PDA

View Full Version : Graphics Card folding??


Spanky Deluxe
05-19-2005, 12:06 PM
Hi,

I'll hopefully be building my new machine tomorrow but I was wondering if there was any way to fold using graphics card as the processors?? I swear I saw something along those lines a few months ago on here - I did a quick search for 'graphics card folding' but to no avail.

Since my setup should be SLI, it would be a waste not to use all that extra power. Is there anything??


Phil

sluflyer06
05-19-2005, 12:29 PM
if there is let me know cause that would be awesome considering how powerful GPU's are.

Disposibleteen
05-19-2005, 05:56 PM
GPUs arent that great, it cant hurt though

Magnj
05-19-2005, 06:21 PM
read a thread about htis also. It's possible in theory, but no one wants to code an app to do it do untill some one does we are SOL

matttheniceguy
05-19-2005, 06:24 PM
I also remember reading something like that. The upstream speed of AGP made doing calculations with the GPU pointless, but I think with PCI-e it can work quite well. The GPU is only suited to do certian types of calculations (floating point calculations?) but I did read something about folding and some other possable applications, where the GPU could more than double the speed of the computer.

BrokenWall
05-19-2005, 07:19 PM
think about it this way, when nvidia engineered the NV40 core, they designed it as a dual layer gpu, one of the layers was designed to render real-time video, for example if your working in premier and your cpu aint powerful to render the preview nice and smooth a plugin could be designed to take use of that layer in the NV40, and allow it to do the work, also this would allow you to use the video card to do all the rendering work to a final file, while allowing your cpu to be free to continue to use your computer smoothly, surfing the net, chatting, music, stuff like that.

If someone could capture that power in the NV40 core, then it would be possible.

sluflyer06
05-19-2005, 07:29 PM
considering how many transistors an modern GPU and the fact that a mainstream processor severly handicaps a 6800 or x800 they can't be too powerless....but yea it doesn't help if all that power is written into calculations that don't help.

Ackbar
05-19-2005, 10:02 PM
Interestingly... many facilities that require high computational power are looking into using PC GPUs for additional power. They're quite good with matrix operations and so for some things they're very nice. Unfortunately, for most programs people have to create very odd ports written at a low level for it to work on the GPU. This is not usually worth people's time/effort.

I didn't know that F@H had a GPU port...

Spanky Deluxe
05-20-2005, 03:33 AM
I'm sure its possible. Does anyone remember that guy that wrote some audio software that ran off of Nvidia graphics cards?? I think it was 6 series but it might have been FX. It seriously bumped up the power of the audio processing that was required and I think he retailed the software. It would make a normal computer as powerful as a fully professional audio computer or something. I'll try to find the link...

Spanky Deluxe
05-20-2005, 03:37 AM
Here's the link:

http://www.nforcershq.com/modules.php?name=News&file=article&sid=1853

Spanky Deluxe
05-20-2005, 03:39 AM
And here's the site link with their software: http://www.bionicfx.com/

If they can do this then I'm sure its possible theoretically to create some form of folding app for Nvidia GPUs. If not folding then at least something else of a similar useful distributed computing nature.

BTW Sorry for triple posting. :slap:

Ackbar
05-20-2005, 05:31 PM
And here's the site link with their software: http://www.bionicfx.com/

If they can do this then I'm sure its possible theoretically to create some form of folding app for Nvidia GPUs. If not folding then at least something else of a similar useful distributed computing nature.

BTW Sorry for triple posting. :slap:

Its a non-trivial task to develop code that will use the GPU. There are some resources available. I've looked into it for my own personal computational work, but it needs to be done at a very low level as standard compilers won't use this resource. Take a look at:
http://graphics.stanford.edu/projects/brookgpu/

You'd have to recompile the F@H application on an analogous compiler for it to work. AFAIK, Stanford isn't giving out their code.

Spanky Deluxe
05-21-2005, 03:58 AM
Found this link on their forums over there:

http://www.gpgpu.org/forums/viewtopic.php?t=180&highlight=folding

sc00p
01-02-2006, 01:40 PM
This is advancing all the time... Gromacs@GPUs:slobber:
http://folding.stanford.edu/FAQ-highperformance.html
"December 2005 We are working on optimizing the newest code. We are seeking help from GPU manufactures to tune our code on their hardware."

http://graphics.stanford.edu/%7Emhouston/public_talks/R520-mhouston.pdf
Well... P4@3GHz -> 12GFLOP@peak versus ATI (R520) X1800XT 120GFLOPs. Up to be able use out of that Ati's capacity ~30GFLOPs (2.5 X P4@3GHz) straight to overall speed bump for folding... (if I counted right). And this is the early stage when nothing is really optimized.
...GPUs would be HUGE folders (i guess with some dualie939-Opties).

LOL this has been discussed already here (http://xtremesystems.org/forums/showthread.php?t=80537). How did I missed that thread, sorry.