Lightning Talk: RISC-V and AI: How an Open-Standards ISA Underpins Portability f...
summary
RISC-V claims to make AI/ML kernels portable by standardizing two concrete extension types: an integrated matrix extension that reuses vector state for ~2–4× speedups and an attached matrix extension that adds a tiled accelerator with its own registers and memory ports aimed at data center, edge and HPC. Companies (Meta, Google, Stream Computing, Esperanto, Rivos, Softco) are taping out or shipping silicon; one configuration targets ~32–40 TOPS on 12 nm. The effort expects 95–98% shared infrastructure between vendors.
AI-written summary · about
related talks
20mRISC-V Goes BIG
20mUnderstanding the RISC-V Extensions for AI
10mKeynote: Owning Your Silicon Future Starts with Building on RISC-V
19mKeynote: Designing the Next Billion Chips: How RISC-V is Revolutionizing Hardware
24mOpen Acceleration for RISC-V – Portability, Performance and Partners...
27mGlobal Importance, Adoption, and Opportunity for Europe in RISC-V
17mA dive into RISC-V
11mBoosting AI on Semidynamics RISC-V Cores with Custom Tensor Instructions
