Question 223730 in CentOS Question posted in
The official documentation can be found here.

Centos – GPU not available in docker

I am unable to run GPU on Docker after updating the GPU driver.
When I run nvidia-smi in the host environment (Centos), the GPU is recognized.

docker run --gpus all -it -v $(pwd):/home/workspace test /bin/bash

ERRO[0000] error waiting for container: context canceled 
docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].

From what I have researched, it seems that updating the GPU driver removes the settings, so I referred to NVIDIA container toolkit site and performed the installation procedure again, but the above error is still there. I also rebooted the system just to be sure, but this did not solve the problem.
What should I do?

docker run -it --gpus all nvidia/cuda:11.4.0-base-ubuntu20.04 nvidia-smi
This also generated the above error as well.

Tags:

2

Answers


  1. 0

    Supposing you followd well installation procedure:

    Now, from Tutorial: how to attach NVIDIA GPU with docker
    https://www.howtogeek.com/devops/how-to-use-an-nvidia-gpu-with-docker-containers/

    Use the nvidia base image instead of your image test

    docker run -it --gpus all nvidia/cuda:11.4.0-base-ubuntu20.04 nvidia-smi /bin/bash

    If it’s not running well, your image test needs further debugging

    Login or Signup to reply.
  2. 0

    What made the difference for me was to follow the instructions from Docker on how to access an nvidia gpu

    1. Follow the instructions to add nvidia-container-runtime to apt
    2. apt-get install nvidia-container-runtime
    3. Ensure the nvidia-container-runtime-hook is accessible from $PATH: which nvidia-container-runtime-hook
    4. Reboot
    Login or Signup to reply.
Please signup or login to give your own answer.

Back To Top
Search