Advancing AI Application Development on RISC-V Platforms
summary
A company ships a 64-bit 8‑core RISC‑V AI SoC board (motherboard starting at $299) that runs Whisper, VLC speech‑to‑text, webLLM and text‑to‑image workloads entirely on RISC‑V hardware; current hardware runs 4B and 7B models “okay,” and a planned 32‑core / 256‑LBDDR mini‑ITX with four chips aims to host 70B models. Porting uses Vulkan/OpenCL backends and RVV 1.0 DSP instruction emulation; software and upstream kernel work remain the bottleneck. Sold ~1,000 units, targeting 10,000.
AI-written summary · about
chapters (10)
- 0:00RISC-V AI hardware innovation
- 3:16Software to hardware with RISC-V
- 7:29RISC-V future for AI and open hardware
- 10:42High-performance RISC-V SOC challenges
- 13:57AI role in RISC-V platforms
- 18:42Hardware-software collaboration for RISC-V
- 23:50Call to action for open-source support
- 27:08Educating RISC-V developers
- 31:30AI opportunity for RISC-V growth
- 35:17Q&A
related talks
13mKeynote: Taking RISC-V Intelligence Everywhere
24mReal World Results using Thousands of RISC-V Cores for AI and Beyond
19mWhy RISC-V?
22mKeynote: RISC-V at NVIDIA: One Architecture, Dozens of Applications, Billions of Processors
40mKeynote Panel: The Future of High Perfor...
19mDemocratizing Inference of Open-weight Models on RISC-V Manycore Acc...
19mAccelerating Edge AI with the Help of RISC-V
10mRISC-V Revolutionizing in AI
