Folding Forum

I've been reading a fair about about a lot of people building out GPU based folding farms and I remember reading a while ago on the FAQ's that GPU clients can only do a rather limited subset of all of the work/calculations/simulations that the entire F@H project needs, and that a CPU client run slower, but can process more varied types of WUs and projects.

So, from Pande Group, which is their preference: people running the GPU clients or people running the CPU clients?

While, I am sure that someone is going to come back say that any and all donations are being accepted, what, if any, is the preference from the group?

With all of these new GPU F@H farms going up, will there be a shortage of GPU WUs long before there'd be a shortage of CPU WUs?

I know you asked for a PG response, but what you asked is a common question, with common answers.

You are correct, all donations are welcomed, in whatever form they come.

Vijay has responded in the past that asking him to choose a prefered client is like asking him to pick a favorite child. They are all his favorite children. (source: dl.tv video interview)

Unfortunately, there will never be a lack of work to do until all of the diseases are cured. Besides, a few hundred extra GPUs is no comparison to a few hundred thousand CPU clients. Neither will run short any time soon.

And while Vijay is not likely to comment on this further than I have (as in the past), you can infer from the current level of development activity on the various clients as to where the most potential for the project is at this time (although development activity priorities change all the time, so don't infer too much).

7im wrote:I know you asked for a PG response, but what you asked is a common question, with common answers.

You are correct, all donations are welcomed, in whatever form they come.

Vijay has responded in the past that asking him to choose a prefered client is like asking him to pick a favorite child. They are all his favorite children. (source: dl.tv video interview)

Unfortunately, there will never be a lack of work to do until all of the diseases are cured.

As I've mentioned, I've seen a fair number of GPU folding farms going up (more so than CPU farms, for understandable reasons). Therefore; because the GPU client only do a subset of the work that needs to get done, would it stand to reason that the GPU WUs would run out sooner than CPU WUs?

It might appear that way at casual glance, but that is not the case.

IIRC, Vijay discussed this somewhat in his Blog (Project News page). This is what I took away from what he posted... For some of the research (not all, and no detail as to how much was given) the GPU (and SMP to a degree) clients are used as chainsaws to chew through large chunks of simulations, like a wood sculptor would cut away large sections of a statue during the initial stages of carving. Then CPU clients are tasked with doing or re-doing the finer and more accurate simulations, as that same sculptor does when in moving to chisels, knives, and finally sanding to finish the details of the job.

I'm sure all client types are used in many different ways, but the above demonstrates the cooperative nature of the various client types, and so neither GPU or CPU would run out of work before the other.

Vijay and the others do read the forum, so I'll let them comment or correct as needed, if needed.

alpha754293 wrote:As I've mentioned, I've seen a fair number of GPU folding farms going up (more so than CPU farms, for understandable reasons). Therefore; because the GPU client only do a subset of the work that needs to get done, would it stand to reason that the GPU WUs would run out sooner than CPU WUs?

Folding farms are good, whether they're GPU or CPU or PS3. There are A LOT of proteins to be studied. It's unlikely that there will ever be enough computers of any category to finish folding all those proteins within our lifetime.