CUDALibrarySamples icon indicating copy to clipboard operation
CUDALibrarySamples copied to clipboard

Published 20 hours ago • verified publisher icon NVIDIA

Which is faster for cublas or cublaslt when multiplying matrices of float, half, and int8

Open dingjingzhen opened this issue 2 years ago • 1 comments

hi， Which is faster for cublas or cublaslt when multiplying matrices of float, half, and int8

dingjingzhen avatar Feb 21 '23 07:02 dingjingzhen

Hi @dingjingzhen, are you asking if cublas is faster than cublaslt (or vice-versa) for those three precisions or are you asking which precision has the highest throughput?

hbabak avatar May 09 '23 15:05 hbabak