After several minutes of running q2 rtx, GPU seems to crash with an XID 31. The exact fault messages seem to vary. Good news for me is that I can ssh and kill the process and the desktop comes back to normal.
Current driver: 430.26
Seemed to be the same with 418.56
nvidia-bug-report.log.gz (1.09 MB)
Please check if this applies:
https://devtalk.nvidia.com/default/topic/1056067/linux/virtualbox-freezes-my-system-when-on-nvidia-but-not-when-on-intel-hd/
It feels like it ran a little longer this way: yesterday 40 mins standing still, had to turn off.
Today I tried playing, so it lasted 15 mins and then crashed with a similar message. Attached.
nvidia-bug-report.log.gz (1.06 MB)
OK, something very weird is going on. DOOM (2016) crashes during the first 2 seconds of launching the game (vulkan version) with XID 13, while witcher 3 managed to run ~20 mins before crash (XID 13).
In all cases I had export __GL_MaxFramesAllowed=1.
Interestingly, XID 13 in these cases, while quake rtx was persistent with XID 31.
doom-nvidia-bug-report.log.gz (1.04 MB)
witcher3-nvidia-bug-report.log.gz (1.03 MB)
Now that could be anything.
kwin: did you try to lauch an X session with e.g. openbox and check if it still crashes?
driver: IIRC, the latest Vulkan beta 418.52.10 has some fixes not mainlined yet
hw: running cuda-memtest shouldn’t harm either.
Indeed, there seem to be three problems here: two of them is a conflict of kwin with nvidia, namely crashes in quake 2 rtx and crash in witcher 3. They are gone by using openbox.
On the other hand, doom crashes on the launch on openbox too. Most likely a vulkan extension bug.
Which kwin version are you running?
In the doom case, you can only test the current vulkan beta driver to see if this is a new bug or a fixed bug not mainlined yet.