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.
- 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
- CentOS 7
- RedHat 7
-
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/hoststo match the target server -
edit the definition file for the role
group_vars/docker-hostto deactivate/activate Nvidia-GPU installation steps
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!
Log onto the target system and try to start an example Nvidia CUDA container: sudo docker run --rm nvidia/cuda nvidia-smi
- MIT license (check file
LICENSEfor more details)
Please use the Github Repository Issue tracker to report us bugs.
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.
- Markus Rekkenbeil - (Datadrivers GmbH)