How to enable NVIDIA GPU on CoolerControl?

Posted by ABOhiccups

If you have NVIDIA GPU running in SparkBox like Custom Single Slot Cooler by n3rdware. Open Terminal and SSH into SparkBox. Enter this command. sudo nano /opt/sparkbox/state/overrides/coolercontrol.yml or sudo msedit /opt/sparkbox/state/overrides/coolercontrol.yml Paste this into yml file and save. services: coolercontrold: deploy: resources: reservations: devices: - driver: nvidia count: all capabilities: [gpu] Make sure you have NVIDIA Container Toolkit installed. Enter the following commands. sudo nvidia-ctk runtime configure --runtime=docker sudo sparkbox restart Congratulations! You've successfully enabled NVIDIA GPU in CoolerControl.

3 replies

Chris wrote:

Thanks for writing this up. Two steps need adding so it works on other boxes: 1. The nvidia-ctk command only edits Docker's configuration, and Docker has to restart before it takes effect. Run this after it and before the SparkBox restart (it briefly restarts every app on the box, so do it when nothing is downloading or transcoding): sudo systemctl restart docker 2. The CoolerControl module sets NVIDIAVISIBLEDEVICES to empty by default. Add this to the same override file, under services, coolercontrold: environment: - NVIDIAVISIBLEDEVICES=all Then run sudo sparkbox restart as you did. If the GPU still does not show in CoolerControl, reply here and tell us which step you were on.

ABOhiccups wrote:

environment: - NVIDIAVISIBLEDEVICES=all is not compatible with CoolerControl. The other custom docker-compose override setup used in this tutorial post is compatible. services: coolercontrold: deploy: resources: reservations: devices: - driver: nvidia count: all capabilities: [gpu]

Chris wrote:

I'm sorry, the environment line I added was a wrong step. I suggested it as required without testing it on an NVIDIA box, and you've found it breaks CoolerControl on your setup. Please remove the environment lines from the override so it matches your post again. The GPU reaches the container through the device reservation in that override, and the docker restart after nvidia-ctk is still needed. What I knew about this has been corrected, so the next person won't be sent to that line. One check: after you remove it, run sudo sparkbox restart coolercontrol and tell us whether the NVIDIA GPU shows up in CoolerControl.