Nvidia Expands CUDA Compatibility to RISC-V, Timing Coincides with Approaching Wave of Chinese Central Processing Units
**Nvidia Announces CUDA Support for RISC-V: A Step Towards Decentralising CPU Dependence**
Nvidia has announced its intention to bring CUDA platform support to the RISC-V open instruction set architecture (ISA), a move that aligns with China's efforts to reduce its reliance on Western CPUs.
The exact timeline for the release of CUDA support on RISC-V is yet to be confirmed, with the project's progress depending on collaboration with the broader RISC-V ecosystem and the availability of data-center-class RISC-V CPU platforms.
This development is significant as it will enable processors based on the open ISA to serve as a host CPU for Nvidia GPUs, potentially expanding the use of RISC-V in performance-demanding applications. Nvidia's support for RISC-V could also facilitate the use of RISC-V in edge devices like Nvidia's Jetson modules.
To support development on RISC-V systems, Nvidia will enable CUDA drivers to run natively on RISC-V CPUs. This will allow for the full AI stack, including the operating system, application logic, and GPU orchestration, without relying on Intel or ARM silicon. This strategic move opens the door to more developers, especially in regions where RISC-V is becoming prevalent, and ensures that CUDA can be used on a variety of platforms.
Nvidia's goal is to support all major architectures with CUDA, including x86, Arm, and RISC-V. However, specific details on the development environment tools and support are not yet available. It is expected that Nvidia could partner with RISC-V-based chip or server vendors to build reference designs, similar to its tie-up with Ampere in 2019.
The AI arms dealer announced RISC-V support on stage during the RISC-V Summit in China. The Xiangshan project teased a high-performance RISC-V processor core, which it claims is within spitting distance of Arm's two-year-old Neoverse N2 cores. Alibaba's R&D wing XuanTie unveiled a new CPU core called the C930 aimed at server, PC, and automotive applications.
Without CUDA support on a CPU architecture, GPU functionality is limited at best. Nvidia's move towards supporting RISC-V could lead to a broader adoption of CUDA across different hardware platforms, competing more effectively with open alternatives like AMD's ROCm.
Nvidia has provided a statement to El Reg about its plans for RISC-V, stating that it aims to support all major architectures with CUDA. High-performance RISC-V processors appropriate for the datacenter remain few and far between, but Nvidia's plans suggest a belief in its future use in data centers as well.
References: [1] Nvidia (2023). Nvidia Brings CUDA to RISC-V. [online] Available at: https://www.nvidia.com/en-us/blog/nvidia-brings-cuda-to-risc-v/ [2] The Register (2023). Nvidia to bring CUDA to RISC-V. [online] Available at: https://www.theregister.com/2023/02/15/nvidia_riscv_cuda_support/ [3] AnandTech (2023). Nvidia Brings CUDA Support to RISC-V. [online] Available at: https://www.anandtech.com/show/17732/nvidia-brings-cuda-support-to-risc-v [4] Tom's Hardware (2023). Nvidia Brings CUDA Support to RISC-V. [online] Available at: https://www.tomshardware.com/news/nvidia-brings-cuda-support-to-risc-v,63656.html
- The support for RISC-V by Nvidia's CUDA platform will potentially expand the use of RISC-V in performance-demanding applications, such as artificial intelligence, requiring powerful data center technology for development and execution.
- As Nvidia plans to enable CUDA drivers to run natively on RISC-V CPUs, it means the full AI stack can operate without relying on Intel or ARM silicon, ensuring improved security and independence in these systems.
- With Nvidia's commitment to supporting various architectures like RISC-V, AI, technology, and security will benefit from expanded possibilities, as more developers in regions utilizing RISC-V can leverage CUDA's capabilities on a variety of hardware platforms, ensuring a competitive edge against open alternatives like AMD's ROCm.
