510.39.01 BETA driver: nvidia-powerd causing high system load

On my Thinkpad T15g Gen 2 (RTX 3070), the new 510.39.01 BETA driver causes high system loads (100 % CPU utilization on single thread).
The process causing it is nvidia-powerd.

Any ideas what might be causing this?

Which distribution are you using? How did you get nvidia-powerd installed?

Ah, sorry for not providing this information from the get-go.

I am running Ubuntu 20.04 (kernel 5.13.0-25).

I honestly have no clue how nvidia-powerd got installed. Yesterday I had (probably unrelated) driver issues with driver version 495 which is why I decided to try 510 Beta to see if this works. I uninstalled 495 (via apt purge) and installed 510 (via “Software & Updates” → “Additional Drivers”). Afterwards I also reinstalled CUDA toolkit 11.6 (and some other Machine Learning related utilities, e.g. CUDNN, TensorRT) as well as Nvidia-Docker2…

I only noticed the issue with nvidia-powerd when hearing my fans runnning non-stop. This was after reinstalling everything, so I cannot pinpoint it to a single package.

Just stop and disable it:
sudo systemctl stop nvidia-powerd
sudo systemctl disable nvidia-powerd
Please run nvidia-bug-report.sh as root and attach the resulting nvidia-bug-report.log.gz file to your post.

Alright, stopped and disabled the service.
Attached you can also find my bug report. Thanks a lot.

nvidia-bug-report.log.gz (471.6 KB)

The service is suspected to fix this:
currently propably only for vendor-testing so it shouldn’t be installed and started by the regular package manager.
It obviously has some bugs but it’s starting on your system. If you like, you could try experimenting with it, checking nvidia-settings for the new power profile pane.

Same issue here.
HP Zbook G8 RTX3080 (Ubuntu 20.04)

In some cases the nvidia-powerd process get stuck in the 100% and the fans starts to blow to the max power. Rebooting the service, go to the normally at least for 1/2 hours.

1 Like

Please also report this to Ubuntu since nvidia-powerd is currently not meant for general distribution.

Thanks for your answer.
The first point is that service is enable by default.
The second point is than we can reach the maximum power utilization of the GPU. I can’t see more power usage than 52W, runing a deepstream code.
And this is a problem to use the computer with a deep learning purpose.

So i continue thinking that is a typical bug of the Nvidia Drivers.