-----Original Message----- From: cctalk-bounces at classiccmp.org [mailto:cctalk-bounces at classiccmp.org] On Behalf Of Paul Koning Sent: vrijdag 23 september 2005 15:31 To: cctalk at classiccmp.org Cc: cctech at classiccmp.org Subject: Re: State of the art Chuck> I was catching up on my reading and came across an interesting Chuck> article in the September 2005 IEEE Computer Society "Computer" Chuck> magazine. The gist of the particular article was that the Chuck> graphics processors on many high-end PC video cards are Chuck> overlooked for applications requiring heavy number-crunching. Chuck> What caught my eye was the chart that illustrated that the Chuck> Nvidia G70 graphics processor now performs at about 170 Chuck> GFlops! (A dual-core Pentium 4 running at 3 GHz, by contrast, Chuck> will do about 20 GFlops). 20 GFlops with a 3 GHz processor? How can that be -- that would require 3 FP functional units per core, each issuing one instruction per clock cycle. Did you mean 2 GFlops? Chuck> Granted, this is 32-bit vector floating point arithmetic, but Chuck> the raw numbers are pretty stunning. Sure is. There may be as many as three processors tucked away in your PC that are more powerful (though more narrowly focused) than the main processor. The graphics engine is one -- the other two are the digital signal processing engines in the disk read channel and (if you have one) the Gigabit Ethernet interface. There's a group somewhere (can't remember the name or URL) working on developing parallel processing algorithms that run on your graphics card -- essentially treating it as a vector coprocessor for your PC. paul