Skip to content

Latest commit

 

History

History
34 lines (26 loc) · 1.12 KB

nvidia-gpu-operator.md

File metadata and controls

34 lines (26 loc) · 1.12 KB
Raw

Nvidia GPU operator installation

After a new k8s installation, the Nvidia GPU operator must be installed to run GPU tasks. Visit the official repository to get more info about the operator: https://github.com/NVIDIA/gpu-operator/

Prerequisites

  • Nodes must not be pre-configured with NVIDIA components (driver, container runtime, device plugin).
  • Node Feature Discovery (NFD) is required on each node. By default, NFD master and worker are automatically deployed.

Installation

Note: Helm 3 should be already installed on the pharma console host, if it's not then please install it by executing the following script:

$ curl https://raw.githubusercontent.com/helm/helm/master/scripts/get-helm-3 | bash

Add and update Nvidia repository

$ helm repo add nvidia https://nvidia.github.io/gpu-operator
$ helm repo update

Install the GPU operator

$ helm install nvidia-gpu-operator nvidia/gpu-operator --devel --wait

Deploy a test GPU job Note: assigning a job to a worker node may take about 3 minutes

kubectl apply -f https://nvidia.github.io/gpu-operator/notebook-example.yml
kubectl get pod tf-notebook