Skip to content

B200 Attention Kernel from Scratch to Near-SOTA in 60 Diagrams

iaroslavelistratov.github.io
24 pointsmagoghm3 comments
On HN

Comments

Pretty cool, it's missing how these concepts map to ThunderKittens code (my favorite CUDA toolkit)

One of the best AI GPU Attention Kernel how-to-implement articles I've read in a long time!

Tremendous work, tremendous effort was placed into writing this article -- and it shows!

Well done!

Really interesting project, appreciated.

AboutSource Built by g1lg1l

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