market-standing standing.fps_mean.min_delta_pct (-12.1%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: market-standing, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); market-standing: 2 game starts a side, standing 40 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 21b606e | |---|---|---| | market-standing · `RX_VK_PROF_FINE=0` · standing · frame rate | 65.4 | 57.4 (−12%) | | market-standing · `RX_VK_PROF_FINE=0` · standing · frame time (p99) (ms) | 19.8 | 21.4 (+8%) | | market-standing · `RX_VK_PROF_FINE=0` · forbidden log lines | 0.0 | 0.0 | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame rate | 62.8 | — | | market-standing · `RX_VK_PROF_FINE=1` · standing · frame time (p99) (ms) | 20.2 | — | | market-standing · `RX_VK_PROF_FINE=1` · forbidden log lines | 0.0 | 0.0 | ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_PROF_FINE` | switch | `0`, `1` | main's default | the PR's default | market-standing, the PR at 0 → at 1: not measured | untested | ``` </details> </details>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP
perf(vk): an indexed draw's index buffer bound at its buffer's start, its place given as the first index; diag: RX_VK_CALL_COUNT
Some checks failed
CI / Workflow lint (actionlint) (pull_request) Successful in 12s
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 10s
Website / Website on Cloudflare (pull_request_target) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
Third-party parts / As their notes say (pull_request) Successful in 14s
CI / Conventional Commit title (pull_request) Successful in 7s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
beanstalk/windows+nvidia 1/1 fail
CI / Website build and checks (pull_request) Has been cancelled
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+amd 1/1 fail
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 2/1 fail
beanstalk/linux+nvidia 0/1 pass, 1 running, 0 queued
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/amd+igpu 0/1 pass, 1 running, 1 queued
69263e4d66
On an RX 7900 XTX in the Market (2,500 draws a frame), the replay bound an index buffer for 0.97 of its draws: a pool's
draws share one VkBuffer at their own offsets, and with the offset in the bind each draw rebound it, a driver call each,
where AMD's drivers spend about twice NVIDIA's per call. With the offset given as the draw's first index (when it is a
whole number of indices) only a new buffer binds. RX_VK_INDEX_REBASE=0: as before. RX_VK_CALL_COUNT=1 (diag) counts the
replay's Vulkan calls by kind, a frame and a draw.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP
Member
Beanstalk test farm · an earlier run: commit 69263e4 (the newer results are in a comment below)

Commit 69263e4 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

Two starts per side compare main and the PR in Kokiri Forest, Hyrule Field and Market with V-Sync off, the frame cap uncapped and Threaded rendering Auto. The plan measures FPS, p99 frame time and stutters while standing and after camera turns, and compares pictures; crashes, performance regressions or picture differences fail. The requested AMD-only diagnostic cells and measurements during turns cannot be represented by this plan schema, so they are omitted. Each target takes about 23 minutes; Intel targets may wait for a matching tester. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure kokiri-standing for 20 s (after 5 s warm-up), take the picture kokiri-standing to compare with main, move the stick, measure kokiri-after-turn for 15 s (after 0 s warm-up), take the picture kokiri-after-turn to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field-standing to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-after-turn to compare with main, teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market-standing to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-after-turn to compare with main; 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate.
  • The PR fails if: it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 0%; or stutters is more than 20% above main; or frame time (p99) is more than 10% above main.
  • Time: about 28 min of game runs per tester (4 game starts), plus the two builds.

Started at 02:48 UTC.

Estimate: done in about 5 h, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd · 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 1/1 1 none 12.8 vs 13.1 (−2%) — fail
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0 2/1 1 2 1 PR run 13.4 vs 12.9 (+4%) — fail
windows+intel 0/1 — — — queued
linux+amd · 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7; 2: linux+amd+radv+dgpu AMD Steam Machine (RADV NAVI33), 26.1.99 2/1 1 2 none 27.1 vs 20.3 (+33%) — fail ⚠ outlier
linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 1/1 1 none — — running
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — queued
amd+igpu 0/1 — — — running

Measurements · windows+amd

Verdict: fail on windows+amd (1/1 report): three-areas market-standing.fps_mean.min_delta_pct (-25.2%); three-areas screenshot.max_diff failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 69263e4
three-areas · field-standing · frame rate 92.2 92.9 (+1%)
three-areas · kokiri-standing · frame rate 84.2 93.4 (+11%)
three-areas · market-standing · frame rate 64.4 48.1 (−25%)
three-areas · field-after-turn · frame rate 85.5 86.1 (+1%)
three-areas · kokiri-after-turn · frame rate 81.3 83.0 (+2%)
three-areas · market-after-turn · frame rate 76.6 81.2 (+6%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · kokiri-standing · stutters 0.0 0.0
three-areas · market-standing · stutters 1.5 0.5 (−67%)
three-areas · field-after-turn · stutters 0.0 0.0
three-areas · kokiri-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.5 0.0 (−100%)
three-areas · field-standing · frame time (p99) (ms) 13.1 12.8 (−2%)
three-areas · kokiri-standing · frame time (p99) (ms) 13.6 12.9 (−5%)
three-areas · market-standing · frame time (p99) (ms) 22.5 21.6 (−4%)
three-areas · field-after-turn · frame time (p99) (ms) 14.1 14.1 (−0%)
three-areas · kokiri-after-turn · frame time (p99) (ms) 14.4 15.1 (+5%)
three-areas · market-after-turn · frame time (p99) (ms) 17.4 17.1 (−2%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-25.2%)`; `three-areas screenshot.max_diff` failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 69263e4 |
|---|---|---|
| three-areas · field-standing · frame rate | 92.2 | 92.9 (+1%) |
| three-areas · kokiri-standing · frame rate | 84.2 | 93.4 (+11%) |
| three-areas · market-standing · frame rate | 64.4 | 48.1 (−25%) |
| three-areas · field-after-turn · frame rate | 85.5 | 86.1 (+1%) |
| three-areas · kokiri-after-turn · frame rate | 81.3 | 83.0 (+2%) |
| three-areas · market-after-turn · frame rate | 76.6 | 81.2 (+6%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · kokiri-standing · stutters | 0.0 | 0.0 |
| three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) |
| three-areas · field-after-turn · stutters | 0.0 | 0.0 |
| three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.5 | 0.0 (−100%) |
| three-areas · field-standing · frame time (p99) (ms) | 13.1 | 12.8 (−2%) |
| three-areas · kokiri-standing · frame time (p99) (ms) | 13.6 | 12.9 (−5%) |
| three-areas · market-standing · frame time (p99) (ms) | 22.5 | 21.6 (−4%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 14.1 | 14.1 (−0%) |
| three-areas · kokiri-after-turn · frame time (p99) (ms) | 14.4 | 15.1 (+5%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 17.4 | 17.1 (−2%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · windows+nvidia

Verdict: fail on windows+nvidia (2/1 report): three-areas crash; three-areas market-standing.fps_mean.min_delta_pct (-9.1%); three-areas screenshot.max_diff failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0 AMD Ryzen 7 7800X3D 8-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 69263e4
three-areas · field-standing · frame rate 95.0 92.0 (−3%)
three-areas · kokiri-standing · frame rate 87.1 95.0 (+9%)
three-areas · market-standing · frame rate 74.9 68.1 (−9%)
three-areas · field-after-turn · frame rate 102.7 107.1 (+4%)
three-areas · kokiri-after-turn · frame rate 77.0 76.0 (−1%)
three-areas · market-after-turn · frame rate 93.0 100.1 (+8%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · kokiri-standing · stutters 5.5 0.0 (−100%)
three-areas · market-standing · stutters 1.0 1.0 (+0%)
three-areas · field-after-turn · stutters 0.0 1.0
three-areas · kokiri-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.0 2.0
three-areas · field-standing · frame time (p99) (ms) 12.9 13.4 (+4%)
three-areas · kokiri-standing · frame time (p99) (ms) 16.9 13.7 (−19%)
three-areas · market-standing · frame time (p99) (ms) 17.5 19.4 (+11%)
three-areas · field-after-turn · frame time (p99) (ms) 13.9 11.8 (−15%)
three-areas · kokiri-after-turn · frame time (p99) (ms) 17.5 18.5 (+6%)
three-areas · market-after-turn · frame time (p99) (ms) 15.3 14.6 (−5%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas market-standing.fps_mean.min_delta_pct (-9.1%)`; `three-areas screenshot.max_diff` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0` `AMD Ryzen 7 7800X3D 8-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 69263e4 |
|---|---|---|
| three-areas · field-standing · frame rate | 95.0 | 92.0 (−3%) |
| three-areas · kokiri-standing · frame rate | 87.1 | 95.0 (+9%) |
| three-areas · market-standing · frame rate | 74.9 | 68.1 (−9%) |
| three-areas · field-after-turn · frame rate | 102.7 | 107.1 (+4%) |
| three-areas · kokiri-after-turn · frame rate | 77.0 | 76.0 (−1%) |
| three-areas · market-after-turn · frame rate | 93.0 | 100.1 (+8%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · kokiri-standing · stutters | 5.5 | 0.0 (−100%) |
| three-areas · market-standing · stutters | 1.0 | 1.0 (+0%) |
| three-areas · field-after-turn · stutters | 0.0 | 1.0 |
| three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.0 | 2.0 |
| three-areas · field-standing · frame time (p99) (ms) | 12.9 | 13.4 (+4%) |
| three-areas · kokiri-standing · frame time (p99) (ms) | 16.9 | 13.7 (−19%) |
| three-areas · market-standing · frame time (p99) (ms) | 17.5 | 19.4 (+11%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 13.9 | 11.8 (−15%) |
| three-areas · kokiri-after-turn · frame time (p99) (ms) | 17.5 | 18.5 (+6%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 15.3 | 14.6 (−5%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+amd

Verdict: fail on linux+amd (2/1 report): three-areas market-standing.fps_mean.min_delta_pct (-15.6%); three-areas screenshot.max_diff; three-areas kokiri-standing.stutter_count.max_delta_pct failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM; 2: linux+amd+radv+dgpu AMD Steam Machine (RADV NAVI33), 26.1.99 AMD Custom CPU 1772, 16 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 068fb8e After: this PR @ 69263e4
three-areas · field-standing · frame rate 67.3 77.1 (+15%)
three-areas · kokiri-standing · frame rate 60.1 78.9 (+31%)
three-areas · market-standing · frame rate 42.2 36.4 (−14%)
three-areas · field-after-turn · frame rate 69.2 82.8 (+20%)
three-areas · kokiri-after-turn · frame rate 61.0 75.9 (+24%)
three-areas · market-after-turn · frame rate 69.9 84.3 (+21%)
three-areas · field-standing · stutters 2.2 4.0 (+78%)
three-areas · kokiri-standing · stutters 0.8 0.8 (+0%)
three-areas · market-standing · stutters 71.0 100.5 (+42%)
three-areas · field-after-turn · stutters 0.2 6.8 (+2600%)
three-areas · kokiri-after-turn · stutters 0.5 21.8 (+4250%)
three-areas · market-after-turn · stutters 6.5 0.8 (−88%)
three-areas · field-standing · frame time (p99) (ms) 20.3 27.1 (+33%)
three-areas · kokiri-standing · frame time (p99) (ms) 20.5 16.3 (−20%)
three-areas · market-standing · frame time (p99) (ms) 52.9 74.5 (+41%)
three-areas · field-after-turn · frame time (p99) (ms) 18.2 27.2 (+49%)
three-areas · kokiri-after-turn · frame time (p99) (ms) 19.9 29.0 (+45%)
three-areas · market-after-turn · frame time (p99) (ms) 31.7 16.4 (−48%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd (2/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-15.6%)`; `three-areas screenshot.max_diff`; `three-areas kokiri-standing.stutter_count.max_delta_pct` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Steam Machine (RADV NAVI33), 26.1.99` `AMD Custom CPU 1772, 16 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 068fb8e | After: this PR @ 69263e4 |
|---|---|---|
| three-areas · field-standing · frame rate | 67.3 | 77.1 (+15%) |
| three-areas · kokiri-standing · frame rate | 60.1 | 78.9 (+31%) |
| three-areas · market-standing · frame rate | 42.2 | 36.4 (−14%) |
| three-areas · field-after-turn · frame rate | 69.2 | 82.8 (+20%) |
| three-areas · kokiri-after-turn · frame rate | 61.0 | 75.9 (+24%) |
| three-areas · market-after-turn · frame rate | 69.9 | 84.3 (+21%) |
| three-areas · field-standing · stutters | 2.2 | 4.0 (+78%) |
| three-areas · kokiri-standing · stutters | 0.8 | 0.8 (+0%) |
| three-areas · market-standing · stutters | 71.0 | 100.5 (+42%) |
| three-areas · field-after-turn · stutters | 0.2 | 6.8 (+2600%) |
| three-areas · kokiri-after-turn · stutters | 0.5 | 21.8 (+4250%) |
| three-areas · market-after-turn · stutters | 6.5 | 0.8 (−88%) |
| three-areas · field-standing · frame time (p99) (ms) | 20.3 | 27.1 (+33%) |
| three-areas · kokiri-standing · frame time (p99) (ms) | 20.5 | 16.3 (−20%) |
| three-areas · market-standing · frame time (p99) (ms) | 52.9 | 74.5 (+41%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 18.2 | 27.2 (+49%) |
| three-areas · kokiri-after-turn · frame time (p99) (ms) | 19.9 | 29.0 (+45%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 31.7 | 16.4 (−48%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+nvidia

Verdict: running on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Nothing was measured: no metric came back.

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** running on linux+nvidia (1/1 report).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Nothing was measured: no metric came back.

