Quick Start

Get HAMi up and running in minutes by deploying the Helm chart and submitting your first shared GPU workload.

Prerequisites

Before deploying HAMi, ensure your GPU nodes meet the following prerequisites:

• Helm v3+
• kubectl v1.23+
• CUDA v10.2+
• NVIDIA Driver v440+
• NVIDIA Container Toolkit (with nvidia-container-runtime set as default runtime)

1. Label your nodes

Label the target GPU nodes with gpu=on. Nodes without this label will not be managed by HAMi:

kubectl label nodes <node-name> gpu=on

2. Deploy HAMi using Helm

Add the official HAMi Helm repository and deploy the chart:

helm repo add hami-charts https://project-hami.github.io/HAMi/
helm repo update
helm install hami hami-charts/hami -n kube-system

Verify that the hami-scheduler and hami-device-plugin pods are running:

kubectl get pods -n kube-system | grep hami

3. Submit a vGPU Workload

Save the following manifest as gpu-pod.yaml to request 1 vGPU with 10240 MiB of GPU memory limit:

apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: ubuntu-container
      image: ubuntu:22.04
      command: ["bash", "-c", "sleep 86400"]
      resources:
        limits:
          nvidia.com/gpu: 1
          nvidia.com/gpumem: 10240

Apply the manifest and wait for the Pod to become ready:

kubectl apply -f gpu-pod.yaml
kubectl wait --for=condition=Ready pod/gpu-pod --timeout=120s

4. Verify GPU Memory Isolation

Execute nvidia-smi inside the running container:

kubectl exec -it gpu-pod -- nvidia-smi

Expected output showing HAMi-core hard memory limit (10240MiB):

[HAMI-core Msg]: Initializing.....
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15              Driver Version: 550.54.15      CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|=========================================+========================+======================|
|   0  Tesla V100-PCIE-32GB           On  |   00000000:3E:00.0 Off |                    0 |
| N/A   29C    P0             24W /  250W |       0MiB /  10240MiB |      0%      Default |
+-----------------------------------------------------------------------------------------+

Cleanup

Delete the test Pod:

kubectl delete pod gpu-pod

Next steps

• Verify your setup in detail with Verify HAMi Installation.
• Learn how to customize your deployment parameters in the Configuration Guide.
• Understand how GPU sharing works under the hood in Device Sharing.