Game: performance stays lower after changing areas until a restart or DLSS toggle #89

New issue

Open

opened 2026-11-13 02:27:28 +00:00 by mooki · 0 comments

mooki commented

2026-11-13 02:27:28 +00:00

Longshot version

Original report: Hookshot v2.1.0-rc.3-256. Later reports persist on builds containing the memory fixes, including e0e60920. Longshot main 6b2dafefe inherits those changes; the full symptom is not established as fixed.

What kind of problem

Low fps or stutter

What happened

Ekamnoor reported a drop from about 60 to 35–40 fps entering town from Hyrule Field, while restarting directly there gives about 55–60. They later found that switching DLSS off/on restores performance until another area change. Other reporters describe similar persistent drops, not necessarily with the same cause.

I investigated the area-change and toggle behavior on Linux/RTX 4090. Earlier diagnostic runs before and after the imported-memory trimming and eviction fixes showed less pressure, not its disappearance. These were shared-load observations, not a valid FPS comparison or proof the report was fixed.

There is also a distinct confound: DLSS Balanced 58% becomes 50% after switching to the game's upscaler and back. Longshot pull request 55 fixes that saved-scale loss only, with two fresh Longshot correctness pairs. It does not close this performance report.

The inherited import trimming and excess-only eviction do not settle it. In later matched 8-GB-card tests, klikso still found the Market slower after a tour. He then withdrew his free-block-space explanation; do not carry that forward as the cause.

A separate 10-GB-card case from DFN-master improves with excess-only eviction disabled. klikso explicitly owns that regression fix; no competing fix is claimed here. The remaining post-tour cost and this eviction regression need their separate settings and evidence.

For a useful retest, retain a matched direct-load versus post-area route, exact build, render scale before/after any toggle, how long DLSS was off, frame-generation state and logs. The original reporter's failed log upload does not provide that comparison.

Where in the game

Hyrule Field → Castle Town Market; other reports mention Field → Kakariko and returning to Kokiri Forest. Later diagnostics include Kakariko → Lake Hylia.

How often

Repeated on affected routes; not every transition for every reporter.

Graphics card and driver version

Original reporter: RTX 4080 Laptop 12 GB, “latest” driver without a number. My observations: RTX 4090 24 GB, NVIDIA 615.78.08. Later 8-GB/10-GB measurements belong to klikso and DFN-master, not me.

CPU, RAM and operating system

Original reporter: Core i9-14900HX, 32 GB RAM, Windows 11. My observations used a Core i9-13900K and Arch Linux.

Changes from the defaults

Original reporter: 1440p, 7.5-GB memory setting, DLSS M/Balanced, frame generation 2×, anisotropy 16×, ultrawide enabled on a 16:9 display. My windowless observations did not exercise active frame generation or a real 8-GB heap.

Other software hooked into the game

Not established for all reporters. No single cause or absence of external software is inferred across the whole thread.

Reported on GitHub as NotRexed/Hookshot#2780; taken on there by mooki.

### Longshot version Original report: Hookshot v2.1.0-rc.3-256. Later reports persist on builds containing the memory fixes, including `e0e60920`. Longshot main `6b2dafefe` inherits those changes; the full symptom is not established as fixed. ### What kind of problem Low fps or stutter ### What happened [Ekamnoor reported](https://github.com/NotRexed/Hookshot/issues/2780) a drop from about 60 to 35–40 fps entering town from Hyrule Field, while restarting directly there gives about 55–60. They later found that switching DLSS off/on restores performance until another area change. Other reporters describe similar persistent drops, not necessarily with the same cause. I [investigated the area-change and toggle behavior](https://github.com/NotRexed/Hookshot/issues/2780#issuecomment-6832527827) on Linux/RTX 4090. Earlier diagnostic runs before and after the imported-memory trimming and eviction fixes showed less pressure, not its disappearance. These were shared-load observations, not a valid FPS comparison or proof the report was fixed. There is also a distinct confound: DLSS Balanced 58% becomes 50% after switching to the game's upscaler and back. [Longshot pull request 55](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/55) fixes **that saved-scale loss only**, with two fresh Longshot correctness pairs. It does not close this performance report. The inherited [import trimming](https://git.magicbean.dev/MagicBeanDev/Longshot/commit/ab2d4cc104066b155498731a7b8a10eef2f72f5e) and [excess-only eviction](https://git.magicbean.dev/MagicBeanDev/Longshot/commit/ba3d3f73ec58c6450e52946143c9e53b6d7aa5ec) do not settle it. In [later matched 8-GB-card tests](https://github.com/NotRexed/Hookshot/issues/2780#issuecomment-6838919319), klikso still found the Market slower after a tour. He then [withdrew his free-block-space explanation](https://github.com/NotRexed/Hookshot/issues/2780#issuecomment-6839688579); do not carry that forward as the cause. A separate 10-GB-card case from DFN-master improves with excess-only eviction disabled. [klikso explicitly owns that regression fix](https://github.com/NotRexed/Hookshot/issues/2780#issuecomment-6841009653); no competing fix is claimed here. The remaining post-tour cost and this eviction regression need their separate settings and evidence. For a useful retest, retain a matched direct-load versus post-area route, exact build, render scale before/after any toggle, how long DLSS was off, frame-generation state and logs. The original reporter's failed log upload does not provide that comparison. ### Where in the game Hyrule Field → Castle Town Market; other reports mention Field → Kakariko and returning to Kokiri Forest. Later diagnostics include Kakariko → Lake Hylia. ### How often Repeated on affected routes; not every transition for every reporter. ### Graphics card and driver version Original reporter: RTX 4080 Laptop 12 GB, “latest” driver without a number. My observations: RTX 4090 24 GB, NVIDIA 615.78.08. Later 8-GB/10-GB measurements belong to klikso and DFN-master, not me. ### CPU, RAM and operating system Original reporter: Core i9-14900HX, 32 GB RAM, Windows 11. My observations used a Core i9-13900K and Arch Linux. ### Changes from the defaults Original reporter: 1440p, 7.5-GB memory setting, DLSS M/Balanced, frame generation 2×, anisotropy 16×, ultrawide enabled on a 16:9 display. My windowless observations did not exercise active frame generation or a real 8-GB heap. ### Other software hooked into the game Not established for all reporters. No single cause or absence of external software is inferred across the whole thread. Reported on GitHub as NotRexed/Hookshot#2780; taken on there by mooki.

mooki referenced this issue

2026-11-13 02:28:12 +00:00

fix(menu): keep render scale when switching upscalers #55