### Flags and settings

None: the plan varied no setting, switch or build option.
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>69263e4</code> (the newer results are in a comment below)</summary> Commit `69263e4` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) Two starts per side compare main and the PR in Kokiri Forest, Hyrule Field and Market with V-Sync off, the frame cap uncapped and Threaded rendering Auto. The plan measures FPS, p99 frame time and stutters while standing and after camera turns, and compares pictures; crashes, performance regressions or picture differences fail. The requested AMD-only diagnostic cells and measurements during turns cannot be represented by this plan schema, so they are omitted. Each target takes about 23 minutes; Intel targets may wait for a matching tester. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure kokiri-standing for 20 s (after 5 s warm-up), take the picture kokiri-standing to compare with main, move the stick, measure kokiri-after-turn for 15 s (after 0 s warm-up), take the picture kokiri-after-turn to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field-standing to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), take the picture field-after-turn to compare with main, teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market-standing to compare with main, measure market-after-turn for 15 s (after 0 s warm-up), take the picture market-after-turn to compare with main; 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate. - **The PR fails if:** it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 0%; or stutters is more than 20% above main; or frame time (p99) is more than 10% above main. - **Time:** about 28 min of game runs per tester (4 game starts), plus the two builds. Started at 02:48 UTC. **Estimate:** done in about 5 h, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd · 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_4cd053887e729564) | none | 12.8 vs 13.1 (−2%) | — | fail | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_7f84777e6f00d967) [2](https://beanstalk.magicbean.dev/reports/rp_d6671f9cc193c331) | 1 PR run | 13.4 vs 12.9 (+4%) | — | fail | | windows+intel | 0/1 | — | — | — | queued | | linux+amd · 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7`; 2: linux+amd+radv+dgpu `AMD Steam Machine (RADV NAVI33), 26.1.99` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_d2f84dc3f782ca1b) [2](https://beanstalk.magicbean.dev/reports/rp_ecb9e9784622e840) | none | 27.1 vs 20.3 (+33%) | — | fail ⚠ outlier | | linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_0d4d810e316340cf) | none | — | — | running | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | queued | | amd+igpu | 0/1 | — | — | — | running | #### Measurements · windows+amd **Verdict:** fail on windows+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-25.2%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 92.2 | 92.9 (+1%) | | three-areas · kokiri-standing · frame rate | 84.2 | 93.4 (+11%) | | three-areas · market-standing · frame rate | 64.4 | 48.1 (−25%) | | three-areas · field-after-turn · frame rate | 85.5 | 86.1 (+1%) | | three-areas · kokiri-after-turn · frame rate | 81.3 | 83.0 (+2%) | | three-areas · market-after-turn · frame rate | 76.6 | 81.2 (+6%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · kokiri-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.5 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 13.1 | 12.8 (−2%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 13.6 | 12.9 (−5%) | | three-areas · market-standing · frame time (p99) (ms) | 22.5 | 21.6 (−4%) | | three-areas · field-after-turn · frame time (p99) (ms) | 14.1 | 14.1 (−0%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 14.4 | 15.1 (+5%) | | three-areas · market-after-turn · frame time (p99) (ms) | 17.4 | 17.1 (−2%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-25.2%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 92.2 | 92.9 (+1%) | | three-areas · kokiri-standing · frame rate | 84.2 | 93.4 (+11%) | | three-areas · market-standing · frame rate | 64.4 | 48.1 (−25%) | | three-areas · field-after-turn · frame rate | 85.5 | 86.1 (+1%) | | three-areas · kokiri-after-turn · frame rate | 81.3 | 83.0 (+2%) | | three-areas · market-after-turn · frame rate | 76.6 | 81.2 (+6%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · kokiri-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.5 | 0.0 (−100%) | | three-areas · field-standing · frame time (p99) (ms) | 13.1 | 12.8 (−2%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 13.6 | 12.9 (−5%) | | three-areas · market-standing · frame time (p99) (ms) | 22.5 | 21.6 (−4%) | | three-areas · field-after-turn · frame time (p99) (ms) | 14.1 | 14.1 (−0%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 14.4 | 15.1 (+5%) | | three-areas · market-after-turn · frame time (p99) (ms) | 17.4 | 17.1 (−2%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · windows+nvidia **Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas market-standing.fps_mean.min_delta_pct (-9.1%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0` `AMD Ryzen 7 7800X3D 8-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 95.0 | 92.0 (−3%) | | three-areas · kokiri-standing · frame rate | 87.1 | 95.0 (+9%) | | three-areas · market-standing · frame rate | 74.9 | 68.1 (−9%) | | three-areas · field-after-turn · frame rate | 102.7 | 107.1 (+4%) | | three-areas · kokiri-after-turn · frame rate | 77.0 | 76.0 (−1%) | | three-areas · market-after-turn · frame rate | 93.0 | 100.1 (+8%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · kokiri-standing · stutters | 5.5 | 0.0 (−100%) | | three-areas · market-standing · stutters | 1.0 | 1.0 (+0%) | | three-areas · field-after-turn · stutters | 0.0 | 1.0 | | three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 2.0 | | three-areas · field-standing · frame time (p99) (ms) | 12.9 | 13.4 (+4%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 16.9 | 13.7 (−19%) | | three-areas · market-standing · frame time (p99) (ms) | 17.5 | 19.4 (+11%) | | three-areas · field-after-turn · frame time (p99) (ms) | 13.9 | 11.8 (−15%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 17.5 | 18.5 (+6%) | | three-areas · market-after-turn · frame time (p99) (ms) | 15.3 | 14.6 (−5%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+nvidia (2/1 report): `three-areas crash`; `three-areas market-standing.fps_mean.min_delta_pct (-9.1%)`; `three-areas screenshot.max_diff` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti SUPER, 621.14.0` `AMD Ryzen 7 7800X3D 8-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 95.0 | 92.0 (−3%) | | three-areas · kokiri-standing · frame rate | 87.1 | 95.0 (+9%) | | three-areas · market-standing · frame rate | 74.9 | 68.1 (−9%) | | three-areas · field-after-turn · frame rate | 102.7 | 107.1 (+4%) | | three-areas · kokiri-after-turn · frame rate | 77.0 | 76.0 (−1%) | | three-areas · market-after-turn · frame rate | 93.0 | 100.1 (+8%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · kokiri-standing · stutters | 5.5 | 0.0 (−100%) | | three-areas · market-standing · stutters | 1.0 | 1.0 (+0%) | | three-areas · field-after-turn · stutters | 0.0 | 1.0 | | three-areas · kokiri-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 2.0 | | three-areas · field-standing · frame time (p99) (ms) | 12.9 | 13.4 (+4%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 16.9 | 13.7 (−19%) | | three-areas · market-standing · frame time (p99) (ms) | 17.5 | 19.4 (+11%) | | three-areas · field-after-turn · frame time (p99) (ms) | 13.9 | 11.8 (−15%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 17.5 | 18.5 (+6%) | | three-areas · market-after-turn · frame time (p99) (ms) | 15.3 | 14.6 (−5%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+amd **Verdict:** fail on linux+amd (2/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-15.6%)`; `three-areas screenshot.max_diff`; `three-areas kokiri-standing.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Steam Machine (RADV NAVI33), 26.1.99` `AMD Custom CPU 1772, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 67.3 | 77.1 (+15%) | | three-areas · kokiri-standing · frame rate | 60.1 | 78.9 (+31%) | | three-areas · market-standing · frame rate | 42.2 | 36.4 (−14%) | | three-areas · field-after-turn · frame rate | 69.2 | 82.8 (+20%) | | three-areas · kokiri-after-turn · frame rate | 61.0 | 75.9 (+24%) | | three-areas · market-after-turn · frame rate | 69.9 | 84.3 (+21%) | | three-areas · field-standing · stutters | 2.2 | 4.0 (+78%) | | three-areas · kokiri-standing · stutters | 0.8 | 0.8 (+0%) | | three-areas · market-standing · stutters | 71.0 | 100.5 (+42%) | | three-areas · field-after-turn · stutters | 0.2 | 6.8 (+2600%) | | three-areas · kokiri-after-turn · stutters | 0.5 | 21.8 (+4250%) | | three-areas · market-after-turn · stutters | 6.5 | 0.8 (−88%) | | three-areas · field-standing · frame time (p99) (ms) | 20.3 | 27.1 (+33%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 20.5 | 16.3 (−20%) | | three-areas · market-standing · frame time (p99) (ms) | 52.9 | 74.5 (+41%) | | three-areas · field-after-turn · frame time (p99) (ms) | 18.2 | 27.2 (+49%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 19.9 | 29.0 (+45%) | | three-areas · market-after-turn · frame time (p99) (ms) | 31.7 | 16.4 (−48%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd (2/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-15.6%)`; `three-areas screenshot.max_diff`; `three-areas kokiri-standing.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`; 2: linux+amd+radv+dgpu `AMD Steam Machine (RADV NAVI33), 26.1.99` `AMD Custom CPU 1772, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 068fb8e | After: this PR @ 69263e4 | |---|---|---| | three-areas · field-standing · frame rate | 67.3 | 77.1 (+15%) | | three-areas · kokiri-standing · frame rate | 60.1 | 78.9 (+31%) | | three-areas · market-standing · frame rate | 42.2 | 36.4 (−14%) | | three-areas · field-after-turn · frame rate | 69.2 | 82.8 (+20%) | | three-areas · kokiri-after-turn · frame rate | 61.0 | 75.9 (+24%) | | three-areas · market-after-turn · frame rate | 69.9 | 84.3 (+21%) | | three-areas · field-standing · stutters | 2.2 | 4.0 (+78%) | | three-areas · kokiri-standing · stutters | 0.8 | 0.8 (+0%) | | three-areas · market-standing · stutters | 71.0 | 100.5 (+42%) | | three-areas · field-after-turn · stutters | 0.2 | 6.8 (+2600%) | | three-areas · kokiri-after-turn · stutters | 0.5 | 21.8 (+4250%) | | three-areas · market-after-turn · stutters | 6.5 | 0.8 (−88%) | | three-areas · field-standing · frame time (p99) (ms) | 20.3 | 27.1 (+33%) | | three-areas · kokiri-standing · frame time (p99) (ms) | 20.5 | 16.3 (−20%) | | three-areas · market-standing · frame time (p99) (ms) | 52.9 | 74.5 (+41%) | | three-areas · field-after-turn · frame time (p99) (ms) | 18.2 | 27.2 (+49%) | | three-areas · kokiri-after-turn · frame time (p99) (ms) | 19.9 | 29.0 (+45%) | | three-areas · market-after-turn · frame time (p99) (ms) | 31.7 | 16.4 (−48%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+nvidia **Verdict:** running on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** running on linux+nvidia (1/1 report). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> </details>
Author
Owner

@Beanstalk (new head 69263e4: two commits on 21b606e against the Market loss on AMD discrete cards: an indexed draw's index buffer bound once per buffer with its place given as the first index (the fine profile on the RX 7900 XTX showed a bind for 0.97 of its 2,500 draws a frame), and a submission's first part ended sooner so the GPU starts sooner.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main (the index change must draw exactly as main does).

Cells: the PR's defaults on every platform. On linux+amd and windows+amd only, three more PR-only cells: RX_VK_INDEX_REBASE=0; RX_VK_PART_FIRST=0; RX_VK_PROF_FINE=1 (keep its log).

@Beanstalk (new head 69263e4: two commits on 21b606e against the Market loss on AMD discrete cards: an indexed draw's index buffer bound once per buffer with its place given as the first index (the fine profile on the RX 7900 XTX showed a bind for 0.97 of its 2,500 draws a frame), and a submission's first part ended sooner so the GPU starts sooner.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main (the index change must draw exactly as main does). Cells: the PR's defaults on every platform. On linux+amd and windows+amd only, three more PR-only cells: `RX_VK_INDEX_REBASE=0`; `RX_VK_PART_FIRST=0`; `RX_VK_PROF_FINE=1` (keep its log).
HumanBean force-pushed perf/thread-redesign from 69263e4d66
Some checks failed
CI / Workflow lint (actionlint) (pull_request) Successful in 12s
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 10s
Website / Website on Cloudflare (pull_request_target) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
Third-party parts / As their notes say (pull_request) Successful in 14s
CI / Conventional Commit title (pull_request) Successful in 7s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
beanstalk/windows+nvidia 1/1 fail
CI / Website build and checks (pull_request) Has been cancelled
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+amd 1/1 fail
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 2/1 fail
beanstalk/linux+nvidia 0/1 pass, 1 running, 0 queued
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/amd+igpu 0/1 pass, 1 running, 1 queued
to 8ba0d0281b
Some checks failed
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Workflow lint (actionlint) (pull_request) Successful in 12s
Third-party parts / As their notes say (pull_request) Successful in 14s
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 8s
CI / Conventional Commit title (pull_request) Successful in 9s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 12s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+amd 1/1 fail
beanstalk/windows+nvidia 1/1 fail
beanstalk/windows+intel 0/1 pass, 0 running, 2 queued
beanstalk/linux+amd 1/1 fail
beanstalk/linux+nvidia 1/1 pass
beanstalk/linux+intel 0/1 pass, 0 running, 2 queued
beanstalk/steamdeck 0/1 pass, 2 running, 0 queued
beanstalk/amd+igpu 1/1 fail
CI / Website build and checks (pull_request) Has been cancelled
2026-11-20 04:06:03 +00:00
Compare
Member
Beanstalk test farm · an earlier run: commit 8ba0d02 (the newer results are in a comment below)

Commit 8ba0d02 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

On eight requested hardware targets, the plan compares main and the PR over two starts per side, visiting Kokiri Forest, Hyrule Field and Market with V-Sync off and the frame-rate cap uncapped. It measures standing and post-turn performance, stutters and crashes, and compares pictures; the schema cannot measure while a camera-stick step is active. Regressions in FPS, p99 frame time, stutters or pictures fail the plan. The AMD-only switch cells cannot be restricted to those targets by the plan-wide matrix and are omitted; Intel targets will wait because no Intel tester is currently listed. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up); 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate.
  • The PR fails if: it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 5%; or stutters is more than 50% above main; or frame time (p99) is more than 15% above main.
  • Time: about 28 min of game runs per tester (4 game starts), plus the two builds.

