In an exciting move for open-source software and hardware compatibility, NVIDIA has announced the transition of its GPU drivers for Linux to open source with the upcoming R560 driver release.
This change comes two years after the initial release of the R515 driver, which included open-source Linux GPU kernel modules initially targeting data center GPUs.
The transition, which NVIDIA describes as a critical step forward in enhancing the flexibility and security of its GPUs, includes the complete transition to open-source GPU kernel modules.
These modules now support cutting-edge platforms such as NVIDIA’s Grace Hopper and Blackwell, and are recommended for newer GPUs across the Turing, Ampere, Ada Lovelace, and Hopper architectures. They bring several advanced features to NVIDIA’s GPUs:
- Heterogeneous Memory Management: Enhances the ability to manage memory across different types of hardware.
- Confidential Computing: Adds security features that protect the integrity and confidentiality of data while in use.
- Coherent Memory Architectures: Supports the complex memory architectures of the latest NVIDIA Grace platforms.
However, the compatibility of these open-source modules varies by GPU generation. While they are necessary for the newest GPUs, owners of older Maxwell, Pascal, or Volta GPUs should continue using NVIDIA’s proprietary drivers.
For those unsure about compatibility, NVIDIA has introduced a detection helper script to guide users in selecting the appropriate system driver. Here’s how to use it.
First, install the “nvidia-driver-assistant” package using your package manager. After that, you can run the script by executing:
nvidia-driver-assistantCode language: Bash (bash)Changes in Installation Procedures
The most notable change is in the installation process. More specifically, NVIDIA aligns its driver installation with the CUDA Toolkit installations through package managers.
From CUDA Toolkit 12.6 onwards, the open-source driver will be the default installation method, reflecting a shift from the proprietary drivers that were standard up to CUDA Toolkit 12.5.
For instance, when installing the CUDA Toolkit using a package manager, users will now install the open-source NVIDIA driver alongside the toolkit, simplifying the process and ensuring compatibility with the latest software developments.
For more detailed information, visit the official announcement.
