News Nvidia's CUDA platform now supports RISC-V — support brings open source instruction set to AI platforms, joining x86 and Arm

The article said:
At the 2025 RISC-V Summit in China, Nvidia announced that its CUDA software platform will be made compatible with the RISC-V instruction set architecture (ISA) on the CPU side of things.
Everything you need to know about the rationale for this move is contained in that first sentence.
 
basically Nvidia making sure its tech is in the likely future hardware so it can maintain the monopoly in future.
It's just catering to the Chinese market. That's the main thing.

Nvidia does have rather deep history in RISC-V. Every Nvidia GPU has something called a GPU System Processor that acts as an intermediary with the host CPU. I believe that's currently implemented as a 2-way RISC-V core.

They also implemented their NPUs (called NVDLA), in their edge SoC's, as RISC-V with wide vector engines and tensor product units.

If you want to learn more, check this out:
 
Does anyone have the rough performance comparison between today's best RISC-V vs ARM vs x86?
Oh, it's bad. No contest. RISC-V is still in its infancy, but I'm sure it'll mature quickly. Check back in a couple years.

If you really want to know, here are some benchmarks:

Here's a detailed technical analysis & comparison of the C910:

TL;DR: it looks like a reasonable (if uneven) first effort, establishing a good foundation on which to build.
 

Latest posts