Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

cuBLAS is Nvidia's standard go-to library for accelerating their GEMM code. In this repository I have included my attempts of writing kernels (without using TENSOR_WMMA) to try and meet the cuBLAS performance on my own gpu. The method I have followed is considering naive GEMM implementation as a baseline and then using NSight Compute to find the benchmark metrics.

GFLOPs (Giga-Float operations per second):

GFLOPs = 2 * (Matrix-Size)^3 / (1e6 * time_ms) (and the same for TFLOPs but using 1e9)

gflops_chart

Time metrics: using NCU --metrics gpu__time_duration.min

  1. Linear Scale (Time vs Matirx size)


    time_chart_linear
  2. Log Scale (Time vs Matrix size)


    time_chart

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages