Run a Sample Workload#

As a Kubernetes Cluster Administrator, use this page to verify your installation and run a sample workload. Container User personas can also run the sample workload to confirm the cluster is ready before deploying applications.

Verify your Confidential Container setup by running a basic single-GPU sample workload inside a Confidential Container.

This page assumes that you have completed Prerequisites and either Quickstart Install or Detailed Install Guide. Your cluster should have kata-qemu-nvidia-gpu-snp and kata-qemu-nvidia-gpu-tdx runtime classes installed, and GPU Operator operands (including the Confidential Computing Manager, Kata Sandbox Device Plugin, and VFIO Manager) running on your nodes.

This page intentionally uses the simplest possible manifest so that you can confirm the deployment end-to-end. It is not a production workload template. For runtime class selection, resource type naming, multi-GPU passthrough, and additional manifest patterns, refer to Configuring Workloads.

  1. Create a file named cuda-vectoradd-kata.yaml with a sample manifest for your system:

    apiVersion: v1
kind: Pod
metadata:
  name: cuda-vectoradd-kata
  namespace: default
spec:
  runtimeClassName: kata-qemu-nvidia-gpu-snp
  restartPolicy: Never
  containers:
    - name: cuda-vectoradd
      image: "nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0-ubuntu22.04"
      resources:
        limits:
          nvidia.com/pgpu: "1" # for single GPU passthrough
          memory: 16Gi

    apiVersion: v1
kind: Pod
metadata:
  name: cuda-vectoradd-kata
  namespace: default
spec:
  runtimeClassName: kata-qemu-nvidia-gpu-tdx
  restartPolicy: Never
  containers:
    - name: cuda-vectoradd
      image: "nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0-ubuntu22.04"
      resources:
        limits:
          nvidia.com/pgpu: "1" # for single GPU passthrough
          memory: 16Gi

    The following is a brief list of the options available for the manifest:

    • Runtime class: Use kata-qemu-nvidia-gpu-snp on AMD-based systems or kata-qemu-nvidia-gpu-tdx on Intel-based systems.

    • GPU resource type: The sample requests nvidia.com/pgpu, which is the default resource name advertised by the NVIDIA Kata Sandbox Device Plugin. If your cluster was installed with the P_GPU_ALIAS="" setting, replace it with the model-specific name advertised on your node, for example nvidia.com/GH100_H200_141GB.

    Refer to Configuring Confidential Container Workloads for additional guidance on each option.

  2. Create the pod:

    $ kubectl apply -f cuda-vectoradd-kata.yaml

    Example Output:

    pod/cuda-vectoradd-kata created

  3. Verify the pod is running:

    $ kubectl get pod cuda-vectoradd-kata

    Example Output:

    NAME                  READY   STATUS    RESTARTS   AGE
cuda-vectoradd-kata   1/1     Running   0          10s

    The pod could also say Completed if the container already completed successfully.

    If the pod stays Pending for more than a few minutes, use the Troubleshooting guide for more details on viewing logs and potential causes for the pod being stuck.

  4. View the logs from the pod after the container starts:

    $ kubectl logs -n default cuda-vectoradd-kata

    Example Output:

    [Vector addition of 50000 elements]
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Copy output data from the CUDA device to the host memory
Test PASSED
Done

    The output should include Test PASSED if the container completed successfully. This means that your cluster is configured to deploy GPU workloads in the Kata container runtime on nodes configured for Confidential Computing.

    If you do not see any log output, the pod may still be running. Use the command in the previous step to check the pod status.

  5. Optionally, you can verify the sample app was running the Kata container runtime by checking the pod details:

    $ kubectl describe pod cuda-vectoradd-kata

    Example Output:

    Name:                cuda-vectoradd-kata
Namespace:           default
Priority:            0
Runtime Class Name:  kata-qemu-nvidia-gpu-tdx

...

Events:
Type    Reason     Age   From               Message
----    ------     ----  ----               -------
Normal  Scheduled  95s   default-scheduler  Successfully assigned default/cuda-vectoradd-kata to dellr760tdx
Normal  Pulled     71s   kubelet            Container image "nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0-ubuntu22.04" already present on machine
Normal  Created    71s   kubelet            Created container: cuda-vectoradd
Normal  Started    68s   kubelet            Started container cuda-vectoradd

    The Runtime Class Name should match the runtime class used to create the pod and you will see that the pod ran successfully. The example output shows Runtime Class Name:  kata-qemu-nvidia-gpu-tdx.

To clean up, delete the pod:

$ kubectl delete -f cuda-vectoradd-kata.yaml

Next Steps#

Note

You have verified that a GPU workload runs on the confidential runtime. This is the end of the deployment path covered by this guide.

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.

If you’d like to continue to learn about Confidential Container configuraion, review the following pages:

On this page