Skip to content

Comment on HipKittens: Fast and furious AMD kernelsparent

Comments

By far the easiest way to implement that "small number of algorithms" is with universal number-grinding hardware. Which also protects you against any architectural developments. Hardware takes a damn long time to make.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.