View Single Post
Old 2009-07-12, 21:30   #8
(loop (#_fork))
fivemack's Avatar
Feb 2006
Cambridge, England

11001001010012 Posts

This may be a rude question, but have you actually written much code on your CUDA setups? Google finds me lots of people who have run CUFFT and mention how large the gigaflops numbers that come out are, a few people doing the same for linpack, and the developers of some Russian GIS software which uses CUDA for all the heavy lifting, but nothing which looks like a particularly vibrant CUDA-writing community.

There are lots of papers full of what after a while read like the same commonplaces on how to write good CUDA code, but I've only really seen the one nVidia conference presentation where they take slow code and demonstrate how it may be changed into fast code, and there are several steps there that I don't quite follow and several others where I can't see how they could be made relevant for sieving.
fivemack is online now   Reply With Quote