Folding Forum

7im wrote:WUs have an internal tally of the FLOPS count and is reported with each WU. So yes, and no. Multi-day WUs would reduce the the count, if only a few WUs were running. But with thousands running, it averages out so no drop. Kind of like your PPD averages out over time.

I guess it is all dependent on how they measure it... A user's PPD averages out over time because the running average takes points per day and averages them out over a number of days - to my knowledge the Client Stats page is only recording per day, not over a number of days, so it is quite different to average ppd...

It would be accurate if FLOPS measurement was analogous to calcuating by the number of FLOP performed per WU and then divided by the length of a day. I guess you and bruce are saying their measurement does take this into account and therefore adjusts for the relative 'size' of the WU... Which would make sense seeing as it is a pretty important to a lot of people.

If it was just the speed of the computer in FLOPS as measured by the client (which I use to see when I ran a Linux client) then my understanding would be that if a WU completed every second day it would average out, over thousands of users, at half the rate as every second day your client would report back with its FLOPS rate... Other clients would report back on your 'off day', but they also wouldn't be reporting on your 'on day'. As FLOPS is a instantaneous measurement your client would report the same number of FLOPS on a one day WU as a multi-day WU... If it is this way then any project that takes longer to return would have a marginal difference on reported TFLOPs proportional to the length taken to fold:

i.e. if WUs were smaller than a day every day you'd report 50GFLOPS day 1, 50GFLOPS day 2, 50GFLOPS day 3, 50GFLOPS day 5 => (50+50+50+50) / 4 = 50GFLOPS/Day

However, if larger and your client reported every second day it'd average out as: 50GFLOPS day 1, -- GFLOPS day 2, 50GFLOPS day 3, -- GFLOPS day 4 = > (50+ 0 + 50 + 0) / 4 = 25GFLOPS

That's ignoring how they handle multiple WU returns by individuals of course.

k1wi wrote:i.e. if WUs were smaller than a day every day you'd report 50GFLOPS day 1, 50GFLOPS day 2, 50GFLOPS day 3, 50GFLOPS day 5 => (50+50+50+50) / 4 = 50GFLOPS/Day

However, if larger and your client reported every second day it'd average out as: 50GFLOPS day 1, -- GFLOPS day 2, 50GFLOPS day 3, -- GFLOPS day 4 = > (50+ 0 + 50 + 0) / 4 = 25GFLOPS

Let's suppose your hardware processes 50G operations per day. If you're assigned WUs containing 50G operations, you'll complete one WU per day and it will look like your initial calculation: you'd report 50GFLOPS day 1, 50GFLOPS day 2, 50GFLOPS day 3, 50GFLOPS day 5 => (50+50+50+50) / 4 = 50GFLOPS/Day. Now suppose you're assigned WUs containing 100G operations. You'd complete one WU every other day and it'd average out as: -- GFLOPS day 1, 100GFLOPS day 2, -- GFLOPS day 3, 100GFLOPS day 4 = > (0 +100 + 0 + 100) / 4 = 50GFLOPS

Jesse_V wrote:His last Wikipedia contribution was in 2010. You can always create his Talk page and send him a message that way. Worth a shot, though it may not work. At worst case scenerio, you can always take the image, paste it into MS-Paint or the like, and look at the pixel coordinates and figure out the values from there. But that would be painful...

Hi everyone.

This is kaleb_zero from wikipedia/team 33 [H].
Jesse sent me an e-mail, and we kind of lucked-out because to be honest I don't check that old e-mail address that often these days.
I tried to sign up for this forum here with that e-mail but I couldn't validate it because it's with hotmail, so I've had to go with a slightly modified sign-in name.

Either way, this project you all have started is incredibly awesome. I only wish I still had the data that I had collected back when I made the chart found on wikipedia still on hand so that I could help you fill in some of the blanks from back in the day. I wonder if it would be worth it/appropriate to write a program to approximate the data from the chart, as the graphic is high enough in resolution that we might be able to approximate the data to within 10 or 20 TFLOPS?

