/warp_vs_taichi

compare warp and taichi performance

Primary LanguagePython

warp_vs_taichi

compare warp and taichi performance

Usage:

python3 torch_conversion.py -lang taichi
python3 torch_conversion.py -lang warp

Findings (cost of conversion between pytorch and taichi/warp):

taichi: 0.326s
warp:   0.028s

The recorded timing is the total time, i.e., for converting back-and-forth for 1000 times.

  • Environment: NVIDIA GeForce RTX 3080.
  • Matrix size: 27008 by 3.