Nvidia quietly backs RISC‑V
Reports say Nvidia is backing SiFive’s RISC‑V CPUs and integrating them with its GPUs, signalling a bet on a more open, heterogeneous AI stack rather than tying everything to a single CPU architecture. The move would broaden interoperability options for AI infrastructure and reflect vendor hedging beneath the GPU layer. (ad-hoc-news.de)
RISC-V is the basic instruction language a processor speaks, and Nvidia is now backing SiFive to bring that language closer to its data center graphics chips. (sifive.com 1) (sifive.com 2) SiFive said on April 9 that it raised $400 million in an oversubscribed Series G round at a $3.65 billion valuation, with Nvidia among the investors. Reuters reported the company is using the money to push further into data-center central processors for artificial intelligence systems. (sifive.com) (usnews.com) Three months earlier, on January 15, SiFive said it would integrate Nvidia NVLink Fusion into its high-performance RISC-V data-center platforms. SiFive said that link would let its processors connect directly to Nvidia graphics processors and accelerators over a coherent, high-bandwidth fabric. (sifive.com) That pairing matters because modern artificial intelligence servers do not run on graphics chips alone. They also need central processors to feed data, manage memory, and coordinate workloads across racks of machines. (sifive.com) Most of that central-processor layer in servers has long been dominated by x86 designs from Intel and Advanced Micro Devices, while Nvidia’s own Grace processor uses Arm. SiFive is pitching RISC-V as a third option that customers can customize more freely around specific data-center jobs. (techcrunch.com) (sifive.com) Nvidia’s public line is about interoperability, not replacing every existing server processor. Jensen Huang said in January that Nvidia was bringing NVLink Fusion to “the RISC-V ecosystem,” while SiFive Chief Executive Patrick Little said customers want an “open and customizable” processor platform alongside Nvidia’s artificial intelligence infrastructure. (sifive.com) The timing also lines up with a broader push by Nvidia to make NVLink Fusion a wider standard around its accelerators. ServeTheHome reported in January that SiFive was the first RISC-V vendor to join that program, after earlier support from Arm, Intel, and Amazon Web Services. (servethehome.com) SiFive still sells processor designs, not finished chips, so adoption will depend on cloud companies and chipmakers licensing the technology and building systems around it. Reuters reported SiFive is aiming at the fast-growing market for data-center central processors, where new designs can take years to reach volume deployment. (usnews.com) For now, the clearest signal is that Nvidia is funding a company built around an open instruction set while wiring that technology into its own graphics-chip stack. That gives server builders one more way to pair Nvidia accelerators with something other than x86 or Arm at the processor layer. (sifive.com 1) (sifive.com 2)
