Enabling AI’s potential through Wafer-scale integration
summary
A wafer scale chip called Wafer Scale Engine (WSE) replaces conventional GPUs by scaling a die to ~46,225 mm² and packing 1.2 trillion transistors, 400,000+ cores and 18 GB on‑chip memory per wafer. Compared with the largest GPU it yields roughly 78× cores, 3,000× on‑chip memory, 10,000× memory bandwidth and 133,000× fabric bandwidth, enabling single‑chip models with very high locality and throughput.
AI-written summary · about
related talks
58mKeynote: Enabling the AI Revolution: Next-Generation Memory Solutions for AI Computing(…)
24mReal World Results using Thousands of RISC-V Cores for AI and Beyond
27mComputers for Super-Human AI
10mKeynote Analysis
01h 21mThe Future of Hardware Technologies for Computing
22mCurrent State of AI
13mFrom Silicon to Scale: Training at Scale
21mAnalyst Notebook