Started at 04:07 UTC.

Estimate: done in about 1 h, from the queue and the testers online now; some jobs can't start yet.

âš  No tester has a PC for linux+intel (none seen in 14 days): it waits until one signs in.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd · 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 1/1 1 1 PR run 12.8 vs 12.8 (−0%) — fail
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3080 Ti, 621.24.0 2/1 1 2 none 11.2 vs 11.5 (−2%) — fail
windows+intel 0/1 — — — queued
linux+amd · 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 1/1 1 none 10.7 vs 14.7 (−27%) — fail
linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce GTX 1070, 580.196.7; 2: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070, 615.71.9 2/1 1 2 none 12.3 vs 12.0 (+2%) — pass
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — running
amd+igpu · 1: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99; 2: linux+amd+radv+igpu AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99 2/1 1 2 none 34.3 vs 42.3 (−19%) — fail ⚠ outlier

Measurements · windows+amd

Verdict: fail on windows+amd (1/1 report): three-areas crash; three-areas forest-after-turn.fps_mean.min_delta_pct (-22.3%); three-areas market-standing.fps_mean.min_delta_pct (-12.8%) failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu AMD Radeon RX 9060 XT, 2.0.416 AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 08994a9 After: this PR @ 8ba0d02
three-areas · field-standing · frame rate 92.3 92.2 (−0%)
three-areas · forest-standing · frame rate 84.3 91.9 (+9%)
three-areas · market-standing · frame rate 65.1 56.7 (−13%)
three-areas · field-after-turn · frame rate 85.9 84.5 (−2%)
three-areas · forest-after-turn · frame rate 84.2 65.4 (−22%)
three-areas · market-after-turn · frame rate 79.2 74.6 (−6%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 4.5 0.0 (−100%)
three-areas · market-standing · stutters 0.0 0.0
three-areas · field-after-turn · stutters 0.0 0.0
three-areas · forest-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.0 0.0
three-areas · field-standing · frame time (p99) (ms) 12.8 12.8 (−0%)
three-areas · forest-standing · frame time (p99) (ms) 16.1 13.6 (−16%)
three-areas · market-standing · frame time (p99) (ms) 19.2 24.0 (+25%)
three-areas · field-after-turn · frame time (p99) (ms) 13.8 14.7 (+6%)
three-areas · forest-after-turn · frame time (p99) (ms) 13.6 14.6 (+8%)
three-areas · market-after-turn · frame time (p99) (ms) 17.0 19.1 (+12%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+amd (1/1 report): `three-areas crash`; `three-areas forest-after-turn.fps_mean.min_delta_pct (-22.3%)`; `three-areas market-standing.fps_mean.min_delta_pct (-12.8%)` failed.

Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 08994a9 | After: this PR @ 8ba0d02 |
|---|---|---|
| three-areas · field-standing · frame rate | 92.3 | 92.2 (−0%) |
| three-areas · forest-standing · frame rate | 84.3 | 91.9 (+9%) |
| three-areas · market-standing · frame rate | 65.1 | 56.7 (−13%) |
| three-areas · field-after-turn · frame rate | 85.9 | 84.5 (−2%) |
| three-areas · forest-after-turn · frame rate | 84.2 | 65.4 (−22%) |
| three-areas · market-after-turn · frame rate | 79.2 | 74.6 (−6%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 4.5 | 0.0 (−100%) |
| three-areas · market-standing · stutters | 0.0 | 0.0 |
| three-areas · field-after-turn · stutters | 0.0 | 0.0 |
| three-areas · forest-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.0 | 0.0 |
| three-areas · field-standing · frame time (p99) (ms) | 12.8 | 12.8 (−0%) |
| three-areas · forest-standing · frame time (p99) (ms) | 16.1 | 13.6 (−16%) |
| three-areas · market-standing · frame time (p99) (ms) | 19.2 | 24.0 (+25%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 13.8 | 14.7 (+6%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 13.6 | 14.6 (+8%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 17.0 | 19.1 (+12%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · windows+nvidia

Verdict: fail on windows+nvidia (2/1 report): three-areas market-after-turn.fps_mean.min_delta_pct (-5.3%); three-areas market-after-turn.stutter_count.max_delta_pct failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3080 Ti, 621.24.0 AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 08994a9 After: this PR @ 8ba0d02
three-areas · field-standing · frame rate 102.7 107.3 (+4%)
three-areas · forest-standing · frame rate 90.9 98.5 (+8%)
three-areas · market-standing · frame rate 75.0 77.6 (+3%)
three-areas · field-after-turn · frame rate 108.2 109.0 (+1%)
three-areas · forest-after-turn · frame rate 91.8 96.8 (+5%)
three-areas · market-after-turn · frame rate 87.2 82.6 (−5%)
three-areas · field-standing · stutters 0.5 0.0 (−100%)
three-areas · forest-standing · stutters 0.0 0.0
three-areas · market-standing · stutters 0.5 0.0 (−100%)
three-areas · field-after-turn · stutters 0.0 0.0
three-areas · forest-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.0 0.5
three-areas · field-standing · frame time (p99) (ms) 11.5 11.2 (−2%)
three-areas · forest-standing · frame time (p99) (ms) 14.6 11.9 (−18%)
three-areas · market-standing · frame time (p99) (ms) 16.5 16.6 (+1%)
three-areas · field-after-turn · frame time (p99) (ms) 11.4 11.5 (+0%)
three-areas · forest-after-turn · frame time (p99) (ms) 12.9 12.6 (−2%)
three-areas · market-after-turn · frame time (p99) (ms) 14.6 15.2 (+4%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on windows+nvidia (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-5.3%)`; `three-areas market-after-turn.stutter_count.max_delta_pct` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 08994a9 | After: this PR @ 8ba0d02 |
|---|---|---|
| three-areas · field-standing · frame rate | 102.7 | 107.3 (+4%) |
| three-areas · forest-standing · frame rate | 90.9 | 98.5 (+8%) |
| three-areas · market-standing · frame rate | 75.0 | 77.6 (+3%) |
| three-areas · field-after-turn · frame rate | 108.2 | 109.0 (+1%) |
| three-areas · forest-after-turn · frame rate | 91.8 | 96.8 (+5%) |
| three-areas · market-after-turn · frame rate | 87.2 | 82.6 (−5%) |
| three-areas · field-standing · stutters | 0.5 | 0.0 (−100%) |
| three-areas · forest-standing · stutters | 0.0 | 0.0 |
| three-areas · market-standing · stutters | 0.5 | 0.0 (−100%) |
| three-areas · field-after-turn · stutters | 0.0 | 0.0 |
| three-areas · forest-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.0 | 0.5 |
| three-areas · field-standing · frame time (p99) (ms) | 11.5 | 11.2 (−2%) |
| three-areas · forest-standing · frame time (p99) (ms) | 14.6 | 11.9 (−18%) |
| three-areas · market-standing · frame time (p99) (ms) | 16.5 | 16.6 (+1%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 11.4 | 11.5 (+0%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 12.9 | 12.6 (−2%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 14.6 | 15.2 (+4%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+amd

Verdict: fail on linux+amd (1/1 report): three-areas market-standing.fps_mean.min_delta_pct (-22.9%); three-areas screenshot.max_diff; three-areas field-after-turn.stutter_count.max_delta_pct failed.

Hardware: 1: linux+amd+radv+dgpu AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 08994a9 After: this PR @ 8ba0d02
three-areas · field-standing · frame rate 84.9 104.0 (+22%)
three-areas · forest-standing · frame rate 77.6 110.8 (+43%)
three-areas · market-standing · frame rate 64.4 49.6 (−23%)
three-areas · field-after-turn · frame rate 81.9 112.3 (+37%)
three-areas · forest-after-turn · frame rate 77.5 107.5 (+39%)
three-areas · market-after-turn · frame rate 89.5 103.5 (+16%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 0.5 0.5 (+0%)
three-areas · market-standing · stutters 3.5 47.0 (+1243%)
three-areas · field-after-turn · stutters 0.0 0.5
three-areas · forest-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.0 1.0
three-areas · field-standing · frame time (p99) (ms) 14.7 10.7 (−27%)
three-areas · forest-standing · frame time (p99) (ms) 15.5 11.1 (−28%)
three-areas · market-standing · frame time (p99) (ms) 19.9 65.3 (+228%)
three-areas · field-after-turn · frame time (p99) (ms) 14.7 9.9 (−32%)
three-areas · forest-after-turn · frame time (p99) (ms) 15.7 11.0 (−30%)
three-areas · market-after-turn · frame time (p99) (ms) 15.4 12.5 (−19%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-22.9%)`; `three-areas screenshot.max_diff`; `three-areas field-after-turn.stutter_count.max_delta_pct` failed.

Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 08994a9 | After: this PR @ 8ba0d02 |
|---|---|---|
| three-areas · field-standing · frame rate | 84.9 | 104.0 (+22%) |
| three-areas · forest-standing · frame rate | 77.6 | 110.8 (+43%) |
| three-areas · market-standing · frame rate | 64.4 | 49.6 (−23%) |
| three-areas · field-after-turn · frame rate | 81.9 | 112.3 (+37%) |
| three-areas · forest-after-turn · frame rate | 77.5 | 107.5 (+39%) |
| three-areas · market-after-turn · frame rate | 89.5 | 103.5 (+16%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 0.5 | 0.5 (+0%) |
| three-areas · market-standing · stutters | 3.5 | 47.0 (+1243%) |
| three-areas · field-after-turn · stutters | 0.0 | 0.5 |
| three-areas · forest-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.0 | 1.0 |
| three-areas · field-standing · frame time (p99) (ms) | 14.7 | 10.7 (−27%) |
| three-areas · forest-standing · frame time (p99) (ms) | 15.5 | 11.1 (−28%) |
| three-areas · market-standing · frame time (p99) (ms) | 19.9 | 65.3 (+228%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 14.7 | 9.9 (−32%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 15.7 | 11.0 (−30%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 15.4 | 12.5 (−19%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · linux+nvidia

Verdict: pass on linux+nvidia (2/1 report): no rule failed and no PR run crashed.

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce GTX 1070, 580.196.7 Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAM; 2: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070, 615.71.9 AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 08994a9 After: this PR @ 8ba0d02
three-areas · field-standing · frame rate 114.1 109.9 (−4%)
three-areas · forest-standing · frame rate 108.5 110.5 (+2%)
three-areas · market-standing · frame rate 81.2 79.9 (−2%)
three-areas · field-after-turn · frame rate 128.6 131.9 (+3%)
three-areas · forest-after-turn · frame rate 114.0 116.9 (+3%)
three-areas · market-after-turn · frame rate 107.7 104.0 (−3%)
three-areas · field-standing · stutters 2.5 0.5 (−80%)
three-areas · forest-standing · stutters 5.0 1.5 (−70%)
three-areas · market-standing · stutters 5.0 3.0 (−40%)
three-areas · field-after-turn · stutters 2.0 0.0 (−100%)
three-areas · forest-after-turn · stutters 1.5 0.5 (−67%)
three-areas · market-after-turn · stutters 4.0 2.5 (−38%)
three-areas · field-standing · frame time (p99) (ms) 12.0 12.3 (+2%)
three-areas · forest-standing · frame time (p99) (ms) 13.5 12.4 (−8%)
three-areas · market-standing · frame time (p99) (ms) 19.4 18.6 (−4%)
three-areas · field-after-turn · frame time (p99) (ms) 10.2 9.7 (−5%)
three-areas · forest-after-turn · frame time (p99) (ms) 11.5 10.9 (−5%)
three-areas · market-after-turn · frame time (p99) (ms) 13.8 14.3 (+4%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** pass on linux+nvidia (2/1 report): no rule failed and no PR run crashed.

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce GTX 1070, 580.196.7` `Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAM`; 2: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070, 615.71.9` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 08994a9 | After: this PR @ 8ba0d02 |
|---|---|---|
| three-areas · field-standing · frame rate | 114.1 | 109.9 (−4%) |
| three-areas · forest-standing · frame rate | 108.5 | 110.5 (+2%) |
| three-areas · market-standing · frame rate | 81.2 | 79.9 (−2%) |
| three-areas · field-after-turn · frame rate | 128.6 | 131.9 (+3%) |
| three-areas · forest-after-turn · frame rate | 114.0 | 116.9 (+3%) |
| three-areas · market-after-turn · frame rate | 107.7 | 104.0 (−3%) |
| three-areas · field-standing · stutters | 2.5 | 0.5 (−80%) |
| three-areas · forest-standing · stutters | 5.0 | 1.5 (−70%) |
| three-areas · market-standing · stutters | 5.0 | 3.0 (−40%) |
| three-areas · field-after-turn · stutters | 2.0 | 0.0 (−100%) |
| three-areas · forest-after-turn · stutters | 1.5 | 0.5 (−67%) |
| three-areas · market-after-turn · stutters | 4.0 | 2.5 (−38%) |
| three-areas · field-standing · frame time (p99) (ms) | 12.0 | 12.3 (+2%) |
| three-areas · forest-standing · frame time (p99) (ms) | 13.5 | 12.4 (−8%) |
| three-areas · market-standing · frame time (p99) (ms) | 19.4 | 18.6 (−4%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 10.2 | 9.7 (−5%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 11.5 | 10.9 (−5%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 13.8 | 14.3 (+4%) |

### Flags and settings

None: the plan varied no setting, switch or build option.

