Ansible Playbook Nvidia GPU Docker Demo

This playbook is to show off an easy method to install Docker with Nvidia GPU support on a supported linux host. The playbook can run against a bare metal machine, virtual machine or cloud based machine. Datadrivers GmbH is prefering to use it with an AWS EC2 instance from the P2-, P3- or G3-series.

Requirements

• Ansible 2.1+
• SSH access and sudo permissions (login user!)
• target host with installed linux OS with systemd
• for nvidia-docker usage: a host with a compatible Nvidia graphic card
• Internet access to get the packages and gpg keys

Supported platforms

• CentOS 7
• RedHat 7

Usage

Preparations

• open a terminal / console

• git clone the repository on your system and switch into the directory

• use ansible-galaxy to install the required role: ansible-galaxy install -v -r requirements.yml

• edit the inventory file inventory/hosts to match the target server

• edit the definition file for the role group_vars/docker-host to deactivate/activate Nvidia-GPU installation steps

Run the playbook

Please use instead of user centos, the right user to get SSH access and sudo permissions on the target system.

• start the playbook run: ansible-playbook demo.yml --user centos --diff -v

• If you install the Nvidia-docker2 parts, please reboot the target system after the completion!

Test the Docker environment

Log onto the target system and try to start an example Nvidia CUDA container: sudo docker run --rm nvidia/cuda nvidia-smi

License

• MIT license (check file LICENSE for more details)

Issue Tracker

Please use the Github Repository Issue tracker to report us bugs.

Contributions

Feel free to clone the code, make a branch and to hack your stuff. After that make a pull request and we can check it. We are open for improvements.

Author Information

• Markus Rekkenbeil - (Datadrivers GmbH)