Any thoughts?

Also, this cart, once more complete, would be invaluable to Wikipedia as a way to capture and showcase the computing power of Folding@Home for the history books.

kaleb0 wrote:
I tried to sign up for this forum here with that e-mail but I couldn't validate it because it's with hotmail, so I've had to go with a slightly modified sign-in name.

Welcome to the site, kaleb. Please check your private messages when you have a chance.

Sortofageek has been kind enough to correct my name issues.
In the meantime, I've managed to find a few snapshots from back in the day of the stats page and am adding them in now. Sept. 5, 2008 was on some czech website. I'm finding some decent data going through google image search with some search-fu!

kaleb_zero wrote:In the meantime, I've managed to find a few snapshots from back in the day of the stats page and am adding them in now. Sept. 5, 2008 was on some czech website. I'm finding some decent data going through google image search with some search-fu!

Ninja - sounds like your have some search skills, thanks for helping the 'project'!

Glad you're here with us, kaleb-zero! It's unfortunate that you don't have the data anymore, but thank you for your contributions nonetheless.


Does anyone have any connection to someone at the Pande Group who would know if archives of the stats page are kept??

In regards to the new Efficiency graph, see http://folding.stanford.edu/English/FAQ-PS3#ntoc13

We stress that one should not divide "current TFLOPS" by "active clients" to estimate the performance of that hardware running without interruption. Note that if donors suspend the FAH client (e.g. to play a game, watch a movie, etc) they enlarge the time between getting the WU and delivering the result. This in turn reduces the FLOPS value, as more time was needed to deliver the result.

I suspect that the Efficiency graph isn't very accurate.

I am impressed by the tweaking of the other graphs. I'm especially pleased that you resolved that glitch with the overall x86 FLOPS!

Why is everything plummeting? Ahhhh!

Jesse_V wrote:Why is everything plummeting? Ahhhh!

It is summer that is why.

Jesse_V wrote:
Why is everything plummeting? Ahhhh!

4P machines.

Jesse_V wrote:Why is everything plummeting? Ahhhh!

Summer is at least part of the answer - and for the participation graph - is actually worse because the active client count has quite a long average and for every client that gets shut down for a month others get shut down for parts of the day or parts of the week...

Given how severe the decline is (significantly lower than this time last year, where stats are available) part of me hopes that there is something else going on, such as the V7 client not passing on the information...

After all, despite all the improvements in computer hardware we're below mid-2009 FLOP levels. Between November and today the equivalent of 33,000 stock i7 2600k's switched off... :/

[Edit:} Actually, I think i was overestimating the power of the 2600k - the number is possibly more like 70,000

The efficiency graph was just an idea I decided to play around with and probably won't stick around unless a more accurate way to measure is found.

As for the declines, I hate to be a pessimist but we are also entering a 'post consumer-PC' world slowly but surely. People are getting more and more portable, and I know a number of people that after graduating university never bothered to buy another notebook or desktop and basically do everything on their smart phone - this is still the minority but it's definitely a trend.

7im wrote:

Jesse_V wrote:
Why is everything plummeting? Ahhhh!

4P machines.

Why 4P machines?

He is referring to people shutting off multiple single processor computers for fewer multisocket computers, because of the incentives for faster returns. the net result being, I presume he believes, a decrease in computational power.

k1wi wrote:He is referring to people shutting off multiple single processor computers for fewer multisocket computers, because of the incentives for faster returns. the net result being, I presume he believes, a decrease in computational power.

But why the sudden sharp drop?
I was looking at the native FLOPS milestones, the most recent was crossing 6.0 PFLOPS in November 2011, and now we're at 3727 TFLOPS.
But if what you say is true, shutting off multiple single processor computers for fewer multisocket computers, that should at least increase efficiency