Measurements · amd+igpu

Verdict: fail on amd+igpu (2/1 report): three-areas screenshot.max_diff; three-areas market-after-turn.stutter_count.max_delta_pct (+550.0%); three-areas market-after-turn.frametime_p99_ms.max_delta_pct (+68.4%) failed.

Hardware: 1: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 AMD Custom APU 0405, 12 GB RAM; 2: linux+amd+radv+igpu AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99 AMD Ryzen 7 7840HS w/ Radeon 780M Graphics, 24 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Before: main @ 08994a9 After: this PR @ 8ba0d02
three-areas · field-standing · frame rate 30.8 35.5 (+15%)
three-areas · forest-standing · frame rate 32.4 37.7 (+16%)
three-areas · market-standing · frame rate 28.6 28.0 (−2%)
three-areas · field-after-turn · frame rate 32.2 39.6 (+23%)
three-areas · forest-after-turn · frame rate 31.0 35.3 (+14%)
three-areas · market-after-turn · frame rate 30.8 39.1 (+27%)
three-areas · field-standing · stutters 0.0 0.0
three-areas · forest-standing · stutters 0.0 0.0
three-areas · market-standing · stutters 1.5 0.5 (−67%)
three-areas · field-after-turn · stutters 0.0 0.0
three-areas · forest-after-turn · stutters 0.0 0.0
three-areas · market-after-turn · stutters 0.8 3.2 (+333%)
three-areas · field-standing · frame time (p99) (ms) 42.3 34.3 (−19%)
three-areas · forest-standing · frame time (p99) (ms) 42.8 37.1 (−13%)
three-areas · market-standing · frame time (p99) (ms) 64.8 59.9 (−8%)
three-areas · field-after-turn · frame time (p99) (ms) 40.4 31.6 (−22%)
three-areas · forest-after-turn · frame time (p99) (ms) 41.4 36.6 (−12%)
three-areas · market-after-turn · frame time (p99) (ms) 49.8 70.7 (+42%)

Flags and settings

None: the plan varied no setting, switch or build option.

Markdown for the PR description
## Measurements

**Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff`; `three-areas market-after-turn.stutter_count.max_delta_pct (+550.0%)`; `three-areas market-after-turn.frametime_p99_ms.max_delta_pct (+68.4%)` failed.

Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99` `AMD Ryzen 7 7840HS w/ Radeon 780M Graphics, 24 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

| | Before: main @ 08994a9 | After: this PR @ 8ba0d02 |
|---|---|---|
| three-areas · field-standing · frame rate | 30.8 | 35.5 (+15%) |
| three-areas · forest-standing · frame rate | 32.4 | 37.7 (+16%) |
| three-areas · market-standing · frame rate | 28.6 | 28.0 (−2%) |
| three-areas · field-after-turn · frame rate | 32.2 | 39.6 (+23%) |
| three-areas · forest-after-turn · frame rate | 31.0 | 35.3 (+14%) |
| three-areas · market-after-turn · frame rate | 30.8 | 39.1 (+27%) |
| three-areas · field-standing · stutters | 0.0 | 0.0 |
| three-areas · forest-standing · stutters | 0.0 | 0.0 |
| three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) |
| three-areas · field-after-turn · stutters | 0.0 | 0.0 |
| three-areas · forest-after-turn · stutters | 0.0 | 0.0 |
| three-areas · market-after-turn · stutters | 0.8 | 3.2 (+333%) |
| three-areas · field-standing · frame time (p99) (ms) | 42.3 | 34.3 (−19%) |
| three-areas · forest-standing · frame time (p99) (ms) | 42.8 | 37.1 (−13%) |
| three-areas · market-standing · frame time (p99) (ms) | 64.8 | 59.9 (−8%) |
| three-areas · field-after-turn · frame time (p99) (ms) | 40.4 | 31.6 (−22%) |
| three-areas · forest-after-turn · frame time (p99) (ms) | 41.4 | 36.6 (−12%) |
| three-areas · market-after-turn · frame time (p99) (ms) | 49.8 | 70.7 (+42%) |

### Flags and settings

None: the plan varied no setting, switch or build option.
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>8ba0d02</code> (the newer results are in a comment below)</summary> Commit `8ba0d02` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) On eight requested hardware targets, the plan compares main and the PR over two starts per side, visiting Kokiri Forest, Hyrule Field and Market with V-Sync off and the frame-rate cap uncapped. It measures standing and post-turn performance, stutters and crashes, and compares pictures; the schema cannot measure while a camera-stick step is active. Regressions in FPS, p99 frame time, stutters or pictures fail the plan. The AMD-only switch cells cannot be restricted to those targets by the plan-wide matrix and are omitted; Intel targets will wait because no Intel tester is currently listed. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up); 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate. - **The PR fails if:** it crashes; or frame rate is more than 5% below main; or a picture differs from main by more than 5%; or stutters is more than 50% above main; or frame time (p99) is more than 15% above main. - **Time:** about 28 min of game runs per tester (4 game starts), plus the two builds. Started at 04:07 UTC. **Estimate:** done in about 1 h, from the queue and the testers online now; some jobs can't start yet. ⚠ No tester has a PC for `linux+intel` (none seen in 14 days): it waits until one signs in. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd · 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_1e2a6d44b508a366) | 1 PR run | 12.8 vs 12.8 (−0%) | — | fail | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_f544ba03d5b9a843) [2](https://beanstalk.magicbean.dev/reports/rp_21c02fb7ff77d288) | none | 11.2 vs 11.5 (−2%) | — | fail | | windows+intel | 0/1 | — | — | — | queued | | linux+amd · 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_b47b80f27c0c9ecd) | none | 10.7 vs 14.7 (−27%) | — | fail | | linux+nvidia · 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce GTX 1070, 580.196.7`; 2: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070, 615.71.9` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_a166fe985da3a48a) [2](https://beanstalk.magicbean.dev/reports/rp_2f62a6866fd6272c) | none | 12.3 vs 12.0 (+2%) | — | pass | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | running | | amd+igpu · 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99`; 2: linux+amd+radv+igpu `AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99` | 2/1 [1](https://beanstalk.magicbean.dev/reports/rp_7376d0cf82cdecf4) [2](https://beanstalk.magicbean.dev/reports/rp_32212f8a9379f543) | none | 34.3 vs 42.3 (−19%) | — | fail ⚠ outlier | #### Measurements · windows+amd **Verdict:** fail on windows+amd (1/1 report): `three-areas crash`; `three-areas forest-after-turn.fps_mean.min_delta_pct (-22.3%)`; `three-areas market-standing.fps_mean.min_delta_pct (-12.8%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 92.3 | 92.2 (−0%) | | three-areas · forest-standing · frame rate | 84.3 | 91.9 (+9%) | | three-areas · market-standing · frame rate | 65.1 | 56.7 (−13%) | | three-areas · field-after-turn · frame rate | 85.9 | 84.5 (−2%) | | three-areas · forest-after-turn · frame rate | 84.2 | 65.4 (−22%) | | three-areas · market-after-turn · frame rate | 79.2 | 74.6 (−6%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 4.5 | 0.0 (−100%) | | three-areas · market-standing · stutters | 0.0 | 0.0 | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 0.0 | | three-areas · field-standing · frame time (p99) (ms) | 12.8 | 12.8 (−0%) | | three-areas · forest-standing · frame time (p99) (ms) | 16.1 | 13.6 (−16%) | | three-areas · market-standing · frame time (p99) (ms) | 19.2 | 24.0 (+25%) | | three-areas · field-after-turn · frame time (p99) (ms) | 13.8 | 14.7 (+6%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 13.6 | 14.6 (+8%) | | three-areas · market-after-turn · frame time (p99) (ms) | 17.0 | 19.1 (+12%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+amd (1/1 report): `three-areas crash`; `three-areas forest-after-turn.fps_mean.min_delta_pct (-22.3%)`; `three-areas market-standing.fps_mean.min_delta_pct (-12.8%)` failed. Hardware: 1: windows+amd+amd-proprietary+dgpu `AMD Radeon RX 9060 XT, 2.0.416` `AMD Ryzen 5 5600 6-Core Processor, 32 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 92.3 | 92.2 (−0%) | | three-areas · forest-standing · frame rate | 84.3 | 91.9 (+9%) | | three-areas · market-standing · frame rate | 65.1 | 56.7 (−13%) | | three-areas · field-after-turn · frame rate | 85.9 | 84.5 (−2%) | | three-areas · forest-after-turn · frame rate | 84.2 | 65.4 (−22%) | | three-areas · market-after-turn · frame rate | 79.2 | 74.6 (−6%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 4.5 | 0.0 (−100%) | | three-areas · market-standing · stutters | 0.0 | 0.0 | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 0.0 | | three-areas · field-standing · frame time (p99) (ms) | 12.8 | 12.8 (−0%) | | three-areas · forest-standing · frame time (p99) (ms) | 16.1 | 13.6 (−16%) | | three-areas · market-standing · frame time (p99) (ms) | 19.2 | 24.0 (+25%) | | three-areas · field-after-turn · frame time (p99) (ms) | 13.8 | 14.7 (+6%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 13.6 | 14.6 (+8%) | | three-areas · market-after-turn · frame time (p99) (ms) | 17.0 | 19.1 (+12%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · windows+nvidia **Verdict:** fail on windows+nvidia (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-5.3%)`; `three-areas market-after-turn.stutter_count.max_delta_pct` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 102.7 | 107.3 (+4%) | | three-areas · forest-standing · frame rate | 90.9 | 98.5 (+8%) | | three-areas · market-standing · frame rate | 75.0 | 77.6 (+3%) | | three-areas · field-after-turn · frame rate | 108.2 | 109.0 (+1%) | | three-areas · forest-after-turn · frame rate | 91.8 | 96.8 (+5%) | | three-areas · market-after-turn · frame rate | 87.2 | 82.6 (−5%) | | three-areas · field-standing · stutters | 0.5 | 0.0 (−100%) | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 0.5 | 0.0 (−100%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 0.5 | | three-areas · field-standing · frame time (p99) (ms) | 11.5 | 11.2 (−2%) | | three-areas · forest-standing · frame time (p99) (ms) | 14.6 | 11.9 (−18%) | | three-areas · market-standing · frame time (p99) (ms) | 16.5 | 16.6 (+1%) | | three-areas · field-after-turn · frame time (p99) (ms) | 11.4 | 11.5 (+0%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 12.9 | 12.6 (−2%) | | three-areas · market-after-turn · frame time (p99) (ms) | 14.6 | 15.2 (+4%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on windows+nvidia (2/1 report): `three-areas market-after-turn.fps_mean.min_delta_pct (-5.3%)`; `three-areas market-after-turn.stutter_count.max_delta_pct` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 102.7 | 107.3 (+4%) | | three-areas · forest-standing · frame rate | 90.9 | 98.5 (+8%) | | three-areas · market-standing · frame rate | 75.0 | 77.6 (+3%) | | three-areas · field-after-turn · frame rate | 108.2 | 109.0 (+1%) | | three-areas · forest-after-turn · frame rate | 91.8 | 96.8 (+5%) | | three-areas · market-after-turn · frame rate | 87.2 | 82.6 (−5%) | | three-areas · field-standing · stutters | 0.5 | 0.0 (−100%) | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 0.5 | 0.0 (−100%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 0.5 | | three-areas · field-standing · frame time (p99) (ms) | 11.5 | 11.2 (−2%) | | three-areas · forest-standing · frame time (p99) (ms) | 14.6 | 11.9 (−18%) | | three-areas · market-standing · frame time (p99) (ms) | 16.5 | 16.6 (+1%) | | three-areas · field-after-turn · frame time (p99) (ms) | 11.4 | 11.5 (+0%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 12.9 | 12.6 (−2%) | | three-areas · market-after-turn · frame time (p99) (ms) | 14.6 | 15.2 (+4%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+amd **Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-22.9%)`; `three-areas screenshot.max_diff`; `three-areas field-after-turn.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 84.9 | 104.0 (+22%) | | three-areas · forest-standing · frame rate | 77.6 | 110.8 (+43%) | | three-areas · market-standing · frame rate | 64.4 | 49.6 (−23%) | | three-areas · field-after-turn · frame rate | 81.9 | 112.3 (+37%) | | three-areas · forest-after-turn · frame rate | 77.5 | 107.5 (+39%) | | three-areas · market-after-turn · frame rate | 89.5 | 103.5 (+16%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.5 | 0.5 (+0%) | | three-areas · market-standing · stutters | 3.5 | 47.0 (+1243%) | | three-areas · field-after-turn · stutters | 0.0 | 0.5 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 1.0 | | three-areas · field-standing · frame time (p99) (ms) | 14.7 | 10.7 (−27%) | | three-areas · forest-standing · frame time (p99) (ms) | 15.5 | 11.1 (−28%) | | three-areas · market-standing · frame time (p99) (ms) | 19.9 | 65.3 (+228%) | | three-areas · field-after-turn · frame time (p99) (ms) | 14.7 | 9.9 (−32%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 15.7 | 11.0 (−30%) | | three-areas · market-after-turn · frame time (p99) (ms) | 15.4 | 12.5 (−19%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on linux+amd (1/1 report): `three-areas market-standing.fps_mean.min_delta_pct (-22.9%)`; `three-areas screenshot.max_diff`; `three-areas field-after-turn.stutter_count.max_delta_pct` failed. Hardware: 1: linux+amd+radv+dgpu `AMD Radeon RX 7900 XTX (RADV NAVI31), 26.2.7` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 84.9 | 104.0 (+22%) | | three-areas · forest-standing · frame rate | 77.6 | 110.8 (+43%) | | three-areas · market-standing · frame rate | 64.4 | 49.6 (−23%) | | three-areas · field-after-turn · frame rate | 81.9 | 112.3 (+37%) | | three-areas · forest-after-turn · frame rate | 77.5 | 107.5 (+39%) | | three-areas · market-after-turn · frame rate | 89.5 | 103.5 (+16%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.5 | 0.5 (+0%) | | three-areas · market-standing · stutters | 3.5 | 47.0 (+1243%) | | three-areas · field-after-turn · stutters | 0.0 | 0.5 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.0 | 1.0 | | three-areas · field-standing · frame time (p99) (ms) | 14.7 | 10.7 (−27%) | | three-areas · forest-standing · frame time (p99) (ms) | 15.5 | 11.1 (−28%) | | three-areas · market-standing · frame time (p99) (ms) | 19.9 | 65.3 (+228%) | | three-areas · field-after-turn · frame time (p99) (ms) | 14.7 | 9.9 (−32%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 15.7 | 11.0 (−30%) | | three-areas · market-after-turn · frame time (p99) (ms) | 15.4 | 12.5 (−19%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · linux+nvidia **Verdict:** pass on linux+nvidia (2/1 report): no rule failed and no PR run crashed. Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce GTX 1070, 580.196.7` `Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAM`; 2: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070, 615.71.9` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 114.1 | 109.9 (−4%) | | three-areas · forest-standing · frame rate | 108.5 | 110.5 (+2%) | | three-areas · market-standing · frame rate | 81.2 | 79.9 (−2%) | | three-areas · field-after-turn · frame rate | 128.6 | 131.9 (+3%) | | three-areas · forest-after-turn · frame rate | 114.0 | 116.9 (+3%) | | three-areas · market-after-turn · frame rate | 107.7 | 104.0 (−3%) | | three-areas · field-standing · stutters | 2.5 | 0.5 (−80%) | | three-areas · forest-standing · stutters | 5.0 | 1.5 (−70%) | | three-areas · market-standing · stutters | 5.0 | 3.0 (−40%) | | three-areas · field-after-turn · stutters | 2.0 | 0.0 (−100%) | | three-areas · forest-after-turn · stutters | 1.5 | 0.5 (−67%) | | three-areas · market-after-turn · stutters | 4.0 | 2.5 (−38%) | | three-areas · field-standing · frame time (p99) (ms) | 12.0 | 12.3 (+2%) | | three-areas · forest-standing · frame time (p99) (ms) | 13.5 | 12.4 (−8%) | | three-areas · market-standing · frame time (p99) (ms) | 19.4 | 18.6 (−4%) | | three-areas · field-after-turn · frame time (p99) (ms) | 10.2 | 9.7 (−5%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 11.5 | 10.9 (−5%) | | three-areas · market-after-turn · frame time (p99) (ms) | 13.8 | 14.3 (+4%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** pass on linux+nvidia (2/1 report): no rule failed and no PR run crashed. Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce GTX 1070, 580.196.7` `Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAM`; 2: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070, 615.71.9` `AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 114.1 | 109.9 (−4%) | | three-areas · forest-standing · frame rate | 108.5 | 110.5 (+2%) | | three-areas · market-standing · frame rate | 81.2 | 79.9 (−2%) | | three-areas · field-after-turn · frame rate | 128.6 | 131.9 (+3%) | | three-areas · forest-after-turn · frame rate | 114.0 | 116.9 (+3%) | | three-areas · market-after-turn · frame rate | 107.7 | 104.0 (−3%) | | three-areas · field-standing · stutters | 2.5 | 0.5 (−80%) | | three-areas · forest-standing · stutters | 5.0 | 1.5 (−70%) | | three-areas · market-standing · stutters | 5.0 | 3.0 (−40%) | | three-areas · field-after-turn · stutters | 2.0 | 0.0 (−100%) | | three-areas · forest-after-turn · stutters | 1.5 | 0.5 (−67%) | | three-areas · market-after-turn · stutters | 4.0 | 2.5 (−38%) | | three-areas · field-standing · frame time (p99) (ms) | 12.0 | 12.3 (+2%) | | three-areas · forest-standing · frame time (p99) (ms) | 13.5 | 12.4 (−8%) | | three-areas · market-standing · frame time (p99) (ms) | 19.4 | 18.6 (−4%) | | three-areas · field-after-turn · frame time (p99) (ms) | 10.2 | 9.7 (−5%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 11.5 | 10.9 (−5%) | | three-areas · market-after-turn · frame time (p99) (ms) | 13.8 | 14.3 (+4%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> #### Measurements · amd+igpu **Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff`; `three-areas market-after-turn.stutter_count.max_delta_pct (+550.0%)`; `three-areas market-after-turn.frametime_p99_ms.max_delta_pct (+68.4%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99` `AMD Ryzen 7 7840HS w/ Radeon 780M Graphics, 24 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 30.8 | 35.5 (+15%) | | three-areas · forest-standing · frame rate | 32.4 | 37.7 (+16%) | | three-areas · market-standing · frame rate | 28.6 | 28.0 (−2%) | | three-areas · field-after-turn · frame rate | 32.2 | 39.6 (+23%) | | three-areas · forest-after-turn · frame rate | 31.0 | 35.3 (+14%) | | three-areas · market-after-turn · frame rate | 30.8 | 39.1 (+27%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.8 | 3.2 (+333%) | | three-areas · field-standing · frame time (p99) (ms) | 42.3 | 34.3 (−19%) | | three-areas · forest-standing · frame time (p99) (ms) | 42.8 | 37.1 (−13%) | | three-areas · market-standing · frame time (p99) (ms) | 64.8 | 59.9 (−8%) | | three-areas · field-after-turn · frame time (p99) (ms) | 40.4 | 31.6 (−22%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 41.4 | 36.6 (−12%) | | three-areas · market-after-turn · frame time (p99) (ms) | 49.8 | 70.7 (+42%) | ### Flags and settings None: the plan varied no setting, switch or build option. <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** fail on amd+igpu (2/1 report): `three-areas screenshot.max_diff`; `three-areas market-after-turn.stutter_count.max_delta_pct (+550.0%)`; `three-areas market-after-turn.frametime_p99_ms.max_delta_pct (+68.4%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`; 2: linux+amd+radv+igpu `AMD Radeon 780M Graphics (RADV PHOENIX), 26.1.99` `AMD Ryzen 7 7840HS w/ Radeon 780M Graphics, 24 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) | | Before: main @ 08994a9 | After: this PR @ 8ba0d02 | |---|---|---| | three-areas · field-standing · frame rate | 30.8 | 35.5 (+15%) | | three-areas · forest-standing · frame rate | 32.4 | 37.7 (+16%) | | three-areas · market-standing · frame rate | 28.6 | 28.0 (−2%) | | three-areas · field-after-turn · frame rate | 32.2 | 39.6 (+23%) | | three-areas · forest-after-turn · frame rate | 31.0 | 35.3 (+14%) | | three-areas · market-after-turn · frame rate | 30.8 | 39.1 (+27%) | | three-areas · field-standing · stutters | 0.0 | 0.0 | | three-areas · forest-standing · stutters | 0.0 | 0.0 | | three-areas · market-standing · stutters | 1.5 | 0.5 (−67%) | | three-areas · field-after-turn · stutters | 0.0 | 0.0 | | three-areas · forest-after-turn · stutters | 0.0 | 0.0 | | three-areas · market-after-turn · stutters | 0.8 | 3.2 (+333%) | | three-areas · field-standing · frame time (p99) (ms) | 42.3 | 34.3 (−19%) | | three-areas · forest-standing · frame time (p99) (ms) | 42.8 | 37.1 (−13%) | | three-areas · market-standing · frame time (p99) (ms) | 64.8 | 59.9 (−8%) | | three-areas · field-after-turn · frame time (p99) (ms) | 40.4 | 31.6 (−22%) | | three-areas · forest-after-turn · frame time (p99) (ms) | 41.4 | 36.6 (−12%) | | three-areas · market-after-turn · frame time (p99) (ms) | 49.8 | 70.7 (+42%) | ### Flags and settings None: the plan varied no setting, switch or build option. ``` </details> </details>
Author
Owner

