Game: Severe FPS drops while moving the camera on Death Mountain Trail (RTX 4080, nightly 189) #901

Open
opened 2026-11-15 19:51:14 +00:00 by BeanIssues · 3 comments
Collaborator

Reported on Discord

Longshot version

Nightly 189 (nightly-189-aa95d35f, source 719431ac). A second report is from v3.0.0-rc.5 (source 013b1d3a).

Channel

Nightly (first report); release candidate (second report).

What happened

While turning the camera and moving, base FPS swings between about 12 and 60; the first reporter says it returns to around 60 but does not stabilize unless they stop moving. Footage shows base FPS readings of 55, 35, 19, and 17, with 1% lows around 11–46. The log records 328 FPS samples (minimum 25, median 60, maximum 80); the lowest logged sample was during a brief replay-time spike.

A second player reports the same slowdown at Death Mountain Trail on RC.5, which includes Longshot#994's render-target reuse performance change. The continued report on RC.5 shows that this improvement did not necessarily eliminate this symptom for this player's scene/settings.

Where in the game

Death Mountain Trail, near the cracked-rock pile in the rocky canyon. The second report identifies DeathMountainTrail_Adult (adult).

How often

Occurs while turning the camera/moving; stabilizes when movement stops.

Graphics card and driver version

First report: NVIDIA GeForce RTX 4080, driver 621.14. Second report: NVIDIA GeForce RTX 3090, driver 621.01.

CPU, RAM and operating system

First report: AMD Ryzen 9 5900X, 64 GB RAM, Windows 10. Second report: AMD Ryzen 9 3900X (24 threads), 32 GB RAM, Windows 11 (build 26200).

Graphics settings (second report)

1920x1080 output, Docked, NVIDIA DLSS Native (DLAA), frame generation Auto, NVIDIA Reflex On, fullscreen, frame pacing On; other settings are in the attached bug report.

Logs and profiling measurements

First report's longshot_log.txt reports 3840×2160 output resolution. Its final summary reports about 79 FPS and 4.2 ms GPU time per frame; samples include a short drop to 25 FPS, so this may be intermittent rather than persistent in the captured session.

The second report's report.txt says HOS_FPS_LOG=1, HOS_THREAD_TIMES=5, RX_VK_PASS_TIMING enabled, and RX_VK_SLOW_LOG=50. Its attached longshot_log.txt contains repeated GPU pass timing sections (every 600 presents), but no HOS_THREAD_TIMES lines. RX_VK_PASS_TIMING's value is omitted from report.txt. The report was made 2026-11-16 at 14:34.

Additional evidence

First report footage: 20261115_141540.mp4, showing the performance overlay dropping to base 17–19 FPS near the rocks. Second report includes 2026-11-16_13-52-04.mp4 and picture.png.

Attachments from the second report

report.txt, longshot_log.txt, longshot_log.1.txt, and picture.png were provided in the Discord follow-up. The second log is truncated in the bug report bundle; the first is 1.3 MB. Both logs were checked for thread timing lines; none were present.

Longshot#994 reuses per-frame Vulkan render targets to reduce render-thread time and slow frames. It is included since v3.0.0-rc.5. The RC.5 report here still describes the slowdown, so further profiling/investigation is warranted.

