I’m referring to this doc to work with GPU container.
https://docs.docker.com/compose/gpu-support/#access-specific-devices

If I add privileged option, nvidia-smi gives me full 8 GPUs rather than two(0 and 3).

The reason why I’m using privileged is I want to use Docker in docker (dind) feature.

How can I restrict specific GPUs inside privileged container?

services:
  test:
    image: tensorflow/tensorflow:latest-gpu
    privileged: true          # <== This point!
    deploy:
      resources:
        reservations:
          devices:
          - driver: nvidia
            device_ids: ['0', '3']
            capabilities: [gpu]