Install Overview#

The NVIDIA GPU Operator supports many Kubernetes platforms, licensing models, and deployment configurations. This page helps you choose the install guide that matches your environment. Each linked page contains the step-by-step procedures; start here when you are not sure where to begin.

Before you install, review Prerequisites to confirm that your cluster, hardware, and tooling meet the requirements for your chosen path.

Red Hat OpenShift Container Platform

If you run Red Hat OpenShift, install the GPU Operator through OperatorHub or the OpenShift CLI (OLM), not the Helm quickstart in this book. Refer to NVIDIA GPU Operator on Red Hat OpenShift Container Platform for OpenShift install, upgrade, and disconnected deployment procedures.

GPU Resource Management#

The GPU Operator supports managing GPU resources on your cluster using the following methods:

NVIDIA Device Plugin through the ClusterPolicy custom resource.
DRA Driver for NVIDIA GPUs through the GPUCluster custom resource.

At install time, choose one GPU resource management model for the cluster. Use either ClusterPolicy or GPUCluster; using both in the same cluster is not supported.

Note

Deploying the DRA model through the GPUCluster custom resource is in Technology Preview and is supported only for greenfield (new) deployments. The GPUCluster API is served under nvidia.com/v1alpha1 and is subject to change in future releases. Migrating an existing ClusterPolicy deployment to GPUCluster in place is not supported.

Component	Device-plugin model (`ClusterPolicy`)	DRA model (`GPUCluster`)
GPU allocation	NVIDIA Kubernetes Device Plugin (extended resources)	DRA Driver for NVIDIA GPUs (ResourceClaims)
NVIDIA GPU driver	Managed by `ClusterPolicy` or `NVIDIADriver`	Pre-installed or managed by `NVIDIADriver` (not managed by `GPUCluster`)
Default install path	Install with Helm	Deploying the GPU Operator with DRA Support

Do not deploy ClusterPolicy and GPUCluster as GPU resource management models in the same cluster. Refer to Deploying the GPU Operator with DRA Support for the supported greenfield DRA deployment.

For the fully supported standalone DRA Driver Helm chart (without Operator-managed GPUCluster), refer to DRA Driver for NVIDIA GPUs.

Driver Provisioning#

How the NVIDIA GPU driver is installed is independent of which platform or GPU resource management model you choose.

Model	When to use	Where to configure
Operator-managed (`ClusterPolicy`)	Default for generic Kubernetes and most cloud installs	Install with Helm and Common Deployment Scenarios
`NVIDIADriver` custom resource	Per-node or mixed-OS driver management; use this for Operator-managed drivers with the DRA model	GPU Driver CRD
Pre-installed on the host	Driver already present on GPU nodes (`driver.enabled=false`)	Install with Helm, Microsoft AKS, or Deploying the GPU Operator with DRA Support

NVIDIA Licenses#

If your deployment requires a licensed NVIDIA product or a sovereign-cloud configuration, use one of these guides instead of the default Helm install.

NVIDIA AI Enterprise

Install with NVAIE licensing using the NGC Bash script (vGPU guest) or standard Helm (datacenter driver).

NVAIE

NVIDIA AI Enterprise

NVIDIA vGPU

Build a custom driver image and install with vGPU licensing secrets.

vGPU

Using NVIDIA vGPU

Government Ready Components

Deploy STIG- and FIPS-hardened containers for sovereign and FedRAMP environments. Requires NVAIE licensing.

gov-ready

NVIDIA GPU Operator Government Ready

Platform#

Select the guide for your Kubernetes distribution or cloud platform.

Upstream Kubernetes or Bare Metal

Install with Helm on upstream Kubernetes, bare-metal hosts, or VMs with GPU passthrough. This is the default path for most self-managed clusters.

Helm default

Installing the NVIDIA GPU Operator

Amazon EKS

Prepare Ubuntu node groups and choose between the default EKS GPU stack and Operator-managed drivers.

EKS

NVIDIA GPU Operator with Amazon EKS

Azure AKS

Compare AKS GPU options and install the Operator with --skip-gpu-driver-install or pre-installed drivers.

AKS

NVIDIA GPU Operator with Azure Kubernetes Service

Google GKE

Choose between the Google driver installer plus the Operator, or full Operator driver management on Ubuntu nodes.

GKE

NVIDIA GPU Operator with Google GKE

Red Hat OpenShift

Install through OperatorHub or oc using the OpenShift documentation book.

OLM OpenShift

https://docs.nvidia.com/datacenter/cloud-native/openshift/latest/index.html

Partner Validated Platforms

Validated configurations for partner Kubernetes distributions such as Mirantis MKE and SUSE RKE2.

partner

https://docs.nvidia.com/datacenter/cloud-native/partner-validated/latest/index.html

Google Cloud Anthos

Deploy on Anthos clusters on bare metal or VMware vSphere.

Anthos

https://docs.nvidia.com/datacenter/cloud-native/edge/latest/anthos-guide.html

Specialized Network Environments#

These guides apply in addition to your primary install path when the cluster has network or mesh constraints.

HTTP Proxy

Configure the driver container to reach external endpoints through a proxy.

proxy

Install GPU Operator in Proxy Environments

Air-Gapped Network

Mirror images and package repositories for disconnected clusters.

air-gap

Install NVIDIA GPU Operator in Air-Gapped Environments

Service Mesh

Disable sidecar injection on the gpu-operator namespace before installing.

Istio Linkerd

Install GPU Operator with Service Mesh

Suggested Paths#

If you are new to the GPU Operator, follow one of these common sequences.

flowchart LR subgraph bareMetal["Bare Metal / Generic K8s"] P1[Prerequisites] --> H1[Install with Helm] end subgraph cloud["Cloud Managed K8s"] P2[Prerequisites] --> C1[Platform Guide] C1 --> H2[Install with Helm] end subgraph ocp["OpenShift"] O1[OpenShift Book] end subgraph dra["DRA Greenfield"] P3[Prerequisites] --> D1[DRA Install Guide] end H1 --> V[Verify Installation] H2 --> V O1 --> V D1 --> V

Common journeys:

Bare metal or generic Kubernetes: Prerequisites → Install with Helm → verify
Amazon EKS, Azure AKS, or Google GKE: Prerequisites → platform guide → Install with Helm → verify
Red Hat OpenShift: NVIDIA GPU Operator on Red Hat OpenShift Container Platform (OperatorHub or oc install)
DRA-native cluster (Technology Preview): Prerequisites → Deploying the GPU Operator with DRA Support → verify

If your cluster uses a restricted network, add the matching guide from Specialized Network Environments before or during install.

Next Steps#

After you choose a path:

Review Prerequisites for your platform, license, and GPU resource management model.
Follow the install guide for your chosen path.
Verify the installation using the steps in that guide.
Configure workloads (MIG, time-slicing, RDMA, and others) from the Advanced Operator Configuration section in the sidebar.