Folding on a CPU & GPU?!

Matt555

iMod
Quoted from CUSTOM PC Magazine:
Amid the frantic 3D benchmarking at ATi's recent X1800-series launch, it also emerged that you can use your GPU for some other processing tasks, including folding and physics processing.
Because a high-end GPU can process a huge number of floating point operations in a second, as well has having quick access to on-board memory, it can outperform a CPU in some processing tasks that benefit from data parallelism, ATi used an example of a rolling ocean to demonstrate the physics aspect; the waves stopped and stuttered with a CPU calculating physics, ut rolled smoothly with an X1800XT in charge. According to http://techreport.com there was also talk of using a CrossFire setup with one card handling the graphics and the other handling the physics.
Mike Houston from Stanford University (home of the folding@home project) also presented some performance figures, showing that a Radeon X1800XT could process GROMACS units almost three-and-a-half times as fast as a 3GHz Pentium 4. Nvidia's GeForce 7800GTX, meanwhile, could only perform the job at around half the speed of the Pentium chip. Houston said that the tests showed 'a very promising application for GPUs' and that you could 'combine CPU and CPU processing for a folding monster!' See http://graphics.stanford.edu/~mhouston
Interesting, very interesting!
 
Awesome, really when you concider how much power those beasts have it's no wonder people haven't tried this sooner.
 
Yeah, it's a great thing, you've got your CPU idling and then think I'll do something with it because it's just wasted time with your computer on but not doing anything productive...your graphics cards are the same...
 
Nice find! But why exactly are they able to calculate the physics aspects more efficiently? Is it because the RAM is much faster and easily accessed?
 
StimpE said:
Nice find! But why exactly are they able to calculate the physics aspects more efficiently? Is it because the RAM is much faster and easily accessed?

Graphics cards are VERY powerfull (most have more transistors that the most powerfull desktop processors) and also the fact that they have much much faster ram.

I like this idear :D
 
Yeah they have access to some of the quickest RAM and have very good processors, as publicenemy said, most of them do have more transistors than the most powerful desktop CPU's...
 
The architecture of CPUs and GPUs differ however, although some algorithms the GPU would excel because of minor differences in the actual processing unit sometimes they aren't as suited as well to some pure mathematical operations or certain functions as they focus more on matrix calculations.

Just to say, complexity and performance isn't just about transistor count. A GPU with more transistors than a CPU doesn't mean it performs better as they are designed to do different things. The reason why GPUs have so many transistors is that a lot of them are dedicated to the pipelines. A CPU is designed with every possible bit of performance in mind, and this goes down to individual transistors, while GPUs have design speed in mind; they tend to have a more parallel processing in mind.
 
So how do we use our gpu's for f@h? (what do i need to do :P)

I would spend more time looking into this myself but i have to go and get a bus! (and im sure one of you will have allready worked it out :D )
 
AMD + NVIDIA UNITE!

wow very nice find,

itd be pretty cool if they invented cpus with integrated powerful nvidia cards

more room for pci slots, less heat, and a big as hell cpu frame
 
Back
Top