NVIDIA GPU Operator

Description and Purpose

The NVIDIA GPU Operator deploys the NVIDIA GPU software stack required to expose and manage NVIDIA GPUs in Kubernetes. Use it on K8s Clusters with NVIDIA GPU worker nodes so workloads can request GPU resources through Kubernetes.

Parameters

This application has no application-specific user inputs. During installation, provide the standard release name, target namespace, and namespace-creation option.

The built-in Helm values enable the NVIDIA Container Device Interface integration for the NRI plugin:

cdi:
  nriPluginEnabled: true

Usage and Configuration

To install this application, follow the Managing Applications guide using the catalog ID shown above.

From the OpenNebula Front-end, retrieve the kubeconfig for the target K8s Cluster:

oneks show cluster <CLUSTER_ID> --kubeconfig > kubeconfig

After installation, use the K8s Cluster kubeconfig to verify the operator workloads. The following example assumes that the application was installed with gpu-operator as both the release name and target namespace:

KUBECONFIG=./kubeconfig kubectl get pods -n gpu-operator

After the release reaches ready, verify that GPU resources are advertised by the intended worker nodes:

KUBECONFIG=./kubeconfig kubectl get nodes \
  -o custom-columns=NAME:.metadata.name,GPUS:.status.capacity.nvidia\.com/gpu

For GPU Operator configuration and validation procedures, follow the upstream documentation linked above.