Currently the IncusOS non primary application gpu-support includes GPU firmware blobs, but does not prevent the nouveau driver from binding to NVIDIA hardware at boot after installation. While it includes the NVIDIA firmware, it lacks the kernel driver and userspace components needed for the nvidia.runtime=true config to function
Ideally gpu-support should also bundle the official NVIDIA Linux Open GPU Kernel Modules pre-compiled for IncusOS, as well as the open source container orchestration libraries (libnvidia-container, nvidia-container-toolkit) and proprietary userspace components (nvidia-smi, libcuda.so, libnvidia-ml.so, libnvcuvid.so, libnvidia-encode.so, etc.)
This feature would benefit users with NVIDIA GPUs, enabling them to use a larger range of official NVIDIA features such as CUDA compute and NVENC.
This issue was created as a follow-up to the discussion on the Linux Containers forum Nvidia drivers not enabled after installing gpu-support
Currently the IncusOS non primary application
gpu-supportincludes GPU firmware blobs, but does not prevent thenouveaudriver from binding to NVIDIA hardware at boot after installation. While it includes the NVIDIA firmware, it lacks the kernel driver and userspace components needed for thenvidia.runtime=trueconfig to functionIdeally
gpu-supportshould also bundle the official NVIDIA Linux Open GPU Kernel Modules pre-compiled for IncusOS, as well as the open source container orchestration libraries (libnvidia-container,nvidia-container-toolkit) and proprietary userspace components (nvidia-smi,libcuda.so,libnvidia-ml.so,libnvcuvid.so,libnvidia-encode.so, etc.)This feature would benefit users with NVIDIA GPUs, enabling them to use a larger range of official NVIDIA features such as
CUDA computeandNVENC.This issue was created as a follow-up to the discussion on the Linux Containers forum Nvidia drivers not enabled after installing gpu-support