https://github.com/andrewtwin/ansible-role-install_nvidia_cuda

Ansible role for installing Nvidia drivers and CUDA toolkit
https://github.com/andrewtwin/ansible-role-install_nvidia_cuda

ansible-role cuda-toolkit drivers nvidia

Last synced: 12 days ago
JSON representation

Ansible role for installing Nvidia drivers and CUDA toolkit

• Host: GitHub
• URL: https://github.com/andrewtwin/ansible-role-install_nvidia_cuda
• Owner: andrewtwin
• Created: 2025-01-15T06:40:04.000Z (20 days ago)
• Default Branch: main
• Last Pushed: 2025-01-17T06:33:08.000Z (18 days ago)
• Last Synced: 2025-01-17T07:29:20.299Z (18 days ago)
• Topics: ansible-role, cuda-toolkit, drivers, nvidia
• Language: Jinja
• Homepage:
• Size: 20.5 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
install-nvidia-cuda

=========

Install pre-compiled or dkms Nvidia drivers and cuda-toolkit on:  

- RHEL8  

- RHEL9  

- Ubuntu 22.04  

- Ubuntu 24.04  

- Debian 12  

Other distros in these families may also work.  

  

Based on the instructions from:  

- https://docs.nvidia.com/datacenter/tesla/driver-installation-guide/  

- https://docs.nvidia.com/cuda/cuda-installation-guide-linux/  

Requirements

------------

See Nvidia docs for system requirements.  

Drivers and tool-kit can take up 10Gb+ on some systems.  

Role Variables

--------------

- `driver_version` - Default `propriety`.  

  `open` for the latest open drivers.  

  `propriety` for the latest propriety drivers.  

   Or another value from https://docs.nvidia.com/datacenter/tesla/driver-installation-guide/index.html#precompiled-streams.  

   Open drivers require a 'Turing' (RTX 20xx, 16xx, etc.) or newer Nvidia GPU.  

- `allow_autoremove` - Default `false`. If autoremove should run when packages are installed or removed.  

- `driver_packages` - Default `true`. Install (or remove) Nvidia driver packages.  

- `cuda_packages` - Default `true`. Install (or remove) the CUDA toolkit packages.  

Dependencies

------------

- `community.general.dnf_config_manager` module  

Example Playbook

----------------

Can test the ouput of `lspci` to only run on nodes with Nvidia GPU hardware.

The propriety, pre-compiled drivers should work with older, pre-Turing, GPUs if required.  

```

---

- name: Install CUDA on nodes with GPUs

  hosts: all

  become: true

  gather_facts: true

  pre_tasks:

  - name: Check which nodes have NVIDIA GPUs

    register: lspci_result

    changed_when: false

    check_mode: false

    ansible.builtin.command:

      cmd: "lspci"

  tasks:

  - name: Install CUDA

    vars:

      driver_version: propriety

    when: lspci_result['stdout'] | regex_search('vga.*nvidia', ignorecase=true, multiline=false)

    ansible.builtin.include_role:

      name: install_nvidia_cuda

```

License

-------

BSD-3-Clause