Folding Forum

Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months? At one point it was hovering up in the 4.5 PF range, I believe (didn't it break 5 briefly?), and then it seemed that after the GPU server problems it dropped substantially to its current 3.5ish. I seem to remember the native Nvidia PF being over 2 at its peak, whereas now it's just over 1, which would be the prime contributor to the overall drop. Did that many Nvidia contributors abandon the project because of the server problems?

k1wi wrote:He might be referring to BOINC's total overall FLOPS (4.9PetaFlop) being higher than Fah's 3.874..?

But that's spread out over a wide number of different projects and whether BOINC's reporting NativeFlops or x86 flops I'm not sure.

Oh, I see. Now Boinc is a DC project? I thought it was simply a network architecture upon which projects are run.

I wonder how many FLOPS we can add up with people doing protein simulations? That's more homogenous that boinc...

Folding@Home, Rosetta@Home, Proteins@home, TANPAKU, Predictor@home, SIMAP, HPF, Docking@home, ...

....gpugrid, drugsdiscovery

Zagen30 wrote:Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months?

SMP2/bigadv from cheaper CPUs? They're outproducing GPU farms that haven't really gotten cheaper.

Wrish wrote:

Zagen30 wrote:Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months?

SMP2/bigadv from cheaper CPUs? They're outproducing GPU farms that haven't really gotten cheaper.

That's true.

Many people ran SMP concurrently with GPU2. With the restructuring of the bonus system, many people have found that SMP2 (FahCore_a3) awards more points than SMP2 + GPU2. (That depends on a number of factors so it might not apply to you.) The FLOPS for SMP may have been slightly lower than the FLOPS for SMP2, but they're going to be similar so some of these folks have simply removed their GPU2 client, hence the FLOPS that are being counted goes down by one or more GPUs.

The goal here is to align the points system with the value of the scientific results. There are likely other changes coming and there's no way to know how they will influence the FAH donors, but you do need to remember that FLOPS is not necessarily a good measure of the science being produced. Ultimately, though, it seems counter-intuitive to believe that a CPU-only system can produce more science that a CPU+GPU system, so there should be a way to encourage folks to fold on their GPUs, too.

Zagen30 wrote:Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months? At one point it was hovering up in the 4.5 PF range, I believe (didn't it break 5 briefly?), and then it seemed that after the GPU server problems it dropped substantially to its current 3.5ish. I seem to remember the native Nvidia PF being over 2 at its peak, whereas now it's just over 1, which would be the prime contributor to the overall drop. Did that many Nvidia contributors abandon the project because of the server problems?

Uhhh, the global super-recession happened.

People and businesses decided to reevaluate their electricity usage to save money.

Prettz wrote:

Zagen30 wrote:Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months? At one point it was hovering up in the 4.5 PF range, I believe (didn't it break 5 briefly?), and then it seemed that after the GPU server problems it dropped substantially to its current 3.5ish. I seem to remember the native Nvidia PF being over 2 at its peak, whereas now it's just over 1, which would be the prime contributor to the overall drop. Did that many Nvidia contributors abandon the project because of the server problems?

Uhhh, the global super-recession happened.

People and businesses decided to reevaluate their electricity usage to save money.

If my facts are right, the throes of the recession occurred in 2008, but F@h was growing in FLOPs through that time until around April 2009, and started dropping off this past fall/winter, which doesn't exactly line up.

Zagen30 wrote:

Prettz wrote:

Zagen30 wrote:Somewhat tangential, but what happened to F@h's FLOPs in the past couple of months? At one point it was hovering up in the 4.5 PF range, I believe (didn't it break 5 briefly?), and then it seemed that after the GPU server problems it dropped substantially to its current 3.5ish. I seem to remember the native Nvidia PF being over 2 at its peak, whereas now it's just over 1, which would be the prime contributor to the overall drop. Did that many Nvidia contributors abandon the project because of the server problems?

Uhhh, the global super-recession happened.

People and businesses decided to reevaluate their electricity usage to save money.

