This code has migrated into the RHOSO EDPM Ansible repo, here: https://github.com/openstack-k8s-operators/edpm-ansible/tree/main/roles/edpm_accel_drivers
This repo will be archived after gpu-validation has migrated (OSPRH-22196)
Ansible roles for installing gpu drivers on RHEL 9 systems.
- Ansible 2.9 or higher
- RHEL 9 or compatible system
- Root privileges
| Variable | Default | Description |
|---|---|---|
nvidia_driver_repo_url |
"https://developer.download.nvidia.com/compute/cuda/repos/rhel9/" |
NVIDIA repository URL |
nvidia_driver_url |
"" | Install from a URL instead of a repo |
nvidia_driver_module_version |
"open-dkms" |
NVIDIA Driver module version to install |
nvidia_driver_install_cuda_drivers |
true |
Whether to install CUDA drivers (cuda-drivers package) |
nvidia_driver_install_container_tools |
true |
Whether to install container tools (and CDI config) |
nvidia_driver_container_tools_version_release |
1.17.8-1 |
Version of the container tools to install |
nvidia_driver_generate_cdi_config |
true |
Whether to generate the CDI config file |
nvidia_driver_install_management_library |
true |
Whether to install management library (NVML) |
nvidia_driver_management_library_package_name |
libnvidia-ml |
This package is called "nvidia-driver-NVML" in older versions |
nvidia_driver_blacklist_nouveau |
true |
Whether to blacklist nouveau driver |
nvidia_driver_cleanup_temp_files |
true |
Whether to cleanup temporary files after driver installation |
nvidia_driver_force_host_mode |
false |
Force host mode installation. When enabled, CUDA installation and CDI generation will be skipped |
See playbook.yaml