Skip to main content

Install drivers for GPU node groups in a Managed Kubernetes cluster

Last update:

You can create Managed Kubernetes clusters on a cloud server with GPUs without pre-installed drivers. To install the driver yourself, use the NVIDIA® GPU Operator application.

  1. Connect to the cluster.

  2. Install the Helm package manager version 3.7.0 or higher.

  3. Add the nvidia repository to Helm:

    helm repo add nvidia https://helm.ngc.nvidia.com/nvidia

  4. Update the nvidia repository in Helm:

    helm repo update

  5. Install NVIDIA GPU Operator and specify the correct version of the GPU driver:

    helm install \
      --namespace gpu-operator \
      --create-namespace \
      --set driver.version=<driver_version> \
      gpu-operator nvidia/gpu-operator

    Specify <driver_version> is the NVIDIA® driver version. You can look in the NVIDIA GPU Driver row in the GPU Operator Component Matrix table of the NVIDIA® documentation.

  6. To verify that NVIDIA GPU Operator and the GPU driver are installed correctly, run a GPU application. For example, the CUDA VectorAdd vector addition application:

    cat << EOF | kubectl create -f -
    apiVersion: v1
    kind: Pod
    metadata:
    name: cuda-vectoradd
    spec:
    restartPolicy: OnFailure
    containers:
    - name: cuda-vectoradd
        image: "nvidia/samples:vectoradd-cuda11.2.1"
        resources:
        limits:
            nvidia.com/gpu: 1
    EOF

  7. Check that the CUDA VectorAdd application has completed successfully - the status of the feed should be Completed:

    kubectl get pods

    The cuda-vectoradd feed will have a status of Completed in the response:

    NAME                                  READY   STATUS        RESTARTS   AGE
    cuda-vectoradd                        0/1     Completed     0          51s