100% CPU utilization (RunPod, Vast) when running a Proton game or vkcube

[BranislavF]

Hello,

I am facing a strange issue with 100% CPU utilization for this container even with running a simple "vkcube". Stream is not able to get 60 FPS, and it is choppy. The same applies when trying to run a Proton game with the Steam hotfix for unprivileged containers. It just bumps CPU to 100%. I can do more testing and provide more logs/investigation if needed.

This was observed on RunPod and Vast.AI machines.

Thank you.

[ehfd]

This is very, very likely that the GPU has not been configured correctly. Need the logs as well as output from glxinfo and vulkaninfo.

[BranislavF]

Attaching logs from glxinfo and vulkaninfo + screenshot from idle and vkcube running. Let me know if anything else is required :)

glxinfo.txt

vulkaninfo.txt

[ehfd]

Let's see results for nvidia-smi dmon -i 0 while this is happening.

[BranislavF]

It seems that there is nothing unusual from the output. These are the screens from htop and nvidia-smi dmon -i 0 when trying to launch a Steam Proton game

[BranislavF]

UPDATE: So, there is a change in the output from nvidia-smi dmon -i 0 See attached screens. Maybe it helps. But this issue is present just on RunPod/Vast. I tried that on another cloud provider (with full root access to VM) and a docker run command, and this issue was not there.

[ehfd]

I personally think that it's a driver bug I found some time before, the provider should upgrade the drivers.

[ehfd]

I personally think that it's a driver bug I found some time before, the provider should upgrade the drivers.
Let's see the output for just nvidia-smi just to be sure.

[BranislavF]

I think I found the root cause for this issue. FSYNC needs to be disabled when running steam.

https://github.com/ValveSoftware/Proton?tab=readme-ov-file#runtime-config-options

This needs to be followed and now even steam Proton games are running fine. Not sure why FSYNC is not working but at least this solved it for me. Now RunPod or Vast can be used for gaming, with some tweaks :)

Let me know if you would like to track it somehow, or if I can close it.

[ehfd]

Thanks for the pointer!

I'll keep this open and close when there is sufficient intervention or documentation available.

[kryuchenko]

@BranislavF How did you manage to launch it on RunPod? All UPD ports are closed there.

[BranislavF]

@BranislavF How did you manage to launch it on RunPod? All UPD ports are closed there.

You need external TURN server for running that. Anyway, using runpod for gaming is complicated and tricky to set everything up. A strong host CPU is also needed for correct run.