NVIDIA GPU Operator¤

Helm chart link: 051-nvidia-gpu-operator

Configuration¤

Values¤

nvidia_gpu_operator:
  # NFD (Node Feature Discovery) configuration
  # NFD is a prerequisite for GPU operator
  nfd_namespace: "openshift-nfd"
  nfd_channel: "stable"
  nfd_install_plan: Automatic
  nfd_image: ""  # Optional: custom NFD image

  # GPU Operator configuration
  # Namespace where GPU operator will be installed
  # Default: nvidia-gpu-operator
  gpu_namespace: "nvidia-gpu-operator"

  # GPU Operator subscription channel
  # Default: v24.3
  gpu_channel: "v24.3"

  # NVIDIA GPU driver version
  # Specify the driver version to install
  # Default: 575.57.08
  gpu_driver_version: 575.57.08

  # GPU driver repository path
  # Container registry path for GPU drivers
  # Default: nvcr.io/nvidia
  gpu_driver_repository_path: "nvcr.io/nvidia"

  # GPU Operator install plan approval
  # Options: "Automatic" or "Manual"
  # Default: Automatic
  gpu_install_plan: Automatic

Base Cluster Values¤

This chart inherits common cluster configuration values. The most frequently used base values are:

account:
  id: string                    # AWS account identifier

region:
  id: string                    # AWS region identifier

cluster:
  id: string                    # Unique cluster identifier
  url: string                   # OpenShift cluster API URL
  nonshared: boolean            # Whether cluster is dedicated (true) or shared (false)

sm:                             # Secrets Manager configuration
  aws_access_key_id: string (secret reference)
  aws_secret_access_key: string (secret reference)

For complete documentation of all base cluster values including optional fields like notifications, custom_labels, devops, and cli_image_repo, see the Cluster Base Values Reference.

Usage Examples¤

Basic GPU operator installation:

nvidia_gpu_operator:
  nfd_namespace: "openshift-nfd"
  nfd_channel: "stable"
  nfd_install_plan: Automatic
  gpu_namespace: "nvidia-gpu-operator"
  gpu_channel: "v24.3"
  gpu_driver_version: 575.57.08
  gpu_driver_repository_path: "nvcr.io/nvidia"
  gpu_install_plan: Automatic

With specific driver version:

nvidia_gpu_operator:
  nfd_namespace: "openshift-nfd"
  nfd_channel: "stable"
  nfd_install_plan: Automatic
  gpu_namespace: "nvidia-gpu-operator"
  gpu_channel: "v24.3"
  gpu_driver_version: 550.90.07
  gpu_driver_repository_path: "nvcr.io/nvidia"
  gpu_install_plan: Automatic

With custom driver repository:

nvidia_gpu_operator:
  nfd_namespace: "openshift-nfd"
  nfd_channel: "stable"
  nfd_install_plan: Automatic
  gpu_namespace: "nvidia-gpu-operator"
  gpu_channel: "v24.3"
  gpu_driver_version: 575.57.08
  gpu_driver_repository_path: "my-registry.example.com/nvidia"
  gpu_install_plan: Automatic

Prerequisites¤

• OpenShift cluster with GPU-enabled nodes
• Sufficient cluster resources for GPU workloads
• Node Feature Discovery (NFD) operator (automatically installed by this chart)

Resources Created¤

Resource Type	Resource Name	Namespace	Condition	Installed By
OperatorGroup	nvidia-gpu-operator-group	nvidia-gpu-operator	Always	cluster_admin_role
Subscription	gpu-operator-certified	nvidia-gpu-operator	Always	cluster_admin_role
ClusterPolicy	gpu-cluster-policy	N/A (cluster-scoped)	Always	cluster_admin_role
SecurityContextConstraints	ibm-mas-customscc	N/A (cluster-scoped)	Always	cluster_admin_role