隔离调用

kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: cuda-container
      image: nvidia/cuda:9.0-devel
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU
    - name: digits-container
      image: nvidia/digits:6.0
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU

GPUs 只能通过 limits 选项指定
GPUs 是严格隔离的，不同容器之间不能共享
每个容器可以请求一个或多个 GPUs
GPUs 只能正整数级请求

共享调用

实测发现，只要将resources字段去掉即可

kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: cuda-container
      image: nvidia/cuda:9.0-devel
    - name: digits-container
      image: nvidia/digits:6.0

参考:

K8S GPU支持
Docker 中玩转 GPU