@Beanstalk (new head 8ba0d02: rebased on main with #1815, so Linux starts again; plus comment and doc cleanup.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main.

Matrix, please plan all of these cells. Every platform: the PR's defaults. On linux+amd and windows+amd, three more cells, each on the PR side only against the same main: (a) env RX_VK_INDEX_REBASE=0; (b) env RX_VK_PART_FIRST=0; (c) env RX_VK_INDEX_REBASE=0 RX_VK_PART_FIRST=0. They tell which of the two new changes moves the Market.

@Beanstalk (new head 8ba0d02: rebased on main with #1815, so Linux starts again; plus comment and doc cleanup.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. Matrix, please plan all of these cells. Every platform: the PR's defaults. On linux+amd and windows+amd, three more cells, each on the PR side only against the same main: (a) env `RX_VK_INDEX_REBASE=0`; (b) env `RX_VK_PART_FIRST=0`; (c) env `RX_VK_INDEX_REBASE=0 RX_VK_PART_FIRST=0`. They tell which of the two new changes moves the Market.
Contributor

This is the Steam Deck check that #1725 asked for, run on that PR's commit (fec53338d) before it was folded into this one. It is not a test of this PR's head: it measures only the late-readback default on a handheld, on a build without the thread rewrite. I'm posting it here because 1725's closing comment points its Deck flash check at this PR's Still to test; I can run the same routes on this PR's head next if that is wanted.

Steam Deck run of pull 1725's commit: faster where the readback wait was the cost, no new flashes

On a Steam Deck OLED the default that commit turns on is worth about +14% standing in Hyrule Field and +7% in the Market, nothing in Kokiri Forest, and +5% fps while the camera turns, with lower 1% lows there. The flash detector found no one-frame white or dark picture with that default that the same build without the rule doesn't also show.

What was built. de22a6b9d: main's parent 621dae505 with pull 1725's commit fec53338d cherry-picked on it (clean). Not main's head at the time (0de25c3e4): it did not compile on Linux then (gpu/vk/menu/pipeline_list.cpp used SIGPIPE, sigtimedwait, waitpid and uname without their POSIX headers), since fixed by #1807. Main's head differed from its parent only by that menu change, which touches none of that commit's files. A dev-packs base, RX_MARCH=x86-64-v3, no RX_HOT_OPT, no Tracy. One binary for both sides: D = nothing set (that commit's default), Z = RX_VK_READBACK_LATE=0 (its own "turning it off").

Hardware. Steam Deck OLED, SteamOS 3.9.5, RADV VANGOGH (Mesa 26.2.0-devel), GPU clock manual 1600 MHz, headless (the harness, no one holding it).

Settings. Minimum preset, Render 50%, output 1280x800, frame rate uncapped, vsync off. Fresh options each run, the same warm shader cache for all runs.

Policy lines. Both sides print readback_late_newest=1 in vk: policy: can and measured: ...: the line reports the device's default, and with RX_VK_READBACK_LATE=0 it still prints =1. What tells the sides apart is the rule line, which D prints and Z doesn't: vk: readbacks read without their wait: the default on a handheld (newest 1, 0 sizes). Worth a word in the log, since =1 on a run with the switch set to 0 reads as the rule being on.

Method. Kokiri Forest standing 120 s, then a teleport to Hyrule Field: standing 60 s, 4 camera pans (about 46 s), standing again (about 84 s). The Market: standing, 120 s from 20 s after arrival. Two runs a side per route, alternated D, Z, Z, D. fps from the presents, p99 frame time, 1% low = 1000 / mean of the slowest 1% of frame times.

Window D fps (run 1, run 2) Z fps D vs Z D p99 ms Z p99 ms D 1% low Z 1% low
Kokiri Forest standing 18.78, 19.21 18.69, 19.81 -1.3% 75.2, 71.3 76.2, 65.2 12.65, 12.75 12.05, 14.47
Hyrule Field standing 38.44, 39.35 34.24, 34.25 +13.6% 35.3, 33.8 36.4, 35.8 26.79, 28.24 26.36, 26.22
Hyrule Field pans 36.27, 36.75 34.66, 34.82 +5.1% 53.8, 50.8 49.7, 47.3 14.04, 14.29 16.63, 17.95
Hyrule Field standing after the pans 36.73, 40.01 37.53, 33.42 +8.2% (the runs overlap) 37.8, 33.9 34.4, 38.6 20.52, 27.83 26.19, 21.73
the Market standing 16.87, 16.35 15.28, 15.78 +7.0% 161.1, 156.6 160.4, 161.1 5.73, 5.90 5.74, 5.62
  • Kokiri Forest: no change (the sides overlap).
  • Hyrule Field standing: +13.6%, both D runs above both Z runs, p99 a little lower.
  • The pans: +5% fps but worse frame times. The 1% low is lower (14.0 and 14.3 against 16.6 and 18.0) and p99 higher (51-54 ms against 47-50 ms), on both runs a side. So the late read makes the average frame cheaper and the slowest ones slower while the view changes.
  • The Market: +7%, p99 and 1% lows unchanged.

