Assign task to MIG instance

This example requests two MIG instances with at least 8000 MiB each. The scheduler selects the smallest allowlisted profile that satisfies the memory request, typically 2g.10gb * 2 on A100-40GB-PCIE or 1g.10gb * 2 on A100-80GB-SXM.

apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
  annotations:
    nvidia.com/vgpu-mode: "mig"
    hami.io/gpu-scheduler-policy: "binpack" #(Optional)
spec:
  containers:
    - name: ubuntu-container
      image: ubuntu:22.04
      command: ["bash", "-c", "sleep 86400"]
      resources:
        limits:
          nvidia.com/gpu: 2
          nvidia.com/gpumem: 8000