Skip to content

bounverif/ansible-role-cuda-install

Repository files navigation

Ansible Role: CUDA Install

This role installs CUDA packages from the official Nvidia CUDA repos.

Installation

Ansible Galaxy is the recommended way to install this role in your environment.

Using Ansible Galaxy CLI

# Install the latest version from the repository
ansible-galaxy role install git+https://github.com/bounverif/ansible-role-cuda-install.git,main,bounverif.cuda_install

Using Ansible Galaxy requirements file

# requirements.yml

roles:
  # Install the latest version from GitHub
  - name: bounverif.cuda_install
    src: https://github.com/bounverif/ansible-role-cuda-install
    version: main

  # # Install the latest version from Galaxy Hub (not supported yet)
  # - name: bounverif.cuda_install
ansible-galaxy install -r requirements.yml

Using other methods

See Ansible Galaxy User Guide for more options to install Ansible roles.

Usage

You can use this role to install a CUDA package on a supported Linux distro (cuda_install_distro) on a supported CUDA architecture (cuda_install_nvarch). By default the role will try to setup the correct CUDA repository by gathering your distribution and architecture facts. Then the role will install all listed packages (cuda_install_packages) using the package manager.

---
- name: My CUDA environment setup
  hosts: all
  gather_facts: true
  tasks:
    - name: Install CUDA Packages (Debian)
      ansible.builtin.include_role:
        name: bounverif.cuda_install
      vars:
        cuda_install_packages:
          - cuda-nvcc-12-4

You can owerwrite cuda_install_nvarch and cuda_install_distro variables to directly control the repo setup as follows:

---
- name: My CUDA environment setup
  hosts: all
  gather_facts: true
  tasks:
    - name: Install CUDA Packages (Debian-Arm)
      ansible.builtin.include_role:
        name: bounverif.cuda_install
      vars:
        cuda_install_nvarch: sbsa # Arm server architecture (default: x86_64)
        cuda_install_distro: debian12 # Deduced automatically by default
        cuda_install_packages:
          - libnvpl-sparse-devel

Note that every package is not available for every architecture. Besides package names and versioning schemes can be different from one distribution to another. Please refer to the official repo index for CUDA architectures, distributions, and available packages.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published