The readback lines (vk: readbacks of GPU results: <n> of 3000 waited for their write-back, <ms> ms each, median per window, run 1 / run 2):

Window D waited / ms each Z waited / ms each
Kokiri Forest 574, 516 / 0.03, 0.03 535, 504 / 0.03, 0.03
Hyrule Field standing 2051, 2038 / 0.22, 0.21 1250, 1240 / 0.85, 0.82
Hyrule Field pans 1473, 1428 / 0.19, 0.20 1096, 1069 / 0.80, 0.81
the Market 1092, 892 / 0.43, 0.29 896, 989 / 0.89, 0.74

Where the gain is, each wait is about a quarter as long with the rule (0.2 ms against 0.8 ms in the field). In Kokiri Forest the waits are already 0.03 ms, which fits nothing changing there. With the rule more of the 3000 reads count as waited, but each wait is short.

Flashes (RX_VK_FLASH=1, every present compared with its neighbours, RX_VK_FLASH_REGION=40; one run a side on the main route, two a side on the Lon Lon Ranch song route). I looked at every flagged triplet (the frame before, the flagged one and the one after) and measured the whole picture's mean brightness of each.

  • Main route (Kokiri Forest, Hyrule Field, the pans), D 37 flags, Z 26. All of them fall in the pans, at night by the forest entrance:
    • D: 30 are Navi's glow moving across a dark background close to the camera; 7 are the tunnel's edge, hanging roots and the misty forest beyond sweeping across the picture as the camera turns.
    • Z: 15 Navi, 11 the same tunnel and roots, or the camera passing close to Link.
    • In every triplet the flagged frame's mean brightness is within 4 levels (of 255) of its neighbours' mean. Neither side shows a white or dark picture or an exposure jump. The difference in count follows how often Navi crosses the camera's path in a run, not the switch.
  • Lon Lon Ranch song route, D 10 and 10 flags, Z 14 and 13. Both sides flag the same kinds:
    • the camera moving at the start of Malon's conversation (D 2 and 2, Z 3 and 1, Z's one being the HUD appearing with the ranch's name);
    • in the overhead shot of the ranch after the song, pale lavender-white patches of grass that jump to a new place every frame (D 7 and 5, Z 9 and 5). These are the "brighter" flags. They are the ranch's known pale-foliage flicker, which appears just as often with the rule off;
    • the cut from the close-up of Link holding the ocarina to the wide shot (D 1 and 1, Z 0 and 2);
    • horses or the foal passing close to the camera (D 0 and 2, Z 2 and 5).
    • No flagged frame falls in the stretch where the song plays, on either side. The detector found no darker frame there with D or with Z. Apart from the cuts and a horse filling the picture, each flagged frame's mean brightness is within 3 levels of its neighbours'.

Limits.

  • Two runs a side per route, one Deck. The standing-after-pans and Kokiri windows overlap between sides.
  • Headless: no one watched the screen. The flash check sees only what the detector flags at its thresholds (a 48-level cell change, or a 40-level jump of a 16x9 block); a fainter one-frame change would pass unflagged.
  • The flash runs save every frame and run at about half speed, so their timings are not in the table.
  • Built on main's parent rather than main's head, for the reason above.
This is the Steam Deck check that https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1725 asked for, run on **that PR's commit** (`fec53338d`) before it was folded into this one. It is not a test of this PR's head: it measures only the late-readback default on a handheld, on a build without the thread rewrite. I'm posting it here because 1725's closing comment points its Deck flash check at this PR's Still to test; I can run the same routes on this PR's head next if that is wanted. **Steam Deck run of pull 1725's commit: faster where the readback wait was the cost, no new flashes** On a Steam Deck OLED the default that commit turns on is worth about +14% standing in Hyrule Field and +7% in the Market, nothing in Kokiri Forest, and +5% fps while the camera turns, with lower 1% lows there. The flash detector found no one-frame white or dark picture with that default that the same build without the rule doesn't also show. **What was built.** `de22a6b9d`: main's parent `621dae505` with pull 1725's commit `fec53338d` cherry-picked on it (clean). Not main's head at the time (`0de25c3e4`): it did not compile on Linux then (`gpu/vk/menu/pipeline_list.cpp` used `SIGPIPE`, `sigtimedwait`, `waitpid` and `uname` without their POSIX headers), since fixed by https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1807. Main's head differed from its parent only by that menu change, which touches none of that commit's files. A dev-packs base, `RX_MARCH=x86-64-v3`, no `RX_HOT_OPT`, no Tracy. One binary for both sides: **D** = nothing set (that commit's default), **Z** = `RX_VK_READBACK_LATE=0` (its own "turning it off"). **Hardware.** Steam Deck OLED, SteamOS 3.9.5, RADV VANGOGH (Mesa 26.2.0-devel), GPU clock manual 1600 MHz, headless (the harness, no one holding it). **Settings.** Minimum preset, Render 50%, output 1280x800, frame rate uncapped, vsync off. Fresh options each run, the same warm shader cache for all runs. **Policy lines.** Both sides print `readback_late_newest=1` in `vk: policy: can and measured: ...`: the line reports the device's default, and with `RX_VK_READBACK_LATE=0` it still prints `=1`. What tells the sides apart is the rule line, which D prints and Z doesn't: `vk: readbacks read without their wait: the default on a handheld (newest 1, 0 sizes)`. Worth a word in the log, since `=1` on a run with the switch set to `0` reads as the rule being on. **Method.** Kokiri Forest standing 120 s, then a teleport to Hyrule Field: standing 60 s, 4 camera pans (about 46 s), standing again (about 84 s). The Market: standing, 120 s from 20 s after arrival. Two runs a side per route, alternated D, Z, Z, D. fps from the presents, p99 frame time, 1% low = 1000 / mean of the slowest 1% of frame times. | Window | D fps (run 1, run 2) | Z fps | D vs Z | D p99 ms | Z p99 ms | D 1% low | Z 1% low | |---|---|---|---|---|---|---|---| | Kokiri Forest standing | 18.78, 19.21 | 18.69, 19.81 | -1.3% | 75.2, 71.3 | 76.2, 65.2 | 12.65, 12.75 | 12.05, 14.47 | | Hyrule Field standing | 38.44, 39.35 | 34.24, 34.25 | **+13.6%** | 35.3, 33.8 | 36.4, 35.8 | 26.79, 28.24 | 26.36, 26.22 | | Hyrule Field pans | 36.27, 36.75 | 34.66, 34.82 | +5.1% | 53.8, 50.8 | 49.7, 47.3 | 14.04, 14.29 | 16.63, 17.95 | | Hyrule Field standing after the pans | 36.73, 40.01 | 37.53, 33.42 | +8.2% (the runs overlap) | 37.8, 33.9 | 34.4, 38.6 | 20.52, 27.83 | 26.19, 21.73 | | the Market standing | 16.87, 16.35 | 15.28, 15.78 | **+7.0%** | 161.1, 156.6 | 160.4, 161.1 | 5.73, 5.90 | 5.74, 5.62 | - Kokiri Forest: no change (the sides overlap). - Hyrule Field standing: +13.6%, both D runs above both Z runs, p99 a little lower. - The pans: +5% fps but worse frame times. The 1% low is lower (14.0 and 14.3 against 16.6 and 18.0) and p99 higher (51-54 ms against 47-50 ms), on both runs a side. So the late read makes the average frame cheaper and the slowest ones slower while the view changes. - The Market: +7%, p99 and 1% lows unchanged. **The readback lines** (`vk: readbacks of GPU results: <n> of 3000 waited for their write-back, <ms> ms each`, median per window, run 1 / run 2): | Window | D waited / ms each | Z waited / ms each | |---|---|---| | Kokiri Forest | 574, 516 / 0.03, 0.03 | 535, 504 / 0.03, 0.03 | | Hyrule Field standing | 2051, 2038 / 0.22, 0.21 | 1250, 1240 / 0.85, 0.82 | | Hyrule Field pans | 1473, 1428 / 0.19, 0.20 | 1096, 1069 / 0.80, 0.81 | | the Market | 1092, 892 / 0.43, 0.29 | 896, 989 / 0.89, 0.74 | Where the gain is, each wait is about a quarter as long with the rule (0.2 ms against 0.8 ms in the field). In Kokiri Forest the waits are already 0.03 ms, which fits nothing changing there. With the rule more of the 3000 reads count as waited, but each wait is short. **Flashes** (`RX_VK_FLASH=1`, every present compared with its neighbours, `RX_VK_FLASH_REGION=40`; one run a side on the main route, two a side on the Lon Lon Ranch song route). I looked at every flagged triplet (the frame before, the flagged one and the one after) and measured the whole picture's mean brightness of each. - **Main route** (Kokiri Forest, Hyrule Field, the pans), D 37 flags, Z 26. All of them fall in the pans, at night by the forest entrance: - D: 30 are Navi's glow moving across a dark background close to the camera; 7 are the tunnel's edge, hanging roots and the misty forest beyond sweeping across the picture as the camera turns. - Z: 15 Navi, 11 the same tunnel and roots, or the camera passing close to Link. - In every triplet the flagged frame's mean brightness is within 4 levels (of 255) of its neighbours' mean. Neither side shows a white or dark picture or an exposure jump. The difference in count follows how often Navi crosses the camera's path in a run, not the switch. - **Lon Lon Ranch song route**, D 10 and 10 flags, Z 14 and 13. Both sides flag the same kinds: - the camera moving at the start of Malon's conversation (D 2 and 2, Z 3 and 1, Z's one being the HUD appearing with the ranch's name); - in the overhead shot of the ranch after the song, pale lavender-white patches of grass that jump to a new place every frame (D 7 and 5, Z 9 and 5). These are the "brighter" flags. They are the ranch's known pale-foliage flicker, which appears just as often with the rule off; - the cut from the close-up of Link holding the ocarina to the wide shot (D 1 and 1, Z 0 and 2); - horses or the foal passing close to the camera (D 0 and 2, Z 2 and 5). - No flagged frame falls in the stretch where the song plays, on either side. The detector found no darker frame there with D or with Z. Apart from the cuts and a horse filling the picture, each flagged frame's mean brightness is within 3 levels of its neighbours'. **Limits.** - Two runs a side per route, one Deck. The standing-after-pans and Kokiri windows overlap between sides. - Headless: no one watched the screen. The flash check sees only what the detector flags at its thresholds (a 48-level cell change, or a 40-level jump of a 16x9 block); a fainter one-frame change would pass unflagged. - The flash runs save every frame and run at about half speed, so their timings are not in the table. - Built on main's parent rather than main's head, for the reason above.
perf(vk): the recorder thread back, from 6 cores: the driver's calls off the render thread (the Market on AMD)
Some checks reported errors
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
Third-party parts / As their notes say (pull_request) Successful in 12s
Website / Website on Cloudflare (pull_request_target) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
PR description / Description follows the template (pull_request_target) Failing after 4s
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+nvidia 0/1 pass, 1 error
beanstalk/windows+intel 0/1 pass, 0 running, 0 queued
beanstalk/linux+amd 0/1 pass, 0 running, 0 queued
beanstalk/steamdeck 0/1 pass, 0 running, 0 queued
beanstalk/windows+amd 0/1 pass, 0 running, 0 queued
beanstalk/amd+igpu 0/1 pass, 0 running, 0 queued
beanstalk/linux+nvidia 0/1 pass, 0 running, 0 queued
beanstalk/linux+intel 0/1 pass, 0 running, 0 queued
CI / Website build and checks (pull_request) Failing after 2m3s
4543d04b61
Without the recorder thread the render thread made every one of the driver's vkCmd* calls, and on AMD's drivers (about
twice NVIDIA's CPU time a call) the Market's densest view, ~2,500 draws a frame and ~11 ms of replay on an RX 7900 XTX,
fell 16-25% behind main on the RX 7900 XTX, RX 9070 XT and RX 9060 XT (the Test Farm on #1763). main's recorder
(vk_recorder.h) is back as it was, without the parallel replay's parts (par_active, the RX_VK_PARALLEL refusal): the
render thread packs each vkCmd* into a ring and the recorder thread makes the calls in order, the render thread waiting
for it at each command buffer end, submit, wait, readback and destruction and when it lets replay_mu go (DrainingMutex
again). Threaded rendering's Auto is two tiers again: the render thread from 4 cores, the recorder thread too from 6;
On is both. RX_VK_RECORDER (sync, defer, thread; any other value off) is over the row, for an A/B in one build.

With #1763's pieces: vk_work.h's macros keep their work counts, the profiler's glue spans and the call counts, then call
the recorder's packer; vk_gprof.cpp's own query reset goes through the packer (in order, without timing itself); the
direct recordings (Dear ImGui, NGX, FSR 4, XeSS) drain first again (B_DIRECT); the presenter thread makes no vkCmd*
call, and its present is a call off the replay lock, as on main.

Its known costs stand (threaded_rendering.h, threads.md): woken 20-25 times a frame, its short sleeps let an i7-9700K's
clock control slow every core (Kokiri Forest 15 fps against 48), and kept spinning it cost a Ryzen 9 7900X 3-11%. The
checks for it come back from main, fitted (no parallel cases; the spy of test_inline_write_order on the packers; the
recorder checks' stubs for vk_work.h's thread counts, glue and call counts), and test_threaded_rendering checks the two
tiers and that the segment workers stay gone.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP
Member

Beanstalk test farm

Commit 4543d04 · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

The plan compares main with the PR’s default recorder behavior and with the recorder disabled, because the new head restores that thread on PCs with at least six physical cores. Two starts per side visit Kokiri Forest, Hyrule Field and Market with V-Sync off and the frame rate uncapped, measuring FPS, p99 frame time and stutters and comparing pictures. Camera turns are exercised, but performance is measured immediately after each turn rather than during it. Crashes, significant performance regressions or large picture differences fail the run; each tester’s game runs take roughly 45 minutes. The requested Intel targets have no testers shown online and will wait for matching PCs. No tester online now matches windows+intel, linux+intel: those targets wait for one.

