Next-Generation Edge AI with RISC-V Vector Cores for Vision Applications

on YouTube ↗17m

" summary

RISC‑V vector cores are integrated as two programmable vector engines of 2x8 CVA6V cores with 2MB scratchpad each that move 2D image data locally to avoid PCIe copy stalls and host preprocessing bottlenecks. A 4K vector length, 1.2 GHz operation and dedicated DMAs enable on‑chip pre/postprocessing and midprocessing for vision pipelines, claiming 16 Full HD streams at 30 fps on chip and per‑card performance that beats a mobile GPU core for NMS while the Europa line peaks at up to 629 TOPS at 45 W.

AI-written summary · about

" chapters (8)

1. 0:00Introduction and context
2. 1:32Challenges in edge AI
3. 3:42Evaluating design options
4. 5:37RISC-V vector architecture
5. 7:53Introducing Europa
6. 10:46Performance benchmarks
7. 12:31Software stack and roadmap
8. 15:24Q&A and closing remarks

" related talks

• 13mKeynote: Taking RISC-V Intelligence EverywhereCharlie Su · RISC-V Summit North America 2023
• 19mAccelerating Edge AI with the Help of RISC-VFlorian Zaruba · RISC-V Summit North America 2023
• 10mRISC-V Revolutionizing in AIVince Wu · RISC-V at embedded world 2024
• 24mReal World Results using Thousands of RISC-V Cores for AI and BeyondDave Ditzel · RISC-V Summit 2022
• 11mRISC-V enabled, low-power CNN classification in Edge devicesPer Andersson · RISC-V Summit Europe 2024
• 22mKeynote: RISC-V at NVIDIA: One Architecture, Dozens of Applications, Billions of ProcessorsFrans Sijstermans · RISC-V Summit North America 2024
• 20mRISC-V Spotlight: Expanding the RISC-V Horizon and BeyondCharlie Su · RISC-V Summit 2022
• 24mAndes RISC-V Processors for Control and Data PathsCharlie Su · RISC-V Global Forum 2020