Longbridge Dolphin Research·Oct 12Cerebras: Wafer-scale ‘monster’ chip — can it challenge NVDA?GPUs Idle 99.5% of the Time; Cerebras Removes the Wall
D RAHUL PRASANTH·Sep 2The Great Memory Race: Understanding Flash AttentionWhy is ChatGPT so fast? A simple, math-free breakdown of Flash Attention, the algorithm powering modern AI
Tai·Sep 1Samsung Maps Out HBM5 and zHBM: AI Memory Moves From Side-by-Side to 3DToday’s Three-Line Summary
Tai·Aug 25Samsung Puts a Brain in HBM, SK hynix Pushes Past 20 Layers — Semiconductor News, August 25, 2026My take: custom HBM only pencils out for the top two or three customers — but at NVIDIA and Google volumes, that may be exactly the point.
