Message boards :
Number crunching :
Wow: 106 TeraFlops
Message board moderation
Author | Message |
---|---|
B-Roy Send message Joined: 4 May 03 Posts: 220 Credit: 260,955 RAC: 1 |
BOINC-based projects now have 80,721 participants in 188 countries. Together they supply 106 TeraFLOPS of computing power - far more than any conventional supercomputer. from the BOINC page |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
Interesting....how does 80,721 participants equate to 106 teraflops. I'm interested in how this was calculated, where is your supporting data or links? From the BOINC Statistics for the World web site we appear to have a daily public/semi public boinc RAC of around 11.6 million, based on a very approximate example using my 3.2GHz P4 HT computer whose RAC is approx 500 (give or take 10%). It used to do around 16 sah classic wu's a day with HT on. SETISpy estimates the approx teraflop per wu at about 3.8 (some more some less - but about average). Now doing some calcs around these numbers; 16 wu's / day * 3.8 teraflop = 60.8 teraflop per day (note the drop in the s since we're talking per day) = 704 megaflops (million floating point operations per second). I know this is very approximate but bear with me. Now this same comp has a RAC of 500 = 704 megaflops. So a RAC of 11.9 million = ~16.7 teraflops. Now I could be upto 20% out on this, or made a wrong assumption (like the setispy estimated teraflops), or used the daily RAC incorrectly, or my daily RAC is wrong since it is doing a few projects. But we are talking ballparks here and this number and yours are in 2 totally different ball parks. Live long and crunch! Paul (S@H1 8888) And proud of it! |
cRunchy Send message Joined: 3 Apr 99 Posts: 3555 Credit: 1,920,030 RAC: 3 |
I'm totally useless at these sorts of computer concepts or figures but how does this relate to my profile where my (XP AMD 1.3GHz) PC is measured at: ? * Floating point speed 1105.98 million ops/sec Thinking that the average computer on the system might be similar to mine then the figures work out like this: 1105000,000 (FLOPS) 80,721 (Users) ------------------ 89,196,705,000,000 (Result) Or 89 Tera FLOPS... But my PC may not be typical or some users are running several PCs? cRunchy |
Giordano Kaczynski Send message Joined: 16 Jan 05 Posts: 12 Credit: 1,183 RAC: 0 |
> But my PC may not be typical or some users are running several PCs? The first assumption is true (1.3 ghz is a little below the average computer that everyone now has at home). If you think mine (an Athlon Xp 1900+) has aroun 1420 million floating point ops/sec, you get more than 106 teraflops. <i>"I've seen things, you people wouldn't believe...hmmm... attack ships on fire off the shoulder of Orion. I've watched C Beams glitter in the dark near the Tannhauser Gate. All those moments, will be lost in time like tears in rain..."</i> |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
where is your supporting data or > links? I believe he got the info from "www.setisynergy.com" I was there earlier and noticed the stats, or atleast think that's where he/she got it. tony BOINC News - January 19, 2005 Contributed by Curly Brackets January 19, 2005 BOINC-based projects projects now have 80,721 participants in 188 countries. Together they supply 106 TeraFLOPS of computing power - far more than any conventional supercomputer. Source - http://boinc.berkeley.edu |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
Ok, Here is the real math. One: A cobblestone machine is one capable of performing 1000 MFlops and 1000 MInts per second. (or 1 Gigaflop and 1GigaInt) A cobblestone machine running at 100% produces 100 credits in 24 hours. To get a teraflop/second you would need 1000 cobblestone machines. In 24 hours, these 1000 machines would produce 100,000 credits. Two: (Reverse the math) To calculate how many teraflops/sec are being done, take the number of credits granted in any 24 hour period, and divide by 100,000. |
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
> Two: (Reverse the math) > To calculate how many teraflops/sec are being done, take the number of credits > granted in any 24 hour period, and divide by 100,000. > So an RAC of 11,600,000 / 100,000 = 116 teraflops. So classic was REALLY underestimating their computing power this whole time! [edit - but isn't our credit 100 times the real cobblestone?] Paul (S@H1 8888) And proud of it! |
Toby Send message Joined: 26 Oct 00 Posts: 1005 Credit: 6,366,949 RAC: 0 |
The other thing to keep in mind is that this is NOT the amount of processing done, but rather the amount of credit granted in a day. Usually this should be roughly equal however with the validator backlog that seti has been having, it is possible that there was a sudden jump if they fixed some bottleneck in the system. A member of The Knights Who Say NI! For rankings, history graphs and more, check out: My BOINC stats site |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
