Demo: Neural Network Acceleration on Metis, Powered by RISC-V

on YouTube ↗07m

" summary

Metis is a quad‑AI‑core edge inference accelerator with a central RISC-V controller and a digital array for digital in-memory computing that performs matrix‑vector multiplies, paired with a Voyager SDK. The demo runs YOLOv8 pose + segmentation and fruit recognition cascaded across cores: pose detects a person, crops a high‑resolution ROI, downscales that ROI to 640× for detection, and runs segmentation in parallel with configurable core allocation (two cores for segmentation, one for detection, one for postprocessing). The Metis M.2 card sits in an Orange Pi via PCIe 3.0 and uses the host GPU via OpenCL for pre/postprocessing.

AI-written summary · about

" related talks

• 19mAccelerating Edge AI with the Help of RISC-VFlorian Zaruba · RISC-V Summit North America 2023
• 12mCustomizing RISC-V Cores to Accelerate Neural NetworksJon Taylor · embedded world 2022
• 10mDemo: NEOX TM : An Ultra-Low Power RISC-V GPGPU Enabling 3D & TinyMLVassilis Androutsopoulos · RISC-V Summit North America 2023
• 21mMTIA: Meta's First Generation of AI AcceleratorsRoman Levenstein, Amin Firoozshahian, Joel Coburn, Olivia Wu · AI Infra @Scale 2023
• 06mMETASAT Demonstrator: Mixed Criticality, Accelerated AI Computing for Future Space SystemsLeonidas Kosmidis, Marc Solé Bonet, Jannis Wolf, Matina Maria Trompouki · RISC-V Summit Europe 2025
• 20mComputer Vision at the Edge with OpenVINOKrishnakumar Shetti · ODSC India 2018
• 17mMi-V RISC-V Embedded EcosystemKrishnakumar Ranamoorthi · RISC-V Workshop Chennai 2018
• 21mKeynote: Building Data Center Scale SoC's using RISC-V at MetaPrahlad Venkatapuram · RISC-V Summit North America 2023