NVIDIA to Utilize Open GPU Nuclei Modules, Starting with Turing

NVIDIA announced that in the upcoming release of the NVIDIA 560 drivers, systems with GPUs based on Turing (GeForce RTX 2000) and newer architectures will utilize Open Kernel Modules in Linux. The code for these modules was made open in 2022 under the MIT and GPLV2 licenses and is regularly updated in sync with new proprietary driver releases.

Despite the availability of open modules, when installing NVIDIA proprietary drivers, separate proprietary modules are used, based on a shared code base with the open modules but developed independently. The open and proprietary modules cannot be used simultaneously or installed in the file system together. The open modules are limited to supporting only GPUs equipped with a separate GPU System Processor (GSP), starting from 2018 in GPUs based on Turing, Ampere, Ada, and newer architectures. The proprietary modules will continue to support older GPUs without GSP, such as those based on Maxwell, Pascal, and Volta microarchitectures.

With the release of NVIDIA 560, a change will occur where open variants of the kernel modules nvidia.ko, nvidia-modeset.ko, nvidia-uvm.ko will be installed by default for conventional GPUs starting with Turing and for GPU virtualization starting with Ada, when applicable. To install proprietary modules, users will need to specify the option “-Kernel-Module-Type = Proprietary” when installing the drivers. NVIDIA plans to discontinue support for new GPUs in proprietary modules in the future and focus solely on the development of open modules.

/Reports, release notes, official announcements.