Wednesday, July 7, 2010

GPU performance

I am currently doing a research paper for an English class titled; "Which GPU is Better for Scientific Computing; Nvidia GTX 480 or ATI HD5870?". I am amazed at the stats on the HD 5870. It has a raw computation rate for single point precision numbers rated at 2.72 teraflops per second. That is more than double the Nvidia card. The nvidia card, however, has a wider memory bandwidth which, as I described in my previous post, can severely affect performance, but the raw computational performance is simply amazing. I also read that the Nvidia card handles thread batching and memory latency more efficiently so maybe the real world performance difference isn't as grand. I have a HD5870 in a gaming system at home and I think i may take a look at some OpenCL programming in my spare time. Maybe I'll do a head to head with the GTX 280 in the lab.