Performance increase? Too good to be true?

I have a fairly complex cellular automata application which uses a 154x154 grid.

In standard C++, 500 iterations take 258.22 seconds to complete.

Using CUDA, these 500 iterations only take 17.802 seconds!

On each iteration, a calculation takes place involving locating circular neighbours within a certain radius of each grid cell. So I suppose this application is very suited to GPU programming!

Have any other users founds similar speed ups?! I’m double checking my code as this seems an unbelievable increase External Image

Thom

a speedup of ~14x is nice, but also quite usual. with some problems you can even reach a speedup of more than 100x. :-)
you can take a look at nvidia.com/cuda, where they write a speedup compared to cpu for nearly all featured projects.

nevertheless, great work and keep optimizing. ;-)