 |
| Nvidia Expands Open-Source GPU Kernel Modules Across All Platforms for Enhanced Performance and Linux Integration Image credit: Nvidia Newsroom |
Nvidia announced its plans to expand the open-source transition of its GPU kernel modules, starting with the upcoming R560 driver release. Initially launched in May 2022 for data center compute GPUs, Nvidia will now fully adopt open-source GPU kernel modules across all platforms.
After two years of development and testing, Nvidia asserts that the open-source GPU kernel modules now deliver performance on par with, or superior to, proprietary drivers. This transition is made possible because all Nvidia GPUs share a common driver architecture and capability set, allowing the same fundamental driver to be used for both home and cloud-based AI workloads.
The open-source kernel modules also introduce significant new capabilities, including heterogeneous memory management (HMM) support, confidential computing, and coherent memory architectures for Nvidia's Grace platforms. Nvidia recommends switching to these modules for GPUs from the Turing, Ampere, Ada Lovelace, or Hopper architectures. Grace Hopper and Blackwell GPUs require the open-source driver.
However, older GPUs from the Maxwell, Pascal, and Volta architectures are not supported by the open-source modules. This includes models such as the GTX 750 Series, GTX 960 Series, and Tesla V100 Series.
Nvidia's driver installation methods will generally default to the open-source driver, though certain scenarios require special attention, such as package managers with the CUDA metapackage, runfile installation, helper scripts, Windows Subsystem for Linux, and the CUDA Toolkit.
Switching to open-source GPU kernel modules enhances integration with the Linux operating system, allowing developers to debug, integrate, and contribute to driver improvements. This transition also aids distribution providers in managing ease of use and improving the out-of-the-box user experience.
0 Comments