kbidarkar/ocp-nvidia-vgpu-installer

NVIDIA GRID driver delivery with OpenShift Driver-Container Toolkit

This repository is a container image that is based on the [OpenShift Driver-Container Toolkit}(https://github.com/openshift/driver-toolkit). This container image installs the NVIDIA's generic Linux GRID driver on Red Hat CoreOS nodes.

Requirements

• The generic GRID Linux installer for vGPU should be obtained from the NVIDIA Licensing portal. This repo uses the NVIDIA-Linux-x86_64-470.63-vgpu-kvm.run
• Knowledge of the host's kernel version.
• An image of the OpenShift Driver-Container Toolkit that matches the target cluster version.

Obtaining the driver toolkit base image

$ oc adm release info --image-for=driver-toolkit
quay.io/openshift-release-dev/ocp-v4.0-art-dev@sha256:e0f9b9154538af082596f60af99290b5a751e61fd61100912defb71b6cac15c6

Configure

• Update the FROM in the Dockerfile with the relevant driver toolkit image.
• Update the relevant ADD in the Dockerfile with the relevant NVIDIA installer binary (E.g. NVIDIA-Linux-x86_64-470.63-vgpu-kvm.run)
• Update the entrypoint.sh file with the host's kernel version. (E.g. 4.18.0-305.17.1.el8_4.x86_64)

Building the container image:

podman build -t ocp-nvidia-vgpu-nstaller .
podman push [registry_url]/ocp-nvidia-vgpu-nstaller

Deployment

• Update the 1000-drivercontainer.yaml to point to the container image.
• Use an existing node label or label the relevant nodes where this driver should be installed
  • oc label nodes worker-0 hasGpu=true
• Apply the 1000-drivercontainer.yaml to the cluster
  • oc create -f 1000-driverscontainer.yaml