At the 2025 RISC‑V Summit in China, NVIDIA surprised the world with news on its CUDA environment: it will henceforth fully support RISC‑V processors as host CPUs. Historically, the NVIDIA CUDA host systems comprised x86_64 and AArch64 architectures; earlier support existed for PowerPC, Itanium, and SPARC. RISC‑V-based hosts can now use the CUDA runtime, drivers, and full software stack, though.
This change moves RISC‑V, an open and royalty-free instruction set architecture, to a prominent position inside NVIDIA’s heterogeneous computing systems. Three main components in NVIDIA’s reference design split jobs. A RISC‑V CPU manages NVIDIA CUDA drivers, the operating system, and application logic; an NVIDIA GPU runs parallel compute kernels; and a DPU monitors data movement and networking. This modular design not only simplifies system orchestration but also increases flexibility for edge as well as possible datacenter deployments.
NVIDIA CUDA’s addition to RISC‑V offers several tactical advantages. First of all, developers and system architects are now enabled to build bespoke artificial intelligence and high-performance computing solutions using open-source RISC‑V hosts, therefore removing dependence on closed-source x86 or ARM CPUs. Open-source CPU architectures are especially useful in sectors and educational settings where licensing fees or political considerations increase their appeal.
It encourages invention on edge systems like NVIDIA’s Jetson family by enabling the creation of bespoke accelerator cards based around RISC‑V cores. Launching NVIDIA CUDA support for RISC‑V marks a major change in the plan of the software of the company and demonstrates a firm dedication to open ISA ecosystems. NVIDIA not only highlights its leadership in artificial intelligence acceleration but also opens a new age of open, adaptable computer systems by expanding NVIDIA CUDA’s functionality past x86 and ARM.
Wall Street’s prediction that Amazon could also be inspired by this project to use comparable techniques, which would eventually lead to altering the course of AI system development in 2025 and beyond.
