Everything can cause GPU hangs on 9000 series cards. Here's a reddit post about it that sums up the issue: https://www.reddit.com/r/linux_gaming/comments/1pyiw5l/anyone_else_experiencing_crashesshader_cache
It's an amdgpu issue, it has to do with power limiting not working correctly, it's been happening for months, I guess AMD just doesn't care. So disappointing. If my GPU dies because it randomly starts pulling 550w I'll send Lisa Su the bill 😂
How long does it usually take before it hangs? Do you have a save file for a location where it's likely to happen?
How long does it usually take before it hangs? Do you have a save file for a location where it's likely to happen?
Varies quite a bit so I think it's more tied to specific places in game.
Here are my saves: Shadow of the Tomb Raider.zip
Save files are located in SteamLibrary/steamapps/compatdata/750920/pfx/drive_c/users/steamuser/Documents/Shadow of the Tomb Raider.
Slot 2 "Kuwaq yaku" is on a location that seems to crash consistently.
All graphics settings are on their maximum values and FidelityFX CAS is enabled. Resolution is at 1440p.
Video on how to trigger the crash.
https://github.com/user-attachments/assets/8c38833c-ada4-4d78-a188-eb0dd7c584cf
Thanks, that location hangs consistently for me as well on RX 9070. It doesn't seem to be a recent RADV regression, as I got the same hang with 25.0.0 as I did with 10550046.
It also hangs with RADV_DEBUG=hang or syncshaders, but umr segfaults, so I wasn't able to get wave dumps from this by any means. I get the same shader in my breadcrumbs as well, and the RADV dump contains the RayGenMain shader.
Whether it hangs in f6c7229abf691c7b or 388c22d4bcfb693f seems to vary with framerate/timing. With heavier debugging on RX 9070 or the same heavy graphics settings on RX 7600 it hangs in the second shader instead. When I added debugging to those shaders, I couldn't see any signs that the shader was getting stuck in a loop there, just a bunch of single iterations until the timeout.
I was able to get some wave dumps, but only with AMDVLK or an older RADV (I used 25.1.7 in this case), but when I tried to add ACO_DEBUG=force-waitcnt umr would segfault again. I tried a few different timeout intervals as well, but it didn't make any obvious difference as far as I can tell. Sometimes there are v_illegal instructions in the wave dumps, but I don't know if that's related.
Shadow of the Tomb Raider occasionally causes GPU hangs when ray tracing is enabled. Hangs are mostly reproducible in spots where they occur but can depend on things such as camera pointing to a specific direction.
Software information
Shadow of The Tomb Raider Windows version via Proton.
System information
Log files
steam-750920.log
Vkd3d log includes breadcrumbs. To my knowledge the dimensions for trace_rays on rows 218 and 219 and for dispatch on rows 283,284 and on some others seem suspiciously large.
vkd3d-sottr.log