  • Where: windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports).
  • Matrix: 1 cell, each on main and the PR: RX_VK_RECORDER off (PR only).
  • three-areas: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up); 2 runs on each side.
  • Measured: crashes, frame times, stutters and hitches, draws and frame rate, the game's log.
  • The PR fails if: it crashes; or frame rate is more than 10% below main; or a picture differs from main by more than 10%; or stutters is more than 25% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0.
  • Time: about 28 min of game runs per tester (4 game starts), plus the two builds.

Done (started 05:33 UTC, last report 05:33 UTC). Comment @Beanstalk to run it again.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+amd 0/1 — — — queued
windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 1/1 1 none — — error
windows+intel 0/1 — — — queued
linux+amd 0/1 — — — queued
linux+nvidia 0/1 — — — queued
linux+intel 0/1 — — — queued
steamdeck 0/1 — — — queued
amd+igpu 0/1 — — — queued

Measurements · windows+nvidia

Verdict: could not be judged on windows+nvidia (1/1 report).

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0 AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Nothing was measured: no metric came back.

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (frame rate) Helps / doesn't help / untested
RX_VK_RECORDER switch the PR adds (set on the PR side only) off not in main the PR's default three-areas, main without it → the PR with off: not measured untested
Markdown for the PR description
## Measurements

**Verdict:** could not be judged on windows+nvidia (1/1 report).

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM`  
Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Nothing was measured: no metric came back.

### Flags and settings

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_RECORDER` | switch the PR adds (set on the PR side only) | `off` | not in main | the PR's default | three-areas, main without it → the PR with off: not measured | untested |
<!-- beanstalk:sticky --> ### Beanstalk test farm Commit `4543d04` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1763) **What will be tested** · AI plan (`gpt-6.1-terra`) The plan compares main with the PR’s default recorder behavior and with the recorder disabled, because the new head restores that thread on PCs with at least six physical cores. Two starts per side visit Kokiri Forest, Hyrule Field and Market with V-Sync off and the frame rate uncapped, measuring FPS, p99 frame time and stutters and comparing pictures. Camera turns are exercised, but performance is measured immediately after each turn rather than during it. Crashes, significant performance regressions or large picture differences fail the run; each tester’s game runs take roughly 45 minutes. The requested Intel targets have no testers shown online and will wait for matching PCs. No tester online now matches windows+intel, linux+intel: those targets wait for one. - **Where:** windows+amd (1-2 reports), windows+nvidia (1-2 reports), windows+intel (1-2 reports), linux+amd (1-2 reports), linux+nvidia (1-2 reports), linux+intel (1-2 reports), steamdeck (1-2 reports), amd+igpu (1-2 reports). - **Matrix:** 1 cell, each on main and the PR: RX_VK_RECORDER `off` (PR only). - **three-areas:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest to compare with main, move the stick, measure forest-after-turn for 15 s (after 0 s warm-up), teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up); 2 runs on each side. - **Measured:** crashes, frame times, stutters and hitches, draws and frame rate, the game's log. - **The PR fails if:** it crashes; or frame rate is more than 10% below main; or a picture differs from main by more than 10%; or stutters is more than 25% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0. - **Time:** about 28 min of game runs per tester (4 game starts), plus the two builds. Done (started 05:33 UTC, last report 05:33 UTC). Comment `@Beanstalk` to run it again. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+amd | 0/1 | — | — | — | queued | | windows+nvidia · 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` | 1/1 [1](https://beanstalk.magicbean.dev/reports/rp_6918b8605a0e4fb3) | none | — | — | error | | windows+intel | 0/1 | — | — | — | queued | | linux+amd | 0/1 | — | — | — | queued | | linux+nvidia | 0/1 | — | — | — | queued | | linux+intel | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | queued | | amd+igpu | 0/1 | — | — | — | queued | #### Measurements · windows+nvidia **Verdict:** could not be judged on windows+nvidia (1/1 report). Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_RECORDER` | switch the PR adds (set on the PR side only) | `off` | not in main | the PR's default | three-areas, main without it → the PR with off: not measured | untested | <details><summary>Markdown for the PR description</summary> ```markdown ## Measurements **Verdict:** could not be judged on windows+nvidia (1/1 report). Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3050 Ti Laptop GPU, 621.24.0` `AMD Ryzen 7 5800H with Radeon Graphics, 16 GB RAM` Scene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, forest-standing 20 s after 5 s warm-up, forest-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Nothing was measured: no metric came back. ### Flags and settings | Flag | What it is | Values tried | Before this PR | After this PR | Effect (frame rate) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_RECORDER` | switch the PR adds (set on the PR side only) | `off` | not in main | the PR's default | three-areas, main without it → the PR with off: not measured | untested | ``` </details>
Author
Owner

@Beanstalk (new head 4543d04: the recorder thread is back, from 6 physical cores, without the segment workers. It moves the driver's calls off the render thread, which should recover the Market standing loss on fast GPUs, AMD most.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the kokiri-start save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main.

Matrix, on every platform, two cells: the PR's defaults; and the PR with env RX_VK_RECORDER=off (the same build without the recorder thread), each against main.

@Beanstalk (new head 4543d04: the recorder thread is back, from 6 physical cores, without the segment workers. It moves the driver's calls off the render thread, which should recover the Market standing loss on fast GPUs, AMD most.) Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the `kokiri-start` save: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters and crashes, and compare pictures with main. Matrix, on every platform, two cells: the PR's defaults; and the PR with env `RX_VK_RECORDER=off` (the same build without the recorder thread), each against main.
HumanBean changed title from WIP: perf(vk): the renderer's threads rewritten: one render thread, the presenter on every PC, parts that end at the game's own pass ends to perf(vk): the renderer's threads rewritten: one render thread, the presenter on every PC, parts that end at the game's own pass ends 2026-11-20 05:47:11 +00:00
HumanBean deleted branch perf/thread-redesign 2026-11-20 05:47:20 +00:00
Contributor

Post-merge Steam Deck report (zeldaootr). This is the first Steam Deck report for this change. It compares current main after the merge with stock v3.0.0, and checks the change's turn-back switches against main's defaults.

Summary. On the Deck, main matches v3.0.0 in Kokiri Forest and the Market. In the forest exit toward Hyrule Field it is clearly faster: +25% standing, +14% panning, +33% standing after the pans. This compares two builds 64 commits apart, so the Field gain is not this PR's alone. Against the PR's own turn-back switches on the same build, main's defaults win in the Field (+11% standing). The switches win in Kokiri (one pair only, see below), and the Market is a tie. The Deck runs without the render thread, by its own policy. These headless runs make no swapchain presents, so the presenter thread never started. There were no errors, device losses, validation lines or crashes. Pictures match v3.0.0. The flash detector's flags on main are Navi and the camera passing through Link during the pans, not exposure flashes.

Builds

  • main: d12ddf69b (#1833, one commit after this PR's merge commit a3565ac5b), version v3.0.0-64-gd12ddf69b. Built with dev.py base and dev.py rt --empty, RX_MARCH=x86-64-v3, the release compile levels.
  • v3.0.0: the stock release build.
  • Each side used its own prepared shader cache. Main's shader translator stamp is b9d6fe244df8cef0. Neither side prepared or compiled pipelines during the runs: no vk: preparation: lines, and both logged 47798 program keys kept.

Hardware and settings

  • Steam Deck OLED, SteamOS 3.9.5, RADV VANGOGH, Mesa 26.2.0-devel. GPU clock set by hand to 1600 MHz. Headless at 1280x800.
  • The Deck developers' rows: Minimum (quality_preset=1), handheld_profile=0, Render 50% (resolution=4), frame pacing on, uncapped (frame_rate_cap=6), V-Sync off, frame generation off.
  • Repository save kokiri-start. No switches set on either side (stock defaults), except in the O cell below.
  • Main route: Kokiri Forest standing 120 s, teleport to Hyrule Field, standing 60 s at the forest exit, 4 camera pan cycles (42 to 51 s), standing until 210 s after arrival.
  • Market: Castle Town Market, the window from 20 s to 140 s after arrival, standing.
  • Runs interleaved A B B A, two a side: M V V M on each route.

main against v3.0.0 (median of two runs; each run in brackets)

Window v3.0.0 fps main fps Δ p99 ms (v3.0.0 → main) 1% low fps Frames ≥ 50 ms
Kokiri, 120 s 19.55 (19.95, 19.14) 19.39 (19.72, 19.06) −0.8% 69.7 → 68.6 13.71 → 13.59 1317 → 1405
Field, standing 60 s 34.10 (34.06, 34.14) 42.57 (42.59, 42.54) +24.8% 36.1 → 27.6 26.91 → 34.76 0.5 → 0
Field, pans 34.32 (33.24, 35.40) 39.25 (40.05, 38.45) +14.4% 55.3 → 44.5 15.11 → 17.74 29.5 → 9
Field, standing after pans 33.16 (34.13, 32.19) 44.02 (46.50, 41.54) +32.7% 38.2 → 29.3 24.35 → 30.74 2 → 1
Market, 120 s 16.01 (16.02, 16.00) 16.27 (14.36, 18.17) +1.6% 159.2 → 155.8 5.89 → 5.81 1729 → 1430
  • The p99, 1% low and ≥ 50 ms columns are medians too. The 1% low is 1000 divided by the mean of the slowest 1% of frame times.
  • Kokiri at this spot runs about 20 fps on the Deck, so about half of its frames sit right at 50 ms. The ≥ 50 ms count there is mostly that threshold.
  • main's two Market runs spread widely (14.4 and 18.2). A third main Market run, from the pair below, gave 16.0. Read the Market as no difference.

This PR's turn-back switches (O) against main's defaults (M), one pair a route

The rewrite has no single off switch: the recorder and parallel-replay switches are gone, and the presenter thread has none. So the comparison cell is the set this PR's Risks names for turning it back: RX_VK_PART_MIN=240 RX_VK_PART_MAX=240 RX_VK_PASS_BARRIER=full RX_VK_READBACK_LATE=0. RX_VK_PART_FIRST stays at its default. This measures the parts, the barriers and the late readbacks, not the thread design. It is one run a side, run O then M on each route.

Window O fps M fps M against O
Kokiri, 120 s 21.44 (p99 65.4, ≥ 50 ms 535) 19.58 (p99 67.9, ≥ 50 ms 1341) −8.7%
Field, standing 60 s 38.32 (p99 32.6) 42.61 (p99 27.8) +11.2%
Field, pans 35.67 (p99 51.3) 37.29 (p99 49.5) +4.5%
Field, standing after pans 42.39 43.70 +3.1%
Market, 120 s 16.09 16.03 −0.4%

The O cell's log confirms the switches took effect:

vk: switches set (9): ... RX_VK_PART_MIN=240 (tune), RX_VK_PART_MAX=240 (tune), RX_VK_PASS_BARRIER=full (attempt), RX_VK_READBACK_LATE=0 (kill)
vk: a submission's parts ended: 4164 where the game had ended its render pass, 15836 at 240 draws        (O)
vk: a submission's parts ended: 39790 where the game had ended its render pass, 210 at 512 draws         (M)

The Kokiri gap for O comes from one run. All five other Kokiri runs (M and V) fell between 19.06 and 19.95 fps, so it needs repeating before anyone reads it as a cost of the new parts or barriers there.

What main chose on the Deck

Lines from main's log, each the same in every main run:

vk: Threaded rendering: Auto: off (3 cores of 6 logical processors; the render thread from 4 cores)
vk: the presenter thread waits for its presents to be shown: yes (VK_KHR_present_wait)
vk: readbacks read without their wait: the culling camera block (0x3980) and the newest submission's (a handheld) by default (newest 1, 1 sizes)
vk: policy: can and measured: ... gpu_clock_peak=1 readback_late_newest=1 pcache_small_pc=0
vk: policy: auto settings and threads: auto_tier=1 ...; workers 2, capture helpers 2, texture threads 4 (gather 4), program workers in play 2 threaded_auto_render=0 threaded_auto_recorder=0 drop_modules_small_ram=1
  • Render thread: none. Auto leaves it off on the Deck's 3 usable cores, and no the replay runs on a thread of its own line appears. The replay runs on the game's presenting thread.
  • Presenter thread: did not run. The headless runs make no swapchain presents, so the presenter's first-use line (presenter thread: each present made on a thread of its own ...) and its per-600-presents lines never appear. Only the device-setup line about present-wait does. Whether the presenter thread costs or gains anything on the Deck needs a run with a real swapchain (Game Mode or a window).
  • Late readbacks: on, for handhelds, as this PR says.
  • v3.0.0 for comparison: Threaded rendering: Auto: off (3 cores ...) as well. Its can and measured policy line has presenter_queue=0 and no readback_late_newest, and it logs no rule line about reading without waiting.
  • Errors and crashes: across all 14 runs (main, O and v3.0.0) there was no VK_ERROR, device-lost, validation or [error] line. No crash records, and every exit code was 0. Lines matching "fallback" were the same on both builds: the Deck's compute fallback for mesh shaders (#640), and the reserved core mask's fallback.

Pictures

One screenshot per area, at the same spot after settling, main against v3.0.0. The difference measure is the share of pixels whose largest channel differs by more than 8, and the mean absolute difference.

Area main vs v3.0.0 v3.0.0 vs v3.0.0 (run-to-run noise) By eye
Kokiri 7.1%, mean 3.3 (second pair) 6.9%, mean 3.2 Same picture; Navi in another place. The first main run's shot differs from every other shot (40%, mean 10): Navi had flown into Link's house, lighting its inside, and the exposure followed
Field (forest exit) 13.9%, mean 5.2 22.5%, mean 6.3 Same picture: fog, fireflies and Navi in other places
Market (arrival) 17.3%, mean 5.5 32.7%, mean 9.7 Same picture apart from Navi's place

No difference in lighting, geometry, textures or post-processing was seen.

