NVIDIA is set to begin implementing open kernel modules for GPUs, beginning with Turing architecture

NVIDIA Company announcedthat the release of proprietary NVIDIA 560 drivers will be used by default to support Turing-based GPUs (GeForce RTX 2000) and newer microarchitectures open modules for the Linux kernel. The module code was open sourced in 2022 under the MIT and GPLv2 licenses, and since then updated synchronized with each new release of proprietary drivers.

At the same time, despite the presence of open modules, when installing proprietary NVIDIA drivers, proprietary versions of modules continued to be used, based on a common code base with open ones, but different and developed separately. Open source and proprietary modules could not be used at the same time and could not be installed on the file system at the same time. Open modules are limited to supporting only GPUs equipped with a separate GSP (GPU System Processor) microcontroller, used starting in 2018 in GPU-based Turing, Ampere, Ada and newer microarchitectures. In addition to new GPUs, proprietary modules continue to support older GPUs that are not equipped with GSP, for example, GPUs based on Maxwell, Pascal and Volta microarchitectures.

Starting with the release of NVIDIA 560, the situation will change – for regular GPUs starting with Turing, and for GPU virtualization starting with Ada, open versions of kernel modules nvidia.ko, nvidia-modeset.ko, nvidia-uvm.ko, nvidia- will be installed by default drm.ko and nvidia-peermem.ko, in situations where their use is possible. To install proprietary modules on the system, you will need to specify the “–kernel-module-type=proprietary” option when running the run-archive with drivers. In the future, NVIDIA plans to stop implementing support for new GPUs in proprietary modules and will focus only on the development of open modules.

Thanks for reading: