Google

Sunday, January 23, 2005

General Computing on a Video Card

The website gpgpu.org has some interesting information on general purpose computing using graphics processor units. There are servaral languages that are supposedly high level and do not require much knowledge of the hardware. While the capabilities of GPUs are impressive, it is unkown how it would translate into real results. First I believe ATI's floating point processor is not even single precision. Second, GPUs are very good at vectorized calculations and it may be difficult to program to take advatange of this. Third, from a quick look at newegg it appears that GPU's can cost as much or more as relatively new CPUs. Finally, GPUs have their own RAM which is generally much less than the system RAM and I believe the latency to the hard drive would be greater compared to the CPU.

This is something I'd like to look into, but I don't have the time or money right now. I have a feeling that if you took an equally priced GPU and CPU, took a standard code and compiled both to run but did not optimize it for either, the CPU would come ahead. That's just a hunch though.

1 Comments:

At 11:48 PM, Blogger Max said...

I had a friend who was using the video ram decades ago for improved performances. Basically video RAM is different than other RAM because you can read to and write from an address in a single operation. But these days CPU and RAM are cheap and abundant. No really reason to mess with that now. Instead of spending the dollars on programming optimizations, just spend them on a faster chip. Same result, less headache.

 

Post a Comment

<< Home