NVIDIA GPU Workloads

Mirantis Kubernetes Engine (MKE) supports running workloads on NVIDIA GPU nodes. Current support is limited to NVIDIA GPUs. MKE uses the NVIDIA GPU Operator to manage GPU resources on the cluster.

To enable GPU support, MKE installs the NVIDIA GPU Operator on your cluster.

Prerequisites

Before you can enable NVIDIA GPU support in MKE, you must install the following components on each GPU-enabled node:

Configuration

NVIDIA GPU support is disabled by default. To enable NVIDIA GPU support, configure the nvidiaGPU section of the MKE configuration file under devicePlugins:

devicePlugins:
  nvidiaGPU:
    enabled: true

Running GPU Workloads

Run a simple GPU workload that reports detected NVIDIA GPU devices:

cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  restartPolicy: Never
  containers:
    - name: cuda-container
      image: nvcr.io/nvidia/cloud-native/gpu-operator-validator:v22.9.0
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU
  tolerations:
  - key: nvidia.com/gpu
    operator: Exists
    effect: NoSchedule
EOF

Verify the successful completion of the pod:

kubectl get pods | grep "gpu-pod"

Example output:

NAME                        READY   STATUS    RESTARTS   AGE
gpu-pod                     0/1     Completed 0          7m56s

Upgrading

To upgrade an MKE 3 cluster with GPU enabled, make sure that you complete the GPU prerequisites before you start the upgrade process. Failing to do this will cause the upgrade process to detect the GPU configuration in MKE 3 and incorrectly transfer it to MKE 4.

Last updated on