Detailed Install Guide

As a Kubernetes Cluster Administrator, use these steps to deploy Kata Containers and the NVIDIA GPU Operator to your cluster and configure it for Confidential Containers.

If you want the fastest path and intend to run Confidential Containers on every node in your cluster, use the Quickstart Install instead. Use this guide when you need per-node control, such as running Confidential Containers on some nodes and traditional GPU workloads on others, or when you want to use additional configuration options when installing the GPU Operator.

Install Overview

This guide assumes you completed Prerequisites on an existing Kubernetes cluster with GPU worker nodes.

Install workflow:

1. Prerequisites: prepare worker hosts and cluster software.

2. Label nodes to deploy Confidential Containers components: select GPU workers for Confidential Containers workloads.

3. Install Kata Containers: install runtime classes and node-level Kata components.

4. Install the NVIDIA GPU Operator: deploy Confidential Containers operands on target nodes.

Success criteria: Helm releases report STATUS: deployed, the kata-deploy pod is Running, SNP and TDX runtime classes are available, and all GPU Operator operands are healthy on target nodes.

After completing the installation, you can Run a Sample Workload to verify the deployment.

Label Nodes for Confidential Containers Components

The GPU Operator reads labels to determine what software components to deploy to a node. To configure a node for Confidential Container workloads, you label the node with the nvidia.com/gpu.workload.config=vm-passthrough label. Then, when the GPU Operator is installed in a subsequent step, it will deploy the software components needed to run Confidential Containers to the node.

A node can only run one container runtime at a time, so a node configured for Confidential Container workloads cannot run traditional GPU container workloads. The labeling approach is useful if you want to run Confidential Containers workloads on some nodes and traditional GPU container workloads on other nodes in your cluster.

For more details on how the GPU Operator deploys components to your cluster, refer to the GPU Operator Cluster Topology Considerations section in the architecture overview.

Tip

Skip this section if you plan to use all nodes in your cluster to run Confidential Containers and instead set sandboxWorkloads.defaultWorkload=vm-passthrough when installing the GPU Operator.

1. Get a list of the nodes in your cluster:

   $ kubectl get nodes
   

   Example Output:

   NAME          STATUS   ROLES           AGE   VERSION
   node-01       Ready    <none>          10d   v1.34.0
   node-02       Ready    <none>          10d   v1.34.0
   

   Identify the GPU worker node or nodes you want to configure for Confidential Containers and use its name in the next step.

2. Set the NODE_NAME environment variable to the name of the node you want to configure:

   $ export NODE_NAME="<node-name>"
   

   Note

   Commands in this guide use the $NODE_NAME environment variable to reference this node.

3. Label the node for Confidential Containers:

   $ kubectl label node $NODE_NAME nvidia.com/gpu.workload.config=vm-passthrough
   

   Example Output:

   node/<node-name> labeled
   

   If you see <node-name> not labeled, the label may already be set. Continue to the next step to verify if the label was added.

   Note

   To label multiple nodes at once, pass additional node names to the same command:

   $ kubectl label node <node-01> <node-02> nvidia.com/gpu.workload.config=vm-passthrough
   

4. Verify all target nodes have the label by listing all nodes with the label:

   $ kubectl get nodes -l nvidia.com/gpu.workload.config=vm-passthrough
   

   Example Output:

   NAME      STATUS   ROLES    AGE   VERSION
   node-01   Ready    <none>   10d   v1.34.0
   

Success criteria: All nodes you intend to use for Confidential Container workloads appear in the output above.

After all your desired nodes are labeled, you can continue to the next step to install Kata Containers.

Install the Kata Containers Helm Chart

Install Kata Containers using the kata-deploy Helm chart. The kata-deploy chart installs all required components from the Kata Containers project including the Kata Containers runtime binary, runtime configuration, UVM kernel, and images that NVIDIA uses for Confidential Containers and native Kata containers.

The minimum required version is 3.29.0.

