Hacker News new | past | comments | ask | show | jobs | submit
Duh, China?
We also see improvements like 7x between the current and last Google TPU generations, with different chips for training vs inference.