Quickstart Install#

As a Kubernetes Cluster Administrator, use these steps to install Kata Containers and the NVIDIA GPU Operator with minimal steps. For additional configuration options and install details, refer to the Detailed Install Guide.

This quickstart will configure every node in your cluster for Confidential Containers workloads. This is ideal for evaluation or dedicated Confidential Containers clusters. If you need to run Confidential Containers on only some nodes while keeping traditional GPU workloads on others, or you want more control over the installation, use the Label Nodes for Confidential Containers Components section in the Detailed Install Guide.

This quickstart takes approximately 10 minutes to complete.

Note

Before starting, make sure your cluster meets the Prerequisites.

What You Will Build#

By the end of this quickstart, you will have:

  • Kata Containers running on your cluster.

  • The NVIDIA GPU Operator installed and configured for Confidential Containers.

  • All cluster nodes configured for Confidential Containers workloads.

Install the Kata Containers Helm Chart#

  1. Set the chart version and registry path:

    $ export VERSION="4.0.0"
$ export CHART="oci://ghcr.io/kata-containers/kata-deploy-charts/kata-deploy"

  2. Create a values file, such as kata-nvidia-gpu-values.yaml, to configure the kata-deploy chart for NVIDIA Confidential Containers:

    # Example values file to enable NVIDIA GPU shims for the NVIDIA
# Confidential Containers Reference Architecture.

# Set to true for verbose kata-deploy and Kata runtime logging.
debug: false

deploymentMode: daemonset

# Disable Node Feature Discovery (NFD) deployment by kata-deploy.
# Both kata-deploy and the GPU Operator deploy NFD by default. This
# reference architecture relies on the NFD instance that the GPU Operator
# deploys and manages, so the kata-deploy NFD is turned off to avoid a
# duplicate, conflicting deployment.
nfd:
  enabled: false

# Install the nydus snapshotter on each node alongside containerd.
# The confidential -snp and -tdx shims below use nydus to pull container
# images directly into the confidential VM (guest pull), which keeps image
# contents inside the trusted execution environment (TEE).
snapshotter:
  setup: ["nydus"]

# Disable the chart's default hypervisor/TEE shims and opt in only to
# the NVIDIA GPU shims supported by this reference architecture.
shims:
  disableAll: true

  # Non-confidential NVIDIA GPU passthrough shim used when Confidential
  # Computing mode is off on the node. The runtime class is restricted to
  # nodes where the GPU Operator's Confidential Computing Manager has
  # reported nvidia.com/cc.ready.state=false, so it will not schedule on
  # CC-ready nodes. The empty containerd snapshotter falls back to the
  # default (overlayfs); guest pull is not used for this non-confidential
  # path.
  qemu-nvidia-gpu:
    enabled: true
    supportedArches:
      - amd64
    allowedHypervisorAnnotations: []
    containerd:
      snapshotter: ""
    runtimeClass:
      # This label is automatically added by the GPU Operator.
      nodeSelector:
        nvidia.com/cc.ready.state: "false"

  # Confidential NVIDIA GPU passthrough for AMD SEV-SNP nodes.
  # Scheduled where the GPU Operator reports CC mode is on AND NFD
  # reports SEV-SNP support. Set agent.httpsProxy / agent.noProxy if
  # the guest needs a proxy to reach the registry.
  qemu-nvidia-gpu-snp:
    enabled: true
    supportedArches:
      - amd64
    allowedHypervisorAnnotations: []
    containerd:
      snapshotter: "nydus"
      forceGuestPull: false
    crio:
      guestPull: true
    agent:
      httpsProxy: ""
      noProxy: ""
    runtimeClass:
      # These labels are automatically added by the GPU Operator and NFD
      # respectively.
      nodeSelector:
        nvidia.com/cc.ready.state: "true"
        amd.feature.node.kubernetes.io/snp: "true"

  # Confidential NVIDIA GPU passthrough for Intel TDX nodes.
  # Same selectors and snapshotter behavior as the SNP shim above,
  # but pinned to TDX-capable hosts.
  qemu-nvidia-gpu-tdx:
    enabled: true
    supportedArches:
      - amd64
    allowedHypervisorAnnotations: []
    containerd:
      snapshotter: "nydus"
      forceGuestPull: false
    crio:
      guestPull: true
    agent:
      httpsProxy: ""
      noProxy: ""
    runtimeClass:
      # These labels are automatically added by the GPU Operator and NFD
      # respectively.
      nodeSelector:
        nvidia.com/cc.ready.state: "true"
        intel.feature.node.kubernetes.io/tdx: "true"

# Default shim when a pod does not request a runtime class. Set to the
# non-confidential shim so pods only run in a confidential VM when
# they explicitly request the -snp or -tdx runtime class.
defaultShim:
  amd64: qemu-nvidia-gpu # Can be changed to qemu-nvidia-gpu-snp or qemu-nvidia-gpu-tdx if preferred

# Create one Kubernetes RuntimeClass per enabled shim above
# (kata-qemu-nvidia-gpu, kata-qemu-nvidia-gpu-snp, kata-qemu-nvidia-gpu-tdx).
# createDefault: false suppresses the generic "kata" RuntimeClass since
# you should always reference a specific NVIDIA shim
# by name in pod specs.
runtimeClasses:
  enabled: true
  createDefault: false
  defaultName: "kata"

  3. Install the kata-deploy Helm chart with the values file:

    $ helm install kata-deploy "${CHART}" \
   --namespace kata-system --create-namespace \
   -f kata-nvidia-gpu-values.yaml \
   --wait --timeout 10m \
   --version "${VERSION}"

    Example Output:

    Pulled: ghcr.io/kata-containers/kata-deploy-charts/kata-deploy:4.0.0
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

    It can take 2 to 3 minutes for the command to return and all output to be printed.

    Note

    There is a known Helm issue on single-node clusters that may result in the Helm command finishing before all pods are done initializing. If you are deploying to a single-node cluster, wait a few additional minutes after the command completes.

  4. 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. If the pod does not reach Running, refer to View Kata Containers Logs in Troubleshooting.

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

    After helm install completes, the kata-deploy chart creates the Kata RuntimeClass resources on the cluster. This usually happens within 1-2 minutes after the kata-deploy pod reaches Running. Confirm SNP and TDX classes are present before you continue to Install the NVIDIA GPU Operator.

    $ 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

    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. 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 Troubleshooting.

Success criteria: Helm reports STATUS: deployed and both SNP and TDX runtime classes appear in the output above.

Install the NVIDIA GPU Operator#

  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 configured for Confidential Containers on all nodes:

    $ helm install --wait --timeout 10m --generate-name \
   -n gpu-operator --create-namespace \
   nvidia/gpu-operator \
   --set sandboxWorkloads.enabled=true \
   --set sandboxWorkloads.defaultWorkload=vm-passthrough \
   --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

    It may take 3 to 5 minutes for the Helm release to complete and for all GPU Operator pods to reach the Running state.

    The sandboxWorkloads.defaultWorkload=vm-passthrough flag in the Helm install command sets the default cluster workload type for Confidential Containers.

  3. Verify that all GPU Operator pods 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

Success criteria: All GPU Operator pods are Running or Completed.

Your cluster is now configured to deploy GPU workloads into the SNP and TDX runtime classes on all nodes.

Next Steps#