If my facts are right, the throes of the recession occurred in 2008, but F@h was growing in FLOPs through that time until around April 2009, and started dropping off this past fall/winter, which doesn't exactly line up.

Maybe F@H is a lagging indicator

Or a leading indicator...

Another reason to not F@H:

"My -enter significant other here- sits at the computer for hours trying to tweak his system to maximize something he calls 'PPD' and 'TPF.' He is pale skinned and forgets the name of his children. He uses phrases like "Up your Vcore," and "Stupid A3 Core" all the time. The kids are frightened. Even Dog the Bounty Hunter gives him a wide berth. Is there any end to this madness?!"

Answer THAT smart people!

(There is no answer... it's an addiction that will surely end up being called a disease of it's own in no time)

Embrace it!

John Naylor wrote: Can’t they just use a Supercomputer? /They already have 400,000 processors, how is my one processor going to make any difference?
F@H is more than three times as powerful as the world’s current most powerful supercomputer, in terms of operations per second, so using a supercomputer would be a massive step backwards for the project. Even with that in mind, the project is still restricted by the power available to it and needs all the extra silicon it can get.

This statement is no longer true. Japan's K Computer stands at a whopping 8.162 petaFLOPS. See http://en.wikipedia.org/wiki/K_computer However, the K supercomputer is not dedicated or built for protein folding, so its use for that may be limited at best. A better argument would be that RENTING a supercomputer is expensive, and you probably only have a piece of it, and we need raw CPU power for individual slices of a simulation more than crunching a massive dataset at once. Blue Gene is used for protein folding, and a later version of the supercomputer should come online soon with the performance of 20 petaFLOPS. They may have us beat, but we are way up there. Basically supercomputers are designed to process massive amounts of data extremely fast, but the machine has to be maintained, cooled, and whatnot so they can be costly to run. With Folding@home, each donor has a small data set which they crunch at a slow rate, and so a massive dataset is divided up and processed in parallel. While PG does have to pay for professional development of the software, they don't have to pay for cooling all the machines. That's our job and we're glad to do it since we're doing all this great research anyway!

We are faster than everything under BOINC though: http://boincstats.com/stats/project_graph.php?pr=bo
I made note of this on the F@h Wikipedia article, which gets around 350 views a day by the way.

F@H is irrelevant now due to [Corporation X]’s research. Why bother?

Exactly what research is making some people think F@H is irrelevant?

Barry wrote:

F@H is irrelevant now due to [Corporation X]’s research. Why bother?

Exactly what research is making some people think F@H is irrelevant?

AFAIK there is none. This topic is mostly a collection of excuses for not running F@H. Many of those excuses are inaccurate -- just urban legends -- but that doesn't mean somebody will never claim it to be true.

bruce wrote:

Barry wrote:

F@H is irrelevant now due to [Corporation X]’s research. Why bother?

Exactly what research is making some people think F@H is irrelevant?

AFAIK there is none. This topic is mostly a collection of excuses for not running F@H. Many of those excuses are inaccurate -- just urban legends -- but that doesn't mean somebody will never claim it to be true.

Only 2 reasons really not to use fah first is not knowing about it like me I didn't know about it until afew months ago the second reason is computer can't handle it I know because sadly the computer setup I had could not run it because I was on it about 20 hours a day and the off time would not have been enough to finish a unit in time "I tried it" No other reason besides no internet I can think of

ford316 wrote:...the second reason is computer can't handle it I know because sadly the computer setup I had could not run it because I was on it about 20 hours a day and the off time would not have been enough to finish a unit in time "I tried it" No other reason besides no internet I can think of

What kind of computer do you have that cannot meet the deadlines running 20 hours per day? Certainly there are assignments that you can't complete, but there are also ways to get assignments that you CAN complete. In other words, it's a matter of learning how to configure your client -- and there's certainly work that needs to be done in future clients so that that information is readily available to even the most novice Donor who might find out about FAH.

There's really nothing that can be done about no internet, but that's a pretty unusual circumstance, these days.