protonscr

Death Stranding 2 GPU Reset

vkd3dclosed
HansKristian-Work/vkd3d-proton#3153 · opened 2026-07-06 by Settyness · updated 2026-07-10 · 3 comments · github
SSettyness 2026-07-06 github

Death Stranding 2 appears to trigger a GPU reset which kicks me back into my display manager. This happens reliably enough for me to not have reached the first checkpoint, occurring in roughly 30 seconds to 5 minutes of gameplay.

This does happen once in a blue moon with some other games with one occurrence being tied to a specific graphical setting preset. In the logs, I see certain shader emulation being intentionally disabled due to buggy behavior of my family of cards, so maybe this isn't too surprising. It seems like it's triggered by a particular shader codepath or maybe a few.

dmesg initially blamed picom, which seemed fair, so I disabled it. dmesg then started blaming everything else. The log says: radv/amdgpu: The CS has been cancelled because the context is lost. This context is innocent.

I found this thread searching around: https://github.com/HansKristian-Work/vkd3d-proton/issues/2829.
I tested my VRAM with memtest_vulkan and everything came back clean. Seems okay.

There are also pretty annoying thermal issues with this family of cards with junction temps hitting 110°C. This was happening in DS2 as it does with more demanding games. I have never previously touched power or throttle controls for this card, but to rule out thermal issues, and because this particular card is factory overclocked, I also tested while under-clocked and under-volted, with my temps at a much more reasonable 85°C with no apparent degradation in performance or stability (beyond what was already occurring).

An interesting report I found that seems directly related: https://www.protondb.com/app/3280350#nF5B40pwYS
This user has the same card as myself and is using language that reflects what I'm seeing in my logs.

Apologies for my limited understanding in this, but I tried to gather as much data as I could to rule out my hardware, but I'm not sure I found anything that could definitively point to translation or driver layer bug. If there's any more information you need, I'm happy to share.

Software Information

Death Stranding 2

  • Low and Very High Settings Presets
  • Native AA, 1080p60
  • No Upscale, No Framegen

System Information

  • GPU: RX 5700 XT (GV-R57XTGAMING OC-8GD)
  • Driver: Mesa 26.1.3
  • Wine version: Proton Experimental 11.0-20260701b
  • VKD3D-Proton version: 3.1.0 (4232071c346c0a7)

Log Files

steam-3280350.log
coredump_1783315502.txt
coredump_1783315614.txt

Rrunar-work 2026-07-08 github

I gave this a try with RX 5700 XT, Mesa 26.1.4, 7.1.2-arch3-1, vkd3d-proton 4232071c346c0a7. Tried both native AA with PICO and FSR, low preset and very high preset. No issues here so far after an hour total.

SSettyness 2026-07-09 github

I missed something. Not much after, the aforementioned ProtonDB user filed another report: https://www.protondb.com/app/3280350#8ZyeiLz5GL
Here, they say they've eliminated the issue by enabling RADV debug. I was skeptical this is what fixed it for them, but I went ahead and did it anyway and ran with RADV_DEBUG=hang. The reset is gone; I can play the game now albeit at a hefty performance cost. To narrow it down, I also tried RADV_DEBUG=syncshaders and RADV_DEBUG=nocache, but those resulted in resets quickly.

SSettyness 2026-07-10 github

RADV_DEBUG=nocompute fixes my issue. Output a log for VK validation layers, but it didn't have anything useful unfortunately. Beginning to think I need to take this to Mesa, but I hope this info could be helpful.

Proton versions

Launch options

Upstream links