1. Set the chart version and registry path:

   $ export VERSION="3.29.0"
   $ export CHART="oci://ghcr.io/kata-containers/kata-deploy-charts/kata-deploy"
   

2. Install the kata-deploy Helm chart:

   $ helm install kata-deploy "${CHART}" \
      --namespace kata-system --create-namespace \
      --set nfd.enabled=false \
      --wait --timeout 10m \
      --version "${VERSION}"
   

   Example Output immediately after running the command:

   Pulled: ghcr.io/kata-containers/kata-deploy-charts/kata-deploy:3.29.0
   Digest: sha256:aea41018779716ce2e0bf406d701637d10fb5a0792db51a08dfd3f76701eb933
   

   The --wait flag in the install command instructs Helm to wait until the release is deployed before returning. It can take a 2-3 minutes to return more output.

   Example Output when the release is deployed:

   Pulled: ghcr.io/kata-containers/kata-deploy-charts/kata-deploy:3.29.0
   Digest: sha256:aea41018779716ce2e0bf406d701637d10fb5a0792db51a08dfd3f76701eb933
   LAST DEPLOYED: Wed Apr  1 17:03:00 2026
   NAMESPACE: kata-system
   STATUS: deployed
   REVISION: 1
   DESCRIPTION: Install complete
   TEST SUITE: None
   

   Note

   There is a known Helm issue on single node clusters, that may result in the Helm command finishing before all deployed pods are finished initializing. If you are deploying to a single node cluster, you may need to wait for an additional few minutes after the Helm command completes for the kata-deploy pod to be in the Running state.

   STATUS: deployed confirms the Helm release succeeded and the chart resources were applied. The following steps confirm the Kata components are healthy.

   Note

   Both kata-deploy and the GPU Operator deploy Node Feature Discovery (NFD) by default. The install command includes --set nfd.enabled=false to prevent kata-deploy from deploying NFD. The GPU Operator will deploy and manage NFD in the next step.

3. Verify that the kata-deploy pod is running:

   $ kubectl get pods -n kata-system | grep kata-deploy
   

   Example Output:

   kata-deploy-b2lzs       1/1     Running   0             6m37s
   

   A READY value of 1/1 and a STATUS of Running mean the kata-deploy pod installed the Kata components on the node successfully. If the pod is Pending, ContainerCreating, or CrashLoopBackOff, wait a minute and re-run the command. If the kata-deploy pod does not reach Running, refer to View Kata Containers Logs in Troubleshooting.

4. Verify that the kata-qemu-nvidia-gpu-snp and kata-qemu-nvidia-gpu-tdx runtime classes are available:

   $ kubectl get runtimeclass | grep kata-qemu-nvidia-gpu
   

   Example Output:

   NAME                       HANDLER                    AGE
   kata-qemu-nvidia-gpu       kata-qemu-nvidia-gpu       40s
   kata-qemu-nvidia-gpu-snp   kata-qemu-nvidia-gpu-snp   40s
   kata-qemu-nvidia-gpu-tdx   kata-qemu-nvidia-gpu-tdx   40s
   

   Several runtimes are installed by the kata-deploy chart. The kata-qemu-nvidia-gpu runtime class is used with Kata Containers, in a non-Confidential Containers scenario. The kata-qemu-nvidia-gpu-snp for AMD-based systems or kata-qemu-nvidia-gpu-tdx for Intel-based systems runtime classes are used to deploy Confidential Containers workloads.

   The kata-deploy chart typically creates these runtime classes within 1-2 minutes after the kata-deploy pod reaches Running. If the SNP and TDX runtime classes are not listed immediately, the chart may still be initializing rather than failing. Wait 1-2 minutes and re-run the command. If they are still missing after the kata-deploy pod reports Running, the install did not complete correctly. Refer to View Kata Containers Logs in Troubleshooting for help diagnosing the issue. On a single-node cluster, retry after a few minutes only if Helm returned before the kata-deploy pod reaches Running (refer to the note above). Otherwise, refer to the log steps below.

Success criteria: Helm reports STATUS: deployed, the kata-deploy pod is Running, and both kata-qemu-nvidia-gpu-snp and kata-qemu-nvidia-gpu-tdx are available on the cluster. After all checks pass, continue to Install the NVIDIA GPU Operator.

If the kata-deploy pod does not reach Running or the SNP and TDX runtime classes are missing, refer to View Kata Containers Logs in Troubleshooting.

Install the NVIDIA GPU Operator

Install the NVIDIA GPU Operator and configure it to deploy Confidential Container components. For more details on each of the GPU Operator components, refer to the GPU Operator Cluster Topology Considerations section in the architecture overview.

