Paul

FLOP-head

[Paul @paulWilliamChan]

I wrote a matmul kernel on B200 in pure CUDA/PTX that beats cuBLAS by 6% at M=N=K=8192.

Inspired by @gaunernst's blog on Blackwell instructions with benchmarking done on @modal.

Blog: https://t.co/nifzaMdPJS
Repo: https://t.co/EKjrnCrp0z

[Paul @paulWilliamChan]

@vikhyatk Insanely great

[Paul @paulWilliamChan]

@MainzOnX Sounds great, will let you know when I release it. Also very excited about where this could go, especially in regard to finding deadlocks in multi-gpu programing.

[Paul @paulWilliamChan]

@henrylhtsang

[Paul @paulWilliamChan]

@PatrickToulme @gaunernst @modal Will do 👍

[Paul @paulWilliamChan]

I wrote a matmul kernel on B200 in pure CUDA/PTX that beats cuBLAS by 6% at M=N=K=8192.

Inspired by @gaunernst's blog on Blackwell instructions with benchmarking done on @modal.

Blog: https://t.co/nifzaMdPJS
Repo: https://t.co/EKjrnCrp0z

[Paul @paulWilliamChan]

@clattner_llvm Would you consider something like cluster launch control to be micromanaging?