The interesting part is that this is less "CUDA now runs on RISC-V" and more "Nvidia is defining what a CUDA-capable RISC-V server has to look like."

That could end up being a pretty influential de facto server profile for the architecture.