Hackathon Presentations

on YouTube ↗46m

" summary

A 25‑student, 4.5‑hour hackathon uses KIP/CODAL, the SDK and Renaissance hardware to speed up a neural‑net inference loop; teams add multiply and MAC instructions to an RV32I baseline (≈20–23k cycles) and repeatedly hit ~2,076–2,664 cycles (about 8–10× speedup). Several groups vectorize 8‑bit packing into 32‑bit registers to process multiple MACs at once; one extreme micro‑optimization reduces a tiny kernel to ~48 cycles. Some CODAL changes work, one misaligned load causes a hang.

AI-written summary · about

" chapters (10)

1. 0:00Hackathon introduction
2. 1:35First participant presentation
3. 5:44Second participant presentation
4. 13:34Third participant presentation
5. 18:57Fourth participant presentation
6. 23:25Fifth participant presentation
7. 28:30Sixth participant presentation
8. 34:27Seventh participant presentation
9. 38:17Eighth participant presentation
10. 43:17Event wrap-up

" related talks

• 45mHackathon PresentationsMegan Lane · RISC-V Summit North America 2024
• 12mCustomizing RISC-V Cores to Accelerate Neural NetworksJon Taylor · embedded world 2022
• 13mDemo: RISC-V as an Enabler of Heterogeneous ComputeZdenek Prikryl · RISC-V Summit North America 2023
• 12mDemo: NVIDIA's use of Catapult HLS for Building ML Inference AcceleratorsStuart Swan · RISC-V Summit North America 2023
• 23mKeynote: LaunchpadAndrew Moore · RISC-V Summit North America 2024
• 05mDemo: Optimizing ML for IMU Sensor Action DetectionPeter Robertson, Thomas Hepworth · RISC-V Summit North America 2023
• 13mKeynote: Taking RISC-V Intelligence EverywhereCharlie Su · RISC-V Summit North America 2023
• 53mPoster / Demo PreviewsSamuel Steffl · 7th RISC-V Workshop 2017