Flash detector

One main run of the main route with RX_VK_FLASH=1, RX_VK_FLASH_REGION=40, RX_VK_SHOT_EVERY=1. This is the Deck item in Still to test, with the newest submission's readbacks read late.

  • Flags: 19 frame flashes (15 brighter, 4 darker) and 19 region flashes (16 up, 3 down). All 19 fall inside the 4 camera pan cycles. None fell in the Kokiri stand, the Field stand or the stand after the pans.
  • What each flagged triplet shows:
    • Most of the ones looked at are the camera pressed against or passing through Link at the end of a pan, with Navi a few pixels from the lens. Her glow and wings move across most of the frame between presents. This includes the runs of presents 6203 to 6204, 6622 to 6625, 7039 to 7041 and 7460 to 7464, where she leaves the top of the frame.
    • Others are fireflies and the fog moving with the pan (6423, 7383). Every triplet's luma was measured; five triplets (6203, 6423, 6623, 7383, 7463) were looked at by eye.
  • Whole-frame luma: the middle frame minus the mean of its neighbours is −4.7 to +4.7 out of 255, and each extreme is Navi's glow entering or leaving the frame. No triplet shows a frame-wide brightening or darkening: nothing from the exposure meter.
  • A first flash run (13 flags, same kind and the same pan window) lost its triplets to a missing shots folder. It was repeated, and the numbers above are the repeat.

Limits

  • Two runs a side for main against v3.0.0, and one pair for the turn-back switches. On the Market and in Kokiri, the run-to-run spread is as large as the differences.
  • Headless at 1280x800, so no presenter thread and no frame pacing on a real screen. One Deck.
  • main against v3.0.0 covers everything merged since v3.0.0, not only this PR. The O pair is the closest to this PR's own effect on the Deck.
  • The windows are standing and scripted pans, not walking.
  • The game-free checks on Linux (tests/run_all.py on d12ddf69b, run separately) have 5 failures that start with this PR: checks that read vk_replay.cpp source this PR reshaped. They are test_split_before_draw.py, test_early_split_before_draw.py, test_hold_gpu.py, test_readback_early.py and test_readback_late.py. The other 46 failures already failed on this PR's parent.
**Post-merge Steam Deck report** (zeldaootr). This is the first Steam Deck report for this change. It compares current `main` after the merge with stock v3.0.0, and checks the change's turn-back switches against `main`'s defaults. **Summary.** On the Deck, `main` matches v3.0.0 in Kokiri Forest and the Market. In the forest exit toward Hyrule Field it is clearly faster: +25% standing, +14% panning, +33% standing after the pans. This compares two builds 64 commits apart, so the Field gain is not this PR's alone. Against the PR's own turn-back switches on the same build, `main`'s defaults win in the Field (+11% standing). The switches win in Kokiri (one pair only, see below), and the Market is a tie. The Deck runs without the render thread, by its own policy. These headless runs make no swapchain presents, so the presenter thread never started. There were no errors, device losses, validation lines or crashes. Pictures match v3.0.0. The flash detector's flags on `main` are Navi and the camera passing through Link during the pans, not exposure flashes. ### Builds - **main:** `d12ddf69b` (https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1833, one commit after this PR's merge commit `a3565ac5b`), version `v3.0.0-64-gd12ddf69b`. Built with `dev.py base` and `dev.py rt --empty`, `RX_MARCH=x86-64-v3`, the release compile levels. - **v3.0.0:** the stock release build. - Each side used its own prepared shader cache. Main's shader translator stamp is `b9d6fe244df8cef0`. Neither side prepared or compiled pipelines during the runs: no `vk: preparation:` lines, and both logged 47798 program keys kept. ### Hardware and settings - Steam Deck OLED, SteamOS 3.9.5, RADV VANGOGH, Mesa 26.2.0-devel. GPU clock set by hand to 1600 MHz. Headless at 1280x800. - The Deck developers' rows: Minimum (`quality_preset=1`), `handheld_profile=0`, Render 50% (`resolution=4`), frame pacing on, uncapped (`frame_rate_cap=6`), V-Sync off, frame generation off. - Repository save `kokiri-start`. No switches set on either side (stock defaults), except in the O cell below. - **Main route:** Kokiri Forest standing 120 s, teleport to Hyrule Field, standing 60 s at the forest exit, 4 camera pan cycles (42 to 51 s), standing until 210 s after arrival. - **Market:** Castle Town Market, the window from 20 s to 140 s after arrival, standing. - Runs interleaved A B B A, two a side: M V V M on each route. ### main against v3.0.0 (median of two runs; each run in brackets) | Window | v3.0.0 fps | main fps | Δ | p99 ms (v3.0.0 → main) | 1% low fps | Frames ≥ 50 ms | |---|---|---|---|---|---|---| | Kokiri, 120 s | 19.55 (19.95, 19.14) | 19.39 (19.72, 19.06) | −0.8% | 69.7 → 68.6 | 13.71 → 13.59 | 1317 → 1405 | | Field, standing 60 s | 34.10 (34.06, 34.14) | 42.57 (42.59, 42.54) | **+24.8%** | 36.1 → 27.6 | 26.91 → 34.76 | 0.5 → 0 | | Field, pans | 34.32 (33.24, 35.40) | 39.25 (40.05, 38.45) | **+14.4%** | 55.3 → 44.5 | 15.11 → 17.74 | 29.5 → 9 | | Field, standing after pans | 33.16 (34.13, 32.19) | 44.02 (46.50, 41.54) | **+32.7%** | 38.2 → 29.3 | 24.35 → 30.74 | 2 → 1 | | Market, 120 s | 16.01 (16.02, 16.00) | 16.27 (14.36, 18.17) | +1.6% | 159.2 → 155.8 | 5.89 → 5.81 | 1729 → 1430 | - The p99, 1% low and ≥ 50 ms columns are medians too. The 1% low is 1000 divided by the mean of the slowest 1% of frame times. - Kokiri at this spot runs about 20 fps on the Deck, so about half of its frames sit right at 50 ms. The ≥ 50 ms count there is mostly that threshold. - `main`'s two Market runs spread widely (14.4 and 18.2). A third `main` Market run, from the pair below, gave 16.0. Read the Market as no difference. ### This PR's turn-back switches (O) against main's defaults (M), one pair a route The rewrite has no single off switch: the recorder and parallel-replay switches are gone, and the presenter thread has none. So the comparison cell is the set this PR's Risks names for turning it back: `RX_VK_PART_MIN=240 RX_VK_PART_MAX=240 RX_VK_PASS_BARRIER=full RX_VK_READBACK_LATE=0`. `RX_VK_PART_FIRST` stays at its default. This measures the parts, the barriers and the late readbacks, not the thread design. It is one run a side, run O then M on each route. | Window | O fps | M fps | M against O | |---|---|---|---| | Kokiri, 120 s | 21.44 (p99 65.4, ≥ 50 ms 535) | 19.58 (p99 67.9, ≥ 50 ms 1341) | −8.7% | | Field, standing 60 s | 38.32 (p99 32.6) | 42.61 (p99 27.8) | **+11.2%** | | Field, pans | 35.67 (p99 51.3) | 37.29 (p99 49.5) | +4.5% | | Field, standing after pans | 42.39 | 43.70 | +3.1% | | Market, 120 s | 16.09 | 16.03 | −0.4% | The O cell's log confirms the switches took effect: ``` vk: switches set (9): ... RX_VK_PART_MIN=240 (tune), RX_VK_PART_MAX=240 (tune), RX_VK_PASS_BARRIER=full (attempt), RX_VK_READBACK_LATE=0 (kill) vk: a submission's parts ended: 4164 where the game had ended its render pass, 15836 at 240 draws (O) vk: a submission's parts ended: 39790 where the game had ended its render pass, 210 at 512 draws (M) ``` The Kokiri gap for O comes from one run. All five other Kokiri runs (M and V) fell between 19.06 and 19.95 fps, so it needs repeating before anyone reads it as a cost of the new parts or barriers there. ### What main chose on the Deck Lines from main's log, each the same in every main run: ``` vk: Threaded rendering: Auto: off (3 cores of 6 logical processors; the render thread from 4 cores) vk: the presenter thread waits for its presents to be shown: yes (VK_KHR_present_wait) vk: readbacks read without their wait: the culling camera block (0x3980) and the newest submission's (a handheld) by default (newest 1, 1 sizes) vk: policy: can and measured: ... gpu_clock_peak=1 readback_late_newest=1 pcache_small_pc=0 vk: policy: auto settings and threads: auto_tier=1 ...; workers 2, capture helpers 2, texture threads 4 (gather 4), program workers in play 2 threaded_auto_render=0 threaded_auto_recorder=0 drop_modules_small_ram=1 ``` - **Render thread:** none. Auto leaves it off on the Deck's 3 usable cores, and no `the replay runs on a thread of its own` line appears. The replay runs on the game's presenting thread. - **Presenter thread:** did not run. The headless runs make no swapchain presents, so the presenter's first-use line (`presenter thread: each present made on a thread of its own ...`) and its per-600-presents lines never appear. Only the device-setup line about present-wait does. Whether the presenter thread costs or gains anything on the Deck needs a run with a real swapchain (Game Mode or a window). - **Late readbacks:** on, for handhelds, as this PR says. - **v3.0.0 for comparison:** `Threaded rendering: Auto: off (3 cores ...)` as well. Its `can and measured` policy line has `presenter_queue=0` and no `readback_late_newest`, and it logs no rule line about reading without waiting. - **Errors and crashes:** across all 14 runs (main, O and v3.0.0) there was no `VK_ERROR`, device-lost, validation or `[error]` line. No crash records, and every exit code was 0. Lines matching "fallback" were the same on both builds: the Deck's compute fallback for mesh shaders (https://git.magicbean.dev/MagicBeanDev/Longshot/issues/640), and the reserved core mask's fallback. ### Pictures One screenshot per area, at the same spot after settling, `main` against v3.0.0. The difference measure is the share of pixels whose largest channel differs by more than 8, and the mean absolute difference. | Area | main vs v3.0.0 | v3.0.0 vs v3.0.0 (run-to-run noise) | By eye | |---|---|---|---| | Kokiri | 7.1%, mean 3.3 (second pair) | 6.9%, mean 3.2 | Same picture; Navi in another place. The first `main` run's shot differs from every other shot (40%, mean 10): Navi had flown into Link's house, lighting its inside, and the exposure followed | | Field (forest exit) | 13.9%, mean 5.2 | 22.5%, mean 6.3 | Same picture: fog, fireflies and Navi in other places | | Market (arrival) | 17.3%, mean 5.5 | 32.7%, mean 9.7 | Same picture apart from Navi's place | No difference in lighting, geometry, textures or post-processing was seen. ### Flash detector One `main` run of the main route with `RX_VK_FLASH=1`, `RX_VK_FLASH_REGION=40`, `RX_VK_SHOT_EVERY=1`. This is the Deck item in Still to test, with the newest submission's readbacks read late. - **Flags:** 19 frame flashes (15 brighter, 4 darker) and 19 region flashes (16 up, 3 down). All 19 fall inside the 4 camera pan cycles. None fell in the Kokiri stand, the Field stand or the stand after the pans. - **What each flagged triplet shows:** - Most of the ones looked at are the camera pressed against or passing through Link at the end of a pan, with Navi a few pixels from the lens. Her glow and wings move across most of the frame between presents. This includes the runs of presents 6203 to 6204, 6622 to 6625, 7039 to 7041 and 7460 to 7464, where she leaves the top of the frame. - Others are fireflies and the fog moving with the pan (6423, 7383). Every triplet's luma was measured; five triplets (6203, 6423, 6623, 7383, 7463) were looked at by eye. - **Whole-frame luma:** the middle frame minus the mean of its neighbours is −4.7 to +4.7 out of 255, and each extreme is Navi's glow entering or leaving the frame. No triplet shows a frame-wide brightening or darkening: nothing from the exposure meter. - A first flash run (13 flags, same kind and the same pan window) lost its triplets to a missing shots folder. It was repeated, and the numbers above are the repeat. ### Limits - Two runs a side for main against v3.0.0, and one pair for the turn-back switches. On the Market and in Kokiri, the run-to-run spread is as large as the differences. - Headless at 1280x800, so no presenter thread and no frame pacing on a real screen. One Deck. - `main` against v3.0.0 covers everything merged since v3.0.0, not only this PR. The O pair is the closest to this PR's own effect on the Deck. - The windows are standing and scripted pans, not walking. - The game-free checks on Linux (`tests/run_all.py` on `d12ddf69b`, run separately) have 5 failures that start with this PR: checks that read `vk_replay.cpp` source this PR reshaped. They are `test_split_before_draw.py`, `test_early_split_before_draw.py`, `test_hold_gpu.py`, `test_readback_early.py` and `test_readback_late.py`. The other 46 failures already failed on this PR's parent.
Contributor

Correction to my report above: the link after "the Deck's compute fallback for mesh shaders" is wrong. "issue 640" is the wording of the game's own log line (vk: mesh shaders ...: task and mesh programs run as compute shaders (issue 640 fallback)), not Longshot's number 640; ignore the link.

Correction to my report above: the link after "the Deck's compute fallback for mesh shaders" is wrong. "issue 640" is the wording of the game's own log line (`vk: mesh shaders ...: task and mesh programs run as compute shaders (issue 640 fallback)`), not Longshot's number 640; ignore the link.
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
5 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot!1763
No description provided.