Nvidia Fully Switches To Open Source Gpu Core Modules To Improve Linux Performance And Compatibility

In May 2022, NVIDIA released the R515 driver, introducing Linux GPU kernel modules under dual GPL and MIT licenses. Initially, this version targeted data center GPUs, while GeForce and workstation GPUs were in the alpha testing phase. This significant decision marked a fundamental shift in the company’s technical strategy on the Linux platform.

After two years of development, NVIDIA announced that the open-source GPU kernel modules have reached or surpassed the performance of the original drivers, with several new key features:

• Support for Heterogeneous Memory Management (HMM)
• Confidential Computing
• Consistent memory architecture for the Grace platform
• Other enhancements

Based on these advancements, NVIDIA has decided to fully transition to open-source GPU kernel modules in the upcoming R560 driver.

Supported GPU Models

However, not all GPUs are compatible with the open-source kernel modules.

• NVIDIA Grace Hopper and NVIDIA Blackwell platforms: Must use the open-source GPU kernel module as the original driver does not support these platforms.
• New GPUs with Turing, Ampere, Ada Lovelace, or Hopper architecture: NVIDIA recommends switching to the open-source kernel module.
• Older GPUs with Maxwell, Pascal, or Volta architecture: These are not compatible with the open-source kernel module; continue using the original NVIDIA drivers.
• Mixed old and new GPUs in the same system: Continue using the original driver.

If unsure about your GPU model, NVIDIA provides a new detection assistance script to help users select the appropriate driver.

Changes in Installation Methods

All installation methods will default to installing the open-source driver instead of the original driver. Note the following scenarios:

• Using package manager for CUDA metapackages: Previously, installing the CUDA Toolkit would also install the original driver. From CUDA 12.6, the open-source driver will be installed by default.
• Using runfile installation: During CUDA or NVIDIA driver installation, the installer will automatically detect the hardware and install the most suitable driver. Users can also choose between the original or open-source driver via the UI or command line.
• Using installation assistance script: If unsure which driver to choose, NVIDIA’s assistance script can help users make the selection.

Specific Installation Guides for Different Distributions

To ensure a consistent user experience, NVIDIA recommends using package managers to install CUDA toolkits and drivers. Below are the specific installation guides for various Linux distributions:

• Ubuntu and Debian-based distributions: Use the apt-get install nvidia-open command.
• Red Hat Enterprise Linux, Fedora, etc.: Use the dnf module install nvidia-driver:open-dkms command.
• SUSE Linux Enterprise Server or OpenSUSE: Use the zypper install nvidia-open command.

For Windows Subsystem for Linux (WSL) users, no changes or actions are needed as WSL uses the NVIDIA core driver from the host Windows OS.

The installation method for the CUDA toolkit remains unchanged and can be installed via the package manager using cuda-toolkit.

The Significance of Open Source Transition

NVIDIA’s full transition to open-source GPU driver modules holds multiple significances:

1. Performance Improvement: Open-source drivers may provide better performance than proprietary drivers in some cases.
2. Enhanced Transparency: Open-source code allows developers and researchers to better understand how NVIDIA hardware works.
3. Community Contribution: The open-source model may attract more developers to contribute to improvements and optimizations.
4. Better Compatibility: Open-source drivers can integrate more easily with various Linux distributions and kernel versions.
5. Security Enhancements: Open-source code can be reviewed by more people, leading to quicker identification and fixing of potential security issues.

NVIDIA’s full transition to open-source GPU driver modules is a significant milestone, reflecting the company’s commitment to the open-source community and its pursuit of technological innovation. This change will not only bring better performance and features to users but also potentially drive the development of the entire GPU ecosystem. However, users need to pay attention to choosing the correct version suitable for their hardware when upgrading or installing drivers.