Version v1.2 of the documentation is no longer actively maintained. The site that you are currently viewing is an archived snapshot. For up-to-date documentation, see the latest version.

NVIDIA GPU (Proprietary drivers)

In this guide we’ll follow the procedure to support NVIDIA GPU using proprietary drivers on Talos.

Enabling NVIDIA GPU support on Talos is bound by NVIDIA EULA. Talos GPU support has been promoted to beta.

These are the steps to enabling NVIDIA support in Talos.

  • Talos pre-installed on a node with NVIDIA GPU installed.
  • Building a custom Talos installer image with NVIDIA modules
  • Upgrading Talos with the custom installer and enabling NVIDIA modules and the system extension

This requires that the user build and maintain their own Talos installer image.

Prerequisites

This guide assumes the user has access to a container registry with push permissions, docker installed on the build machine and the Talos host has pull access to the container registry.

Set the local registry and username environment variables:

export USERNAME=<username>
export REGISTRY=<registry>
export TAG=v1.2.8-nvidia

For eg:

export USERNAME=talos-user
export REGISTRY=ghcr.io

The examples below will use the sample variables set above. Modify accordingly for your environment.

Building the installer image

Start by cloning the pkgs repository.

Now run the following command to build and push custom Talos kernel image and the NVIDIA image with the NVIDIA kernel modules signed by the kernel built along with it.

make kernel nonfree-kmod-nvidia PLATFORM=linux/amd64 PUSH=true

Replace the platform with linux/arm64 if building for ARM64

Now we need to create a custom Talos installer image.

Start by creating a Dockerfile with the following content:

FROM scratch as customization
COPY --from=ghcr.io/talos-user/nonfree-kmod-nvidia:v1.2.8-nvidia /lib/modules /lib/modules

FROM ghcr.io/siderolabs/installer:v1.2.8
COPY --from=ghcr.io/talos-user/kernel:v1.2.8-nvidia /boot/vmlinuz /usr/install/${TARGETARCH}/vmlinuz

Now build the image and push it to the registry.

DOCKER_BUILDKIT=0 docker build --squash --build-arg RM="/lib/modules" -t ghcr.io/talos-user/installer:v1.2.8-nvidia .
docker push ghcr.io/talos-user/installer:v1.2.8-nvidia

Note: buildkit has a bug #816, to disable it use DOCKER_BUILDKIT=0 Replace the platform with linux/arm64 if building for ARM64

Upgrading Talos and enabling the NVIDIA modules and the system extension

Make sure to use talosctl version v1.2.8 or later

First create a patch yaml gpu-worker-patch.yaml to update the machine config similar to below:

- op: add
  path: /machine/install/extensions
  value:
    - image: ghcr.io/siderolabs/nvidia-container-toolkit:515.65.01-v1.10.0
- op: add
  path: /machine/kernel
  value:
    modules:
      - name: nvidia
      - name: nvidia_uvm
      - name: nvidia_drm
      - name: nvidia_modeset
- op: add
  path: /machine/sysctls
  value:
    net.core.bpf_jit_harden: 1

Now apply the patch to all Talos nodes in the cluster having NVIDIA GPU’s installed:

talosctl patch mc --patch @gpu-worker-patch.yaml

Now we can proceed to upgrading Talos with the installer built previously:

talosctl upgrade --image=ghcr.io/talos-user/installer:v1.2.8-nvidia

Once the node reboots, the NVIDIA modules should be loaded and the system extension should be installed.

This can be confirmed by running:

talosctl read /proc/modules

which should produce an output similar to below:

nvidia_uvm 1146880 - - Live 0xffffffffc2733000 (PO)
nvidia_drm 69632 - - Live 0xffffffffc2721000 (PO)
nvidia_modeset 1142784 - - Live 0xffffffffc25ea000 (PO)
nvidia 39047168 - - Live 0xffffffffc00ac000 (PO)
talosctl get extensions

which should produce an output similar to below:

NODE           NAMESPACE   TYPE              ID                                                                 VERSION   NAME                       VERSION
172.31.41.27   runtime     ExtensionStatus   000.ghcr.io-frezbo-nvidia-container-toolkit-510.60.02-v1.9.0       1         nvidia-container-toolkit   510.60.02-v1.9.0
talosctl read /proc/driver/nvidia/version

which should produce an output similar to below:

NVRM version: NVIDIA UNIX x86_64 Kernel Module  510.60.02  Wed Mar 16 11:24:05 UTC 2022
GCC version:  gcc version 11.2.0 (GCC)

Deploying NVIDIA device plugin

First we need to create the RuntimeClass

Apply the following manifest to create a runtime class that uses the extension:

---
apiVersion: node.k8s.io/v1
kind: RuntimeClass
metadata:
  name: nvidia
handler: nvidia

Install the NVIDIA device plugin:

helm repo add nvdp https://nvidia.github.io/k8s-device-plugin
helm repo update
helm install nvidia-device-plugin nvdp/nvidia-device-plugin --version=0.11.0 --set=runtimeClassName=nvidia

Apply the following manifest to run CUDA pod via nvidia runtime:

cat <<EOF | kubectl apply -f -
---
apiVersion: v1
kind: Pod
metadata:
  name: gpu-operator-test
spec:
  restartPolicy: OnFailure
  runtimeClassName: nvidia
  containers:
  - name: cuda-vector-add
    image: "nvidia/samples:vectoradd-cuda11.6.0"
    resources:
      limits:
         nvidia.com/gpu: 1
<<EOF

The status can be viewed by running:

kubectl get pods

which should produce an output similar to below:

NAME                READY   STATUS      RESTARTS   AGE
gpu-operator-test   0/1     Completed   0          13s
kubectl logs gpu-operator-test

which should produce an output similar to below:

[Vector addition of 50000 elements]
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Copy output data from the CUDA device to the host memory
Test PASSED
Done