Xid 8 system lockup under vulkan (dxvk), driver 440.44

Graphics / Linux Linux
1

The system will lock up reliably with Xid 8, channel 00000026 when running vulkan accelerated content, this error is reliably reproduced using The Witcher 3: Wild Hunt v1.32 running under Proton 4.11-11, using dxvk – notably in the Velen area after about 1-2 hours of gameplay on average, sometimes less (as fast as 20 minutes), sometimes more (up to 3 hours)

The system will not respond to CTRL+ALT+F1, or otherwise be meaningfully interacted with – even with SysRq+R – I hypothesize it drops to tty just fine but won’t output an appropriate video signal to indicate such due to the GPU entering a hung state (SysRq+S implies as much, since it flushes to a non-responsive tty, or a black screen – 50/50 really)

Jan 14 22:42:45 Valhalla kernel: NVRM: GPU at PCI:0000:1c:00: GPU-fed4696b-9f6c-70db-2f56-dbd8040928b4
Jan 14 22:42:45 Valhalla kernel: NVRM: GPU Board Serial Number: 
Jan 14 22:42:45 Valhalla kernel: NVRM: Xid (PCI:0000:1c:00): 8, pid=704, Channel 00000026

nvidia-bug-report.log.gz (253 KB)

2

Likely a thermal problem with local overheating before the temp sensor is catching up. I fixed it with re-applying thermal paste to the chip. It was dry as a brick and covered only half the chip.