> The other thing to keep in mind is that this is NOT the amount of processing > done, but rather the amount of credit granted in a day. Usually this should > be roughly equal however with the validator backlog that seti has been having, > it is possible that there was a sudden jump if they fixed some bottleneck in > the system. Correct of course, To be accurate, with jumps like the validator, it would be best to average daily credit for some period of time. And deciding the 'period' could be its own thread ;) |
genes Send message Joined: 25 May 99 Posts: 117 Credit: 580,187 RAC: 0 |
If you used credit granted, then wouldn't you have to multiply the result by the Quorum amount (currently 3)? Granted, the *useful* work done is still tied to the credit granted, but the extra work is being done, and the Quorum amount is totally arbitrary and could be changed at any time. If they suddenly changed it to 2, there'd be this big jump in work done. -Gene |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
genes, My understanding of the statistic sites is they report the total credit granted to any and all hosts. Naturally this is only granted after validation, but in seti's case this normally ammounts to 3 hosts getting credit (possibly 4 now) and the sum of those 3 credits is added to the daily statistic total. (Hypothetical: credit granted = 35, hosts for this WU are 3, all 3 hosts are "close enough" for validation, statistic site shows increase of 35 x 3 or +105) If they reduced quorum to 2 hosts, the ammount wouldn't change as same number of hosts out there would be crunching and getting credited...but getting credit in pairs rather than triplets. |
JAF Send message Joined: 9 Aug 00 Posts: 289 Credit: 168,721 RAC: 0 |
> BOINC-based projects now have 80,721 participants in 188 countries. Together > they supply 106 TeraFLOPS of computing power - far more than any conventional > supercomputer. > > from the BOINC page > This is true, to a point. Please don't take this as a "put down" of Seti. It's not meant to be. However, while Seti may indeed be capable of running at xxx teraflops, a true super computer that is capable of 36 or 72 teraflops is a whole different animal. Seti does not control the xxx teraflops it uses. If you owned a 36 teraflops supercomputer, you could put that 36 teraflops to work when and how you wanted. Boinc is edging closer to being a super computer though. Projects can be added to access that potential distributed computing power. <img src='http://www.boincsynergy.com/images/stats/comb-912.jpg'> |
genes Send message Joined: 25 May 99 Posts: 117 Credit: 580,187 RAC: 0 |
My mistake. That's true, "credit granted" is granted to all members of the quorum, so changing the quorum amount wouldn't change the perceived number of Teraflops, it would only change the amount of "unique" work done. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> This is true, to a point. Please don't take this as a "put down" of Seti. It's > not meant to be. However, while Seti may indeed be capable of running at xxx > teraflops, a true super computer that is capable of 36 or 72 teraflops is a > whole different animal. Yes, but the latest generations of supercomputers are all of the same basic type with lots of off-the-shelf computing nodes clustered. So, BOINC fits right in with the single owner supercomputing clusters. Watching the trends in the "PC" arena we are incorporating more and more of the elements os what mainframes and mini-computers had a generation or so back. I remember when if you wanted an FPU you had to buy it seperately. Same thing with cache memory. We have more computing power than the generations of yore in our personal machines, but each year the target moves! :) The only real issue (in Paul's opinion, for what that is worth) is the amount of I/O capability in the machines. The constant refrain that the mainframe is dead because PCs have "equivelent" CPU power almost always misses the point about the Mainframes (say, AS-400, etc.) and that is the amount of I/O capability, where PC class machines still lag. Mostly because PCs don't need that much I/O capability. And of course, the reliability issue ... It is always interesting to see the movement back and forth of some companies that migrate off of mainframes to PCs because they are "cheaper" to see them moving back a few years later ... Maybe I am just easily amused ... |
Benher Send message Joined: 25 Jul 99 Posts: 517 Credit: 465,152 RAC: 0 |
> Maybe I am just easily amused ... Yes Paul, but thats a GOOD thing ;) |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> > Maybe I am just easily amused ... > > > Yes Paul, but thats a GOOD thing ;) Yeah, I also noticed that the trend in each new computing arena has stuck with the one same error. Too small of a memory size. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.