Summary: NVIDIA has announced a significant shift towards open-source GPU kernel modules, marking a major milestone in the company’s commitment to the Linux community. This transition aims to improve the integration of NVIDIA GPUs with the Linux operating system, enhance driver quality and security, and foster a collaborative environment for developers.
Embracing Open Source: NVIDIA’s Journey to Open-Source GPU Kernel Modules
NVIDIA’s journey to open-source GPU kernel modules began with the release of production-ready modules for data center compute GPUs in May 2022. Since then, the company has made significant strides in expanding this initiative across all platforms. With the upcoming R560 driver release, NVIDIA will transition fully to open-source GPU kernel modules, signaling a new era of collaboration and transparency.
The Benefits of Open-Source GPU Kernel Modules
The transition to open-source GPU kernel modules brings several benefits to the table:
- Improved Integration: Open-source modules allow for tighter integration with the Linux operating system, making it easier for developers to debug, integrate, and contribute to enhancing the drivers.
- Enhanced Security: By making the source code available, NVIDIA can leverage the input and reviews from the Linux end-user community to improve driver quality and security.
- New Capabilities: The open-source kernel modules introduce substantial new capabilities, including heterogeneous memory management (HMM) support, confidential computing, and the coherent memory architectures of NVIDIA’s Grace platforms.
Supported GPUs and Exceptions
While this transition is a significant step forward, it’s important to note that not all NVIDIA GPUs are supported by the open-source GPU kernel modules. Specifically:
- Supported GPUs: GPUs from the Turing, Ampere, Ada Lovelace, and Hopper architectures are recommended to switch to the open-source GPU kernel modules.
- Unsupported GPUs: Older GPUs, including those with the Maxwell, Pascal, and Volta architectures, are not supported by the open-source modules and will continue to use proprietary drivers.
Special Considerations
NVIDIA has highlighted several scenarios that require special attention during the transition:
- Package Managers: Users with the CUDA metapackage, runfile, installation helper script, and package manager details need to be aware of the changes.
- Windows Subsystem for Linux: Users of the Windows Subsystem for Linux and the CUDA Toolkit should also take note of the transition.
Community and Collaboration
NVIDIA’s commitment to open-source GPU kernel modules is not just about releasing code; it’s about fostering a collaborative environment. The company has been working closely with partners such as Canonical, Red Hat, and SUSE to ensure smooth integration and support for Linux distributions.
Building and Contributing
For developers interested in building and contributing to the open-source GPU kernel modules, NVIDIA provides detailed instructions and guidelines on its GitHub page. This includes how to build the modules, how to contribute, and how to report issues.
Future Developments
While the current open-source GPU kernel modules are a significant step forward, NVIDIA is committed to continuing this journey. The company plans to work on an upstream approach with the Linux kernel community and partners to further integrate the open-source modules into the Linux kernel.
Supported Target CPU Architectures
| Architecture | Description |
|---|---|
| x86_64 | 64-bit x86 |
| aarch64 | 64-bit ARM |
Supported Linux Kernel Versions
| Version | Description |
|---|---|
| 4.15 or newer | Supported Linux kernel versions |
How to Build
- Build Modules:
make modules -j$(nproc) - Install Modules:
make modules_install -j$(nproc)
How to Contribute
- Create a Pull Request: On https://github.com/NVIDIA/open-gpu-kernel-modules
- Report Issues: Through the GitHub repository issue tracker or existing end-user support forum.
Security Reporting
- Security Issues: Report through the channels listed on the GitHub repository security policy.
Supported GPUs
| Architecture | Description |
|---|---|
| Turing | Supported |
| Ampere | Supported |
| Ada Lovelace | Supported |
| Hopper | Supported |
| Maxwell | Not Supported |
| Pascal | Not Supported |
| Volta | Not Supported |
Example Use Cases
- Data Center GPUs: Production-ready for data center compute GPUs.
- GeForce and Workstation GPUs: Alpha-quality support for GeForce and Workstation GPUs on NVIDIA Turing and Ampere Architecture GPUs.
Partner Ecosystem
- Canonical: Simplifies installs and increases security for Ubuntu users.
- SUSE: Delivers the open-source GPU kernel modules with SUSE Linux Enterprise 15 SP4.
Future Plans
- Upstream Approach: Plans to work on an upstream approach with the Linux kernel community and partners.
- Nouveau Driver: Published source code serves as a reference to help improve the Nouveau driver.
Conclusion
NVIDIA’s transition to open-source GPU kernel modules is a significant milestone that underscores the company’s commitment to the Linux community and open-source principles. This move not only improves the integration and security of NVIDIA GPUs with the Linux operating system but also opens up new opportunities for collaboration and innovation. As NVIDIA continues on this path, it’s clear that the future of GPU computing is more open and collaborative than ever.