Getting Intel GPUs to work in your Kubernetes cluster is fairly straightforward; two helm charts are required one not needing any configuring and the second is extremely simple to configure. There is a hard dependency on cert-manager and a pseudo-dependency on Node Feature Discovery
For this tutorial I assume you have the following:
- A Working K3S cluster (If you don't have this I recommend this video by TechnoTim to get you started)
- At least one node with some sort of Intel GPU
- Helm installed
- Cert Manager installed
For my cluster I use Debian 12
In order to effectively use your Intel GPU for quicksync we need to install a package, intel-media-va-driver-non-free. Since I use Debian's cloud images the step to enable to correct repositories is slightly different for some reason:
sudo sed -i 's/^Components: main$/Components: main non-free non-free-firmware/' /etc/apt/sources.list.d/debian.sources
If you installed Debian normally you would want to edit /etc/apt/sources.list, the first line is likely the correct one to edit:
sudo nano /etc/apt/sources.list
deb http://mirror.csclub.uwaterloo.ca/debian/ bookworm main
Stick non-free non-free-firmware onto the end so that it looks something like this:
deb http://mirror.csclub.uwaterloo.ca/debian/ bookworm main non-free-firmware contrib
Press Ctrl+X to save, then Y to confirm and then enter.
sudo apt-get update
sudo apt-get install intel-media-va-driver-non-free
Optionally you can install vainfo to see what codecs your GPU supports:
$ vainfo
error: can't connect to X server!
libva info: VA-API version 1.17.0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_17
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.17 (libva 2.12.0)
vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 23.1.1 ()
vainfo: Supported profile and entrypoints
VAProfileNone : VAEntrypointVideoProc
VAProfileNone : VAEntrypointStats
VAProfileMPEG2Simple : VAEntrypointVLD
VAProfileMPEG2Simple : VAEntrypointEncSlice
VAProfileMPEG2Main : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointVLD
VAProfileH264Main : VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointFEI
VAProfileH264Main : VAEntrypointEncSliceLP
VAProfileH264High : VAEntrypointVLD
VAProfileH264High : VAEntrypointEncSlice
VAProfileH264High : VAEntrypointFEI
VAProfileH264High : VAEntrypointEncSliceLP
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD
VAProfileJPEGBaseline : VAEntrypointVLD
VAProfileJPEGBaseline : VAEntrypointEncPicture
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
VAProfileH264ConstrainedBaseline: VAEntrypointFEI
VAProfileH264ConstrainedBaseline: VAEntrypointEncSliceLP
VAProfileVP8Version0_3 : VAEntrypointVLD
VAProfileVP8Version0_3 : VAEntrypointEncSlice
VAProfileHEVCMain : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointEncSlice
VAProfileHEVCMain : VAEntrypointFEI
VAProfileHEVCMain10 : VAEntrypointVLD
VAProfileHEVCMain10 : VAEntrypointEncSlice
VAProfileVP9Profile0 : VAEntrypointVLD
VAProfileVP9Profile2 : VAEntrypointVLD
The codecs that have VAEntrypointEncSlice are the ones your gpu can encode, in my case I can encode up to and including H265 10bit. Your results may vary.
The Intel GPU Device Plugin chart we'll be installing will create a DaemonSet with a node selector of intel.feature.node.kubernetes.io/gpu: true so that any pods created by the DaemonSet will only run on nodes with Intel GPUs. Here you have two options:
- Manually label nodes
- Use Node Feature Discovery to label appropriate nodes
Manually tagging your nodes is simple:
kubectl label nodes NODE-NAME intel.feature.node.kubernetes.io/gpu=true
And that's it. Do that for each node that has an Intel GPU.
Node Feature Discovery uses rules to determine what to label and why. The Intel GPU Device Plugin that we'll be installing comes with pre-made rules to automatically label your nodes.
helm repo add node-feature-discovery https://kubernetes-sigs.github.io/node-feature-discovery/charts
helm repo update
helm upgrade --install node-feature-discovery node-feature-discovery/node-feature-discovery --version 0.14.3
helm repo add intel https://intel.github.io/helm-charts/
helm repo update
helm upgrade --install device-plugin-operator intel/intel-device-plugins-operator
helm upgrade --install gpu-device-plugin intel/intel-device-plugins-gpu --values intel/values.yaml
The values.yaml file is quite sparse, having only two key-pairs: sharedDevNum: 4 and nodeFeatureRule: true. Setting sharedDevNum something greater than 1 allows N pods to use the same GPU on a node. nodeFeatureRule: true adds Node Feature Discovery rules for detecting various Intel GPUs
And that's it! You are now ready to use your Intel GPU in a pod.
I have set up Jellyfin to use the GPU of the node it's running on for transcoding. Configuring a container is simple:
containers:
- name: jellyfin
image: jellyfin/jellyfin:10.8.13-1
resources:
limits:
gpu.intel.com/i915: "1"
requests:
gpu.intel.com/i915: "1"
If some but not all of your nodes have an Intel GPU you'll want to set up a node selector:
spec:
nodeSelector:
intel.feature.node.kubernetes.io/gpu: "true"
Apply the manifest then go into the Playback settings of Jellyfin and choose Intel QuickSync.
Sources: