Chapter 2. AI accelerator prerequisites for Openshift Container Platform

For Openshift Container Platform clusters, install the NVIDIA GPU Operator and Node Feature Discovery (NFD) Operator to manage GPU nodes.

Prerequisites

You have installed the Node Feature Discovery (NFD) Operator from OperatorHub.
You have installed the NVIDIA GPU Operator from OperatorHub and created a ClusterPolicy custom resource. For more information, see the NVIDIA GPU Operator on Openshift Container Platform documentation.

Supported GPU types

The following NVIDIA GPU types are supported for Distributed Inference with llm-d on Openshift Container Platform:

After provisioning GPU nodes and installing the GPU Operator, verify GPU discovery:

$ oc get nodes -l nvidia.com/gpu.present=true

Output shows nodes with the nvidia.com/gpu.present=true label.

Verify that the GPU device plugin pods are running:

$ oc get pods -n nvidia-gpu-operator -l app=nvidia-device-plugin-daemonset

Verify allocatable GPU resources on each node:

$ oc describe node <gpu_node_name> | grep -A 5 "nvidia.com/gpu"