7 replies to this topic

[maestro949]

Due to their awesome power for performing floating point operations GPUs are looking like they will be one means of providing much more horsepower to the protein folding and biosimulation efforts. Folding@Home may have its Gromacs engine port to GPUs by the end of the year. From the the FAH Wiki "Recent test data indicate performance gains of up to 40x that of an Intel Pentium 4 CPU are possible (note: this performance varies with different GPU's)."



FAH High Performance FAQ

General Purpose Computation on Graphics Processors - Mike Houston, Stanford University

gpgpu.org

Edited by maestro949, 04 August 2006 - 11:58 PM.

[curious_sle]

ah, finally... i've been waiting for this since shadermodel 3.0 came out :-). At least i have a excuse for a nifty GPU now that i don't use my pc for games anymore.

Now, if they'd port it to say a PlayStation 3/XBox 2 you could "build" your own @home cluster for a few thousand bucks that is amazingly fast...

[caston]

I was heavily considering buying a decent dual core Athlon 64 for e-cell and NAMD but now I'm interested in the cell processor. Should get myself a nifty PS3 when they are available at a sane price.

[maestro949]

Should get myself a nifty PS3 when they are available at a sane price.

I don't think e-cell nor NAMD are planning PS3 ports so I would hesitate going the PS3 route if it's an either-or situation and your intent is running MD software.

[caston]

It shouldn't be a problem as both have source code available so it could be ported to the platform and I'm sure I won't be the only one. I predict a thriving Linux/Cell community in the near future.

OK it might be a problem. A subcontractor tells me there's a lot of assembly optimisations. Still its great for folding @ home at least which is mainly written in C. NAMD has been ported to a number of different architectures so I expect it to be a matter of time before it's available for cell.

Edited by caston, 25 August 2006 - 02:34 PM.

[maestro949]

OK it might be a problem. A subcontractor tells me there's a lot of assembly optimisations.

Correct, the memory architecture for cell makes it quite a challenge to shuffle code around.

Still its great for folding @ home at least which is mainly written in C. NAMD has been ported to a number of different architectures so I expect it to be a matter of time before it's available for cell.

Indeed. Also, it will be intersting to see how well clusters of the cell based processors scale for supercomputers. It sounds like the improvements could be quite profound. We'll need it for larger scale simulations. Fingers crossed.

[Ghostrider]

FAH has already started folding on their GPU test cluster. They indicate that they are currently using ATI cards with plans to support nVidia eventually. Looks promising. Whatever accelerates scientific knowledge of biology without harming humans or animals, I support.

[caston]

The wikipedia page has some useful info about IBM releasing a developers kit to help make things a bit easier and the fact that the Linux kernel has supported cell since 2.6.16 in March.

http://en.wikipedia..../Cell_processor