NVIDIA has made CUDA available for RISC-V CPUs.

1. CUDA and the Host CPU

CUDA is a software stack for utilizing NVIDIA GPUs. Previously, the premise was that x86 (Intel/AMD) or Arm (especially NVIDIA Jetson) CPUs served as hosts. The host CPU orchestrates (control, driver execution, and application logic) the parallel computations executed on the GPU.

2. Key Points of NVIDIA’s Announcement

RISC-V CPUs were officially announced as CUDA host CPUs (July 2025, RISC-V Summit China). The announcement was made during a keynote speech by NVIDIA’s Frans Sijsterman (VP of Hardware Engineering). CUDA’s main components (drivers and system control) now support the RISC-V ISA.

The expected configuration is:

① NVIDIA GPU: Responsible for parallel processing (heavy AI and HPC calculations)

② RISC-V CPU: Runs CUDA system drivers, application logic, and the OS

③ DPU (Data Processing Unit): Responsible for networking and data transfer. This completes the heterogeneous platform.

3. Strategic Implications
(1) Chinese Market Strategy

Due to US-China export restrictions, NVIDIA is unable to ship its latest GPUs, such as the GB200/GB300, to China. At the same time, it wants to maintain and expand the CUDA ecosystem.
→ Therefore, it makes sense to incorporate the growing Chinese RISC-V standard into the host CPU and expand CUDA to “RISC-V compatible.”

(2) Support for Open Architecture

RISC-V is an open ISA, making it attractive to companies (especially Chinese and application-specific semiconductor companies) that want to create custom SoCs. CUDA’s support for RISC-V will enable a hybrid “proprietary GPU + open CPU” architecture, appealing to a wider audience.

(3) Data Center vs. Edge

As the article points out, RISC-V will not be widely adopted in hyperscale data centers anytime soon. However, it is likely to be used in edge AI devices like the Jetson module and custom servers. In the long term, this could pave the way for RISC-V to become a popular CPU choice for HPC/AI.

4. Future Outlook

The CUDA × RISC-V combination will become a “third option” alongside the traditional CUDA × x86/Arm combination. In particular, in China and other countries with strict government regulations, it will become possible to build AI/HPC platforms using a domestic RISC-V CPU + NVIDIA GPU. If this trend spreads, other companies (Intel, AMD, and Arm-based SoC vendors) may also accelerate their support for RISC-V.

To summarize:
CUDA’s official support for RISC-V is a strategic move by NVIDIA aimed at maintaining the ecosystem and addressing the Chinese and open markets. While its use will primarily be focused on edge devices in the short term, this move will have an impact in the long term, as it will make RISC-V a CPU option in data centers as well.

Reference article:

Nvidia’s CUDA platform now supports RISC-V — support brings open source instruction set to AI platforms, joining x86 and Arm.
News by Anton Shilov published July 20, 2025

https://www.tomshardware.com/pc-components/gpus/nvidias-cuda-platform-now-supports-risc-v-support-brings-open-source-instruction-set-to-ai-platforms-joining-x86-and-arm