April 24Apr 24 Community Expert Hey allSo I have a container running ollama. Everything was working when I went to bed last night. This morning, I allowed the server o upgrade to rc1. Immediately thhe ollama container is failing with this:--gpus=all 'ollama/ollama'3c5750fb5f4508a3bb986b33dabcf4b0c48aeaae14ccb3c6b0c95ef0e59b2f58docker: Error response from daemon: failed to create task for container: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running prestart hook #0: exit status 1, stdout: , stderr: Auto-detected mode as 'legacy'nvidia-container-cli: ldcache error: process /sbin/ldconfig failed with error code: 1The container will not start. If I remove gpus=all it works on pure cpu.So I have tried:Reinstalling container.Changing back to r2.Downgrading driver and moving from closed to open, etc.Stop docker and restart docker service.Nothign worked. Maybe I didn't do those steps ina specifc order or something dunno..Thoughts appreciated (update - whoops forgot diagnostics) companion-diagnostics-20260424-1308.zip Edited April 24Apr 24 by enesha
April 24Apr 24 Community Expert Better to ask in [Plugin] Nvidia-Driver - Plugin Support - UnraidAnd also post the diags there.
April 24Apr 24 Author Community Expert Solution Yeah My bad. Didn't think of tha . Ah well. thought I did a search too but I guess I was deficient there.So - I did look there, and was able to find the solution (for me at least) and I will add it here just in case anyone runs across this in the future. It was simple, just change Extra Parameters to: --gpus all --runtime=nvidiathat specific runtime was the keyThanks!
April 24Apr 24 Community Expert FYI it's only ever been --runtime=nvidia in extra parameters.Not sure where people started getting --gpus all from...
April 24Apr 24 Community Expert --gpus=all is the modern, supported method for gpu passthrough as per Dockers documentation (linked).It does work on Unraid, I use it currently in 7.2, but there are regressions in 7.3.x, which look likely to be issues with nvidia-container-toolkit.https://docs.docker.com/engine/containers/gpu/
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.