NVIDIA GPU hardware acceleration for Jellyfin on Docker inside a Proxmox LXC

For virtualizing my own PC with Proxmox I had to add another GPU (a cheap Quadro P620) as the host output, which I also planned on using for hardware acceleration for my media collection in Jellyfin. This is how to make that work.

Overview of the setup

• Proxmox VE host
• Podman (or Docker) in an LXC
• Jellyfin running using the official container

Guide

Important hints upfront if you just came here seeking answers

• This is about using a GPU that is not passed through to a VM, but rather used for the host.
• The NVIDIA driver in the LXC and on the Proxmox host need to be the same version.

NVIDIA driver setup on the Proxmox host

For the hardware acceleration to work with all codecs supported by the GPU ¹², we need the proprietary NVIDIA drivers on the host instead of nouveau.

Blacklist the nouveau driver kernel module from being loaded on boot in /etc/modprobe.d/blacklist.conf ³, it should look like this:

blacklist nouveau

Download the proprietary drivers from NVIDIAs download page, selecting the GPU you want to use and Linux 64-bit.

Install the drivers, choosing “Yes” for registering it with DKMS, e.g.:

chmod +x ./NVIDIA-Linux-x86_64-525.89.02.run
./NVIDIA-Linux-x86_64-525.89.02.run

Reboot, and run nvidia-smi to test if the driver works properly.

We will also need to load the CUDA kernel module. Add the following two lines to /etc/modules-load.d/modules.conf⁴:

nvidia
nvidia_uvm

and update initramfs:

update-initramfs -u

To have the required device nodes we need to pass to the LXC later created, add the following to /etc/udev/rules.d/70-nvidia-rules⁴:

# Create /nvidia0, /dev/nvidia1 and /nvidiactl when nvidia module is loaded
KERNEL=="nvidia", RUN+="/bin/bash -c '/usr/bin/nvidia-smi -L && /bin/chmod 666 /dev/nvidia*'"
# Create the CUDA node when nvidia_uvm CUDA module is loaded
KERNEL=="nvidia_uvm", RUN+="/bin/bash -c '/usr/bin/nvidia-modprobe -c0 -u && /bin/chmod 0666 /dev/nvidia-uvm*'"

Reboot the server, and make sure you see the following device nodes:

/dev/nvidia-uvm
/dev/nvidia-uvm-tools
/dev/nvidia0
/dev/nvidiactl

Setting up the LXC

Create a normal (Unprivileged is fine) LXC in the GUI with the OS of your choice (it does not have to be the same as the host), and make sure nesting and keyctl are enabled in Options -> Features.

Then, edit the LXC config file in /etc/pve/lxc/<id>.conf and add the following to mount the device nodes:

lxc.cgroup2.devices.allow: c 226:0 rwm
lxc.cgroup2.devices.allow: c 226:128 rwm
lxc.mount.entry: /dev/nvidia0 dev/nvidia0 none bind,optional,create=file
lxc.mount.entry: /dev/nvidiactl dev/nvidiactl none bind,optional,create=file
lxc.mount.entry: /dev/nvidia-uvm dev/nvidia-uvm none bind,optional,create=file
lxc.mount.entry: /dev/nvidia-uvm-tools dev/nvidia-uvm-tools none bind,optional,create=file

Setting up NVIDIA drivers inside the LXC

Now in the LXC: Install the same version NVIDIA drivers downloaded before, passing --no-kernel-module since we are using the driver loaded in the host kernel. E.g.:

chmod +x ./NVIDIA-Linux-x86_64-525.89.02.run
./NVIDIA-Linux-x86_64-525.89.02.run --no-kernel-module

Then, run nvidia-smi to ensure the driver works and the GPU shows up.

Setting up Docker/Podman in the LXC and running Jellyfin

Install either Docker or Podman depending on your preference (I recommend Podman as its the replacement for the Docker daemon by RedHat).
Notes for Podman:

1. Enable podman.service to start Podman on boot
2. Enable podman-restart.service to start all containers flagged with restart: always on boot.

Install the NVIDIA Container toolkit according to their official instructions based on what platform your LXC is using.

Follow the instructions on how to start a container with the NVIDIA container toolkit hooks, in my case with podman it looks like this:

podman run -d --name jellyfin --restart=always -v /srv/jellyfin/config:/config -v /srv/jellyfin/cache:/cache -v /mnt/media:/media --security-opt=label=disable --hooks-dir=/usr/share/containers/oci/hooks.d/ -p 8096:8096 -p 8920:8920 -p 1900:1900/udp -p 7359:7359/udp jellyfin/jellyfin:latest

The last step would now be to enable Hardware decoding in Jellyfin and check all options that your GPU supports.

References and credits: