| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home
Thread: TFlops of the WCG ? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 30
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
My dear Dagorath, I sort of wrote it down in my first post in this thread...the finesse is proprietary (back of enveloppe put in round archive). Unfortunately not near Blue Gene and certainly no Peta Flops
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
David Autumns produced an estimate a few months ago, also. I didn't mention it because again, I didn't quite follow how he worked it out. It involved calculating the power of the "average" host equivalent. I searched the WCG forums for past discussions of WCG's computing power and ran across David's estimate. Correct, he did attempt to calculate the power of the "average" host. Although it may be possible to calculate the power of the "average" host if you have information on a broad sample of hosts, David simply assumed his own computer is the average host. We have no evidence that suggests his computer is average therefore we have no reason to believe his calculations are correct. The 3e13 is in the work unit info, if you crack open client_state.xml. It hasn't been officially confirmed by anyone, but WCG are going to use the best estimate they have, since the time estimation relies on it (initially). Please don't start on the flaws in the time estimation. Thanks for info on where the 3e13 number came from. I agree, WCG will use the best estimate they have. We have additional evidence that it's pretty darn close when we see that the product of the CPU benchmark and CPU time for a WU gives darn close to 3e13, on average. I also hope this discussion doesn't degenerate into a rant about flaws in time estimation. If anybody wants to discsuss that then please feel free to start a separate thread on it because it is off topic in this thread. I would still like to hear from Sekerob regarding his method. He has a knack for numbers too and his 65 is not far from Didactylos's 45. I think we're in the ballpark either way. ADDED: Oops! Sekerob got his post in before me. OK, Sekerob, my opinion is that your method suffers the same flaw as David's method... it takes a "stab" at the computing power of the average host and then extrapolates from that. I think without info on a broad sample of hosts it's just a stab in the dark and the extrapolation could go anywhere. Sorry, nice try but I can't buy it. I vote for 45 TFLOPS. [Edit 5 times, last edit by Former Member at Jun 24, 2006 12:42:55 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Not quite my dear watson.....there is one real hard factual very large number plus the stab actually had a bit more to it than the back of enveloppe, which i will try to retrieve......
----------------------------------------Since i'm less fuzzy now after reading a piece of David Autumns directions, in fact i think i can narrow it down further....i know the median is somewhere between 62 and 45.....a new variable was added last night.....60% Throttle.....will build it into the model, B.o.S. 
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
lets sign up more people and make it peta
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
rbonneau, I work every day on signing up new members.
Sekerob, yes the 60% throttle changes the picture. In 2 weeks we should have enough returned results to calculate a reasonable average daily results figure. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I searched the WCG forums for past discussions of WCG's computing power and ran across David's estimate. Correct, he did attempt to calculate the power of the "average" host. Although it may be possible to calculate the power of the "average" host if you have information on a broad sample of hosts, David simply assumed his own computer is the average host. We have no evidence that suggests his computer is average therefore we have no reason to believe his calculations are correct. It wasn't so simplistic. He scaled the value based on his average runtime per day and the aggregate average. I'm wary of any estimate that relies on the runtime, since we know it is one of our least accurate metrics (member/host count being the other totally off the wall figure). My estimate completely sidesteps that, and hopefully also adequately handles variability in work unit size. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
AT 1 MILION PC PARTICIPATING WE WOULD SURE BEET ANY SUPER
COMPUTER AT ANY KINDA TEST OR MATH U DO,. SPEED OR TERAFLOPS PER SECOND, SOO TO SETLE THE ISUE, I SUGEST AD THE REAMINING 650 000 PCS:) What is the computal power of WCG ? with 350,000 units... We have ~130,000 results per day. (TFlop needed for a result) X 130,000 / 24*60*60 EDIT: Nobody knows, i will try. I tested my CPU with Sandra 2007, a 3700+ core san diego, 2.2Ghz 1MB L1. my score is 6700 MFLOPS (6.7 GFLOPS) (with iSSE3) It takes me 5h30 to complete a task, 19800 seconds. so, 19800*6.7 = 132 TFLOP for one task if we have 130,000 results per day : (132*130,000) / (24*60*60) WCG = 200 TFLOPS (some days there is a peak at 150k results, 230 TFLOPS) For example, Earth Simulator computes 35 TFLOPS, and the best supercomputer in the world, bluegene does 280 TFLOPS. http://www.top500.org/system/7747 So ~200-230 is GREAT ! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sadly, that's not true.
The current top of the TOP500 list is BlueGene/L rated at a theoretical peak performance of 367 TFLOPS. However, the TOP500 list is about to release a new version (and I haven't kept up to date with supercomputer advances in the last year, so I don't know if there's anything faster than BlueGene/L out there). Also, we don't know about any classified supercomputer installations - and I'd be surprised if there weren't a few we didn't know about. However, with 1M members, we would be safely in the top 10 (even using our effective speed) and probably one of the top three supercomputers. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Okay, lets pick up this thread again. Did not find my back of envelope which and captured some nanosecond thoughts (probably already being recycled at the papermill). Rather than reconstructing, I went in search of some 'independent' source and found: BOINCstats. They publish their version for all projects on this system, which for WCG is set at a present average of 9.188TFlops. Given that BOINC represents about 15% of the total WCG results, the simple extrapolation is 61.25tflops. Give or take 61/62, its the return to the back of the lost envelope (without seasonal or throttle correction).....
----------------------------------------As Dagorath is on a Stampede this weekend, its save to echo rbonneau's call to go forward and hunt for the Peta Flops!  PS, None today.....but to say that my inconclusive and pending validation queue was completely empty last night and every new WU done since went up that PV lane.....no points today, a big dent in WCG stats  
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|