1. Add and update the NVIDIA Helm repository:

   $ helm repo add nvidia https://helm.ngc.nvidia.com/nvidia \
      && helm repo update
   

   Example Output:

   "nvidia" has been added to your repositories
   Hang tight while we grab the latest from your chart repositories...
   ...Successfully got an update from the "nvidia" chart repository
   Update Complete. ⎈Happy Helming!⎈
   

2. Install the GPU Operator with the following configuration:

   Tip

   Add --set sandboxWorkloads.defaultWorkload=vm-passthrough to configure every worker node for Confidential Containers workloads. Refer to the Label Nodes for Confidential Containers Components section for more details on this use case.

   To customize the installation beyond the command below:

   • Refer to the Common GPU Operator Configuration Settings section on this page for the Confidential Containers-specific configuration options you can specify when installing the GPU Operator.

   • Refer to the Common chart customization options in Installing the NVIDIA GPU Operator for the additional general configuration options you can specify when installing the GPU Operator.

   $ helm install --wait --timeout 10m --generate-name \
      -n gpu-operator --create-namespace \
      nvidia/gpu-operator \
      --set sandboxWorkloads.enabled=true \
      --set sandboxWorkloads.mode=kata \
      --set nfd.enabled=true \
      --set nfd.nodefeaturerules=true \
      --version=v26.3.1
   

   Example Output:

   NAME: gpu-operator
   LAST DEPLOYED: Tue Mar 10 17:58:12 2026
   NAMESPACE: gpu-operator
   STATUS: deployed
   REVISION: 1
   TEST SUITE: None
   

   STATUS: deployed confirms the Helm release succeeded. The --wait flag instructs Helm to wait until the release is deployed before returning. It may take 3-5 minutes for the Helm command to complete.

   Use the following steps to confirm the GPU Operator components are deployed and configured correctly.

3. Verify that all GPU Operator pods, especially the Confidential Computing Manager, Kata Device Plugin and VFIO Manager operands, are running:

   $ kubectl get pods -n gpu-operator
   

   Example Output:

   NAME                                                              READY   STATUS    RESTARTS   AGE
   gpu-operator-1766001809-node-feature-discovery-gc-75776475sxzkp   1/1     Running   0          86s
   gpu-operator-1766001809-node-feature-discovery-master-6869lxq2g   1/1     Running   0          86s
   gpu-operator-1766001809-node-feature-discovery-worker-mh4cv       1/1     Running   0          86s
   gpu-operator-f48fd66b-vtfrl                                       1/1     Running   0          86s
   nvidia-cc-manager-7z74t                                           1/1     Running   0          61s
   nvidia-kata-sandbox-device-plugin-daemonset-d5rvg                 1/1     Running   0          30s
   nvidia-sandbox-validator-6xnzc                                    1/1     Running   0          30s
   nvidia-vfio-manager-h229x                                         1/1     Running   0          62s
   

   Each pod should report a READY value of 1/1 and a STATUS of Running or Completed. The nvidia-cc-manager, nvidia-kata-sandbox-device-plugin-daemonset, and nvidia-vfio-manager operands are specific to Confidential Containers and must be present on labeled nodes. Pods may briefly show Pending or Init while they start, which is expected. When all operands are Running or Completed, the GPU Operator components are deployed.

4. If you have host access to the worker node, you can perform the following validation step to confirm that the host uses the vfio-pci device driver for GPUs:

   $ lspci -nnk -d 10de:
   

   Example Output:

   65:00.0 3D controller [0302]: NVIDIA Corporation xxxxxxx [xxx] [10de:xxxx] (rev xx)
            Subsystem: NVIDIA Corporation xxxxxxx [xxx] [10de:xxxx]
            Kernel driver in use: vfio-pci
            Kernel modules: nvidiafb, nouveau
   

   The Kernel driver in use: vfio-pci line in the output confirms the GPU is bound for VFIO passthrough into the confidential virtual machine. If the driver in use is nvidia or nouveau instead, the GPU is not ready for passthrough. Confirm your node meets the Prerequisites section, including removing any NVIDIA GPU drivers on the host.

Success criteria: All GPU Operator pods are Running or Completed. Your cluster is now configured to deploy workloads in Kata Containers. Continue to Run a Sample Workload to confirm everything is working as expected.