Reported on [Discord](https://discord.com/channels/1569849200047427645/1571606182806946034/1571606182806946034) ### Longshot version Nightly 189 (`nightly-189-aa95d35f`, source `719431ac`). A second report is from v3.0.0-rc.5 (source `013b1d3a`). ### Channel Nightly (first report); release candidate (second report). ### What happened While turning the camera and moving, base FPS swings between about 12 and 60; the first reporter says it returns to around 60 but does not stabilize unless they stop moving. Footage shows base FPS readings of 55, 35, 19, and 17, with 1% lows around 11–46. The log records 328 FPS samples (minimum 25, median 60, maximum 80); the lowest logged sample was during a brief replay-time spike. A second player reports the same slowdown at Death Mountain Trail on RC.5, which includes Longshot#994's render-target reuse performance change. The continued report on RC.5 shows that this improvement did not necessarily eliminate this symptom for this player's scene/settings. ### Where in the game Death Mountain Trail, near the cracked-rock pile in the rocky canyon. The second report identifies `DeathMountainTrail_Adult` (adult). ### How often Occurs while turning the camera/moving; stabilizes when movement stops. ### Graphics card and driver version First report: NVIDIA GeForce RTX 4080, driver 621.14. Second report: NVIDIA GeForce RTX 3090, driver 621.01. ### CPU, RAM and operating system First report: AMD Ryzen 9 5900X, 64 GB RAM, Windows 10. Second report: AMD Ryzen 9 3900X (24 threads), 32 GB RAM, Windows 11 (build 26200). ### Graphics settings (second report) 1920x1080 output, Docked, NVIDIA DLSS Native (DLAA), frame generation Auto, NVIDIA Reflex On, fullscreen, frame pacing On; other settings are in the attached bug report. ### Logs and profiling measurements First report's `longshot_log.txt` reports 3840×2160 output resolution. Its final summary reports about 79 FPS and 4.2 ms GPU time per frame; samples include a short drop to 25 FPS, so this may be intermittent rather than persistent in the captured session. The second report's `report.txt` says `HOS_FPS_LOG=1`, `HOS_THREAD_TIMES=5`, `RX_VK_PASS_TIMING` enabled, and `RX_VK_SLOW_LOG=50`. Its attached `longshot_log.txt` contains repeated GPU pass timing sections (every 600 presents), but no `HOS_THREAD_TIMES` lines. `RX_VK_PASS_TIMING`'s value is omitted from report.txt. The report was made 2026-11-16 at 14:34. ### Additional evidence First report footage: `20261115_141540.mp4`, showing the performance overlay dropping to base 17–19 FPS near the rocks. Second report includes `2026-11-16_13-52-04.mp4` and `picture.png`. ### Attachments from the second report `report.txt`, `longshot_log.txt`, `longshot_log.1.txt`, and `picture.png` were provided in the Discord follow-up. The second log is truncated in the bug report bundle; the first is 1.3 MB. Both logs were checked for thread timing lines; none were present. ### Related performance change Longshot#994 reuses per-frame Vulkan render targets to reduce render-thread time and slow frames. It is included since v3.0.0-rc.5. The RC.5 report here still describes the slowdown, so further profiling/investigation is warranted.
Author
Collaborator

Files from the Discord report:

Files from the Discord report: - [longshot_log.txt](https://git.magicbean.dev/attachments/0b25040b-a62e-4311-9294-a1ccca59c49b) - [20261115_141540.mp4](https://git.magicbean.dev/attachments/b75241fe-3ca7-4c6f-8c5a-772ec961a1e1)
Author
Collaborator

From Discord:

A second player reports the same Death Mountain Trail camera/movement FPS drops on v3.0.0-rc.5 (source 013b1d3a). System: NVIDIA GeForce RTX 3090, driver 621.01; Ryzen 9 3900X, 32 GB, Windows 11 build 26200. Settings include frame generation Auto, NVIDIA Reflex On, fullscreen, Docked, DLSS with Native (DLAA, 1920x1080). Attached longshot_log.txt, longshot_log.1.txt, and 2026-11-16_13-52-04.mp4. Neither log contains HOS_THREAD_TIMES or RX_VK_PASS_TIMING measurements.

From [Discord](https://discord.com/channels/1569849200047427645/1571606182806946034/1571979081413820418): A second player reports the same Death Mountain Trail camera/movement FPS drops on v3.0.0-rc.5 (source 013b1d3a). System: NVIDIA GeForce RTX 3090, driver 621.01; Ryzen 9 3900X, 32 GB, Windows 11 build 26200. Settings include frame generation Auto, NVIDIA Reflex On, fullscreen, Docked, DLSS with Native (DLAA, 1920x1080). Attached `longshot_log.txt`, `longshot_log.1.txt`, and `2026-11-16_13-52-04.mp4`. Neither log contains HOS_THREAD_TIMES or RX_VK_PASS_TIMING measurements.
Author
Collaborator

Updated with more from Discord: Added the RC.5 follow-up and its profiling log bundle; documented that the released render-target optimization is present in RC.5 but did not necessarily resolve this reported symptom.

Files:

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1571606182806946034/1571987723374559384): Added the RC.5 follow-up and its profiling log bundle; documented that the released render-target optimization is present in RC.5 but did not necessarily resolve this reported symptom. Files: - [report.txt](https://git.magicbean.dev/attachments/312b9588-30f8-42ab-8dc7-81aff8afcb0b) - [longshot_log.txt](https://git.magicbean.dev/attachments/50adbcce-c222-4426-8907-304bcb186804) - [longshot_log.1.txt](https://git.magicbean.dev/attachments/7a19cc1d-22e8-4e23-aaef-9c46080aba4d) - [picture.png](https://git.magicbean.dev/attachments/6a851092-ba4d-4d2b-b09a-115db8cd21cb)
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
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#901
No description provided.