logoalt Hacker News

juancnyesterday at 5:20 PM0 repliesview on HN

In principle most if not all inference hardware should be usable for training.

Efficiency is the question.