Folding Forum

I am curious if, for purposes of FaH, CPU's are just poor man's GPU's (far lower FLOPS) or if there are certain projects that can't be done on GPU and actually require CPU. No particular reason I'm asking, just a bit curious.

(I am just another user like you. This is what I understand, ideally it is true)

Every CPU project is only a CPU project, GPU Projects are completely separate.

I am vague on this but CPU projects are either Implicit, (protein is assumed to be folding is a continuous fluid) or Explicit (Protein and water molecules are both modeled during folding)

GPU projects had to use Implicit folding last I knew. (this may have been addressed since I was up to date, it could addressed in the future)

So there are more accurate but slower projects on CPUs and faster but less 'real' projects on GPUs.

No researcher want to throw out all his/her existing data, so no one ever switches mid project. Some researchers would not switch anyway as they need the accuracy of Explicit solvation.

The number of CPUs is simply much larger, my employer lets me use 55 PCs, one has a GPU.

GPUs now run explicit solvent projects, too.

FAH's fundamental design goal is to harness whatever Donor hardware that's commonly available for science and there still a A LOT of CPUs that are useful. GPUs are increasingly common in gaming machines and elsewhere. Individually, GPUs contribute more FLOPS, but FAH started on the CPU and collectively, they still do an awful lot of productive work. There's even talk about an upcoming software upgrade that will make use of the AVX feature that's found on many modern CPUs.

bruce wrote:GPUs now run explicit solvent projects, too.

Great! I worried that I was out of date. Glad to learn that the Science gets better and better!

The contributions of the CPU's are <1% of the contributions of the GPU's though. According to this: http://fah-web.stanford.edu/cgi-bin/mai ... e=osstats2

I am going to keep folding regardless, but part of me wonders - if not for the community element which is important - isn't it just a small drop in the bucket?

In a paint by numbers picture of a protein, the GPUs help to fill in the large sections of the picture very quickly. But without the CPUs filling in the fine lines and smaller details, it's just just a bunch of big color spots on the canvas.

Obviously this is an unrealistic oversimplification to suggest an explanation.