Edit: For those who stumble across this with the same issue, I eventually got it working by adding “default-runtime”: “nvidia”, to /etc/docker/daemon.json then restarting the docker service and Jellyfin container.
I am in the process of setting up a new media server on an old PC using Ubuntu Server and CasaOS and have run into my first major roadblock.
To give some background, I formerly had my media server running on my main gaming PC on Windows using Plex and the *arr suite. I’m now trying to do things the right way and set everything back up from scratch on some spare hardware with Jellyfin and all the rest in dockerized containers. I chose CasaOS because I’m not overly familiar with Linux and thought that would be a good way to ease into things.
Everything was going well until I tried to get hardware acceleration enabled in Jellyfin. For the life of me I cannot seem to get the Nvidia drivers properly installed, much less give Jellyfin access to the device. I’m using a GTX 960.
I’m not sure exactly what additional info I need to give here, but here’s something I hope helps:
*****@home-server:/$ nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
*****@home-server:/$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0
*****@home-server:/$ ls /usr/src | grep nvidia
nvidia-srv-535.104.12
*****@home-server:/$ sudo dkms install -m nvidia -v srv-535.104.12
Error! Could not locate dkms.conf file.
File: /usr/src/nvidia-srv-535.104.12/dkms.conf does not exist.
If there’s anything important I’m leaving out - and I probably am - let me know. Also if there’s anywhere else you recommend I post this let me know that as well.
Be careful with your solution. Ever other container you launch will attempt to access the Nvidia runtime, which gives access to the GPU. This may have unintended consequences, versus just running individual containers with Nvidia runtime set, and gpus=all set.