Codes

CUDA

by: Jeroen Bédorf ()

A simple CUDA implementation of the NBabel code. The code can work in double and single precision, see the source for more details.

Download: here

Performance:
Single precision Double precision
N T[s] dE T[s] dE
128 0,190 1,19E-06 0,859 1,73E-06
256 0,256 -0,00064 1,587 -0,00067
512 0,397 -0,04243 3,039 -0,04222
1024 1,023 -0,01742 5,900 -0,0191
2048 1,194 -0,01977 11,780 -0,00858
4096 2,854 -0,00184 44,160 -0,00257
8192 7,715 -0,00193 2m11 -0,00348
16384 25,550 -0,01029 7m12 -0,04431

Show sourceSelect a file