Which settings are you using? Is RT enabled? Can it be reproduced early in the game?
I run the game with settings at 1440p Ultra. The game does not have RT. These settings are the same in Windows 11 Pro and Linux CachyOS. Do you think you could look into this stability and FPS issue? Thank you very much.
Well, everything I'm seeing so far is that it's just heavily CPU bound for some reason, and the game just crashes on boot on Windows (tried everything I could to fix it, but nothing), so I cannot do comparisons myself <_<.
Do you need me to help you with anything to make comparisons or screenshots, logs...??
Was finally able to make game work on windows, and yes, even with GPU fully loaded at 100%, there's some bullshit going on with GPU perf. We'll investigate.
Fun. It seems like AMD is doing some very special app-opt on some of the heavy lighting shaders.
$ ./d3d12-replayer.exe --json /e/tlou/cs_light_froxels_and_temporal/capture.json
[INFO]: Total ticks: 219634544, total timestamps: 1159
[INFO]: Total time per dispatch: 1895.035 us
post@DESKTOP-RT1V3M4 MINGW64 ~/git/d3d12-replayer/build-mingw/output/bin
$ cp d3d12-replayer.exe tlou-i.exe
post@DESKTOP-RT1V3M4 MINGW64 ~/git/d3d12-replayer/build-mingw/output/bin
$ ./tlou-i.exe --json /e/tlou/cs_light_froxels_and_temporal/capture.json
[INFO]: Total ticks: 46513240, total timestamps: 652
[INFO]: Total time per dispatch: 713.393 us
RADV is hitting some extremely bad performance in this particular shader for whatever reason, so that likely explains a lot of why ultra is so slow.
Wow, thanks for the help and the report! I hope it gets fixed with Mesa.
I have two questions:
Update to my earlier comment: I ran a native-Linux control test (Counter-Strike 2, Source 2, no Proton) at the same settings/resolution and it also shows the identical periodic stall (~60-100→15 FPS, same ~7-9s-scale cycle), while Team Fortress 2 (Source 1, native) stays completely clean. So Proton/VKD3D-Proton is not the root cause here — TLOU and RoboCop (both DX12-via-VKD3D-Proton) are affected, but so is a fully native Vulkan title (CS2), while a simpler native renderer (TF2) is unaffected. Looks like the actual common factor is renderer complexity (multi-queue/async-compute usage) interacting with the RTX 5080 + nvidia-open/GSP firmware stack, not the Proton translation layer itself. Filed as NVIDIA/open-gpu-kernel-modules#1260 with full data — leaving this comment for visibility since it still affects this title under Proton.
Hello, first of all, sorry if this doesn't belong here, but I think since it's a dx12 game, it's a good idea to include it.
Well, it turns out I've been testing The Last of Us Part 1's performance on Linux and Windows for several months, and on Linux, it's easily ~20 fps slower. However, The Last of Us Part 2 on Linux runs like Windows or even surpasses it in fps and stability (it's a better port). Would it be possible to investigate why this is happening? It would be great to fix it, if possible.
Software information
Name of the game, settings used etc.
System information
Log files
Please attach Proton or Wine logs as a text file:
PROTON_LOG=1 %command%and attach the correspondingsteam-xxxxx.logfile in your home directory.wine game.exe > game.log 2>&1and attach the resultinggame.logfile.