Omri Weinstein(@WeinsteinOmri):Pearl (@prlnet) has the fastest FP8 MoE Kernels for Blackwell chips! Group-Matrix-Multiplication is one of the most optimized operations in AI. As a former researcher at @nvidia I can say firsthand that pushing the performance frontier of MatMuls is... Hard. Pearl-GEMM is ~5% faster than prior state-of-art Quack (@tri_dao) and ~42% faster than Flashinfer on B200s.

Omri Weinstein

@WeinsteinOmri

Computer scientist, co-founder of @prlnet, ex-Nvidia, Vast Data, Princeton PhD, ¶

가입 August 2020

212 팔로잉 중 3.7K 팬

Omri Weinstein@WeinsteinOmri

2026.09.17 19:12

Pearl (@prlnet) has the fastest FP8 MoE Kernels for Blackwell chips! Group-Matrix-Multiplication is one of the most optimized operations in AI. As a former researcher at @nvidia I can say firsthand that pushing the performance frontier of MatMuls is... Hard. Pearl-GEMM is ~5% faster than prior state-of-art Quack (@tri_dao) and ~42% faster than Flashinfer on B200s.