If you are not seeing the expected output, view the logs for the GPU Operator pods or refer to Troubleshooting.

Common GPU Operator Configuration Settings

The following are the available GPU Operator configuration settings to enable Confidential Containers:

Parameter	Description	Default
sandboxWorkloads.enabled	Enables sandbox workload management in the GPU Operator for virtual machine-style workloads and related operands.	false
sandboxWorkloads.defaultWorkload	Specifies the default type of workload for the cluster, one of container, vm-passthrough, or vm-vgpu. Set to vm-passthrough if you plan to run all or mostly virtual machines in your cluster.	container
sandboxWorkloads.mode	Specifies the sandbox mode to use when deploying sandbox workloads. Accepted values are kubevirt (default) and kata. Set to kata to run Confidential Containers workloads in Kata Containers.	kubevirt
kataSandboxDevicePlugin.env	Optional list of environment variables passed to the NVIDIA Kata Device Plugin pod. Each list item is an EnvVar object with required name and optional value fields. Use the setting to configure P_GPU_ALIAS or NVSWITCH_ALIAS for the Kata sandbox device plugin. Refer to the Configuring GPU or NVSwitch Resource Types Name section for more details.	[] (empty list)

Configuring GPU or NVSwitch Resource Types Name

By default, the NVIDIA GPU Operator creates a resource type for GPUs and NVSwitches, nvidia.com/pgpu and nvidia.com/nvswitch. You can reference this name in your manifests to request GPU or NVSwitch resources for your workload. If you want to use a different name, you can set the P_GPU_ALIAS or NVSWITCH_ALIAS environment variables in the Kata device plugin to your preferred name. In clusters where all GPUs are the same model, a single resource type is typically sufficient.

In heterogeneous clusters, where you have different GPU types on your nodes, you might want to use specific GPU types for your workload. To do this, specify an empty P_GPU_ALIAS environment variable in the Kata sandbox device plugin by adding the following to your GPU Operator installation: --set kataSandboxDevicePlugin.env[0].name=P_GPU_ALIAS and --set kataSandboxDevicePlugin.env[0].value="".

When this variable is set to "", the Kata device plugin creates GPU model-specific resource types, for example nvidia.com/GH100_H200_141GB, instead of the default nvidia.com/pgpu type. Use the exposed device resource types in pod specs by specifying respective resource limits.

Similarly, you can set NVSWITCH_ALIAS to "" to advertise model-specific NVSwitch resource types.

The following example installs the GPU Operator with both P_GPU_ALIAS and NVSWITCH_ALIAS configured:

$ helm install --wait --timeout 10m --generate-name \
     -n gpu-operator --create-namespace \
     nvidia/gpu-operator \
     --set sandboxWorkloads.enabled=true \
     --set sandboxWorkloads.mode=kata \
     --set nfd.enabled=true \
     --set nfd.nodefeaturerules=true \
     --set kataSandboxDevicePlugin.env[0].name=P_GPU_ALIAS \
     --set kataSandboxDevicePlugin.env[0].value="" \
     --set kataSandboxDevicePlugin.env[1].name=NVSWITCH_ALIAS \
     --set kataSandboxDevicePlugin.env[1].value="" \
     --version=v26.3.1

After installing the GPU Operator, you can view the GPU or NVSwitch resource types available on a node by running the following command:

$ kubectl get node $NODE_NAME -o json | grep nvidia.com

Note

The NODE_NAME environment variable was set in the Label Nodes section. If you want to view the resource types for a different node, you can update the NODE_NAME environment variable and run the command again.

Example Output:

"nvidia.com/GH100_H200_141GB": "1"

You should see the resource type information for the GPUs and NVSwitches on the node.

Next Steps

Note

You now have a working Confidential Containers runtime.

Attestation is what cryptographically verifies the TEE and releases secrets to a production workload. For attestation concepts and a local connectivity test, see the Attestation quickstart. For production attestation deployment, refer to the upstream Confidential Containers NVIDIA attestation guide.

• Run a Sample Workload to verify your deployment.

• To help manage the lifecycle of Kata Containers, install the Kata Lifecycle Manager. This Argo Workflows-based tool manages Kata Containers upgrades and day-two operations.