AI Hardware & Semiconductors
A Rising FLOPs-per-Byte Ratio Explains Why Nvidia Split the Chip in Two
FLOPs per byte, tracked across six Nvidia data-center GPU generations from the 2016 Tesla P100 to 2026's Rubin, climbs from under 30 to over 1,500 — and then the ratio forks into two numbers the day Nvidia ships a GPU with no HBM in it at all.