1Pflops milestone
Message boards :
News :
1Pflops milestone
Message board moderation
| Author | Message |
|---|---|
 GDFSend message Joined: 14 Mar 07 Posts: 1958 Credit: 629,356 RAC: 0 Level  Scientific publications     |
GPUGRID has been constantly above the Petaflop in the past few days. Although just a symbolic number it feels good to reach such an impressive target for us. Every volunteer should feel proud too, because it's all thanks to you. gdf |
 Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level  Scientific publications  |
Congratulations! That might be news worthy of pushing them out via the BOINC message system. Although I'd generally advise to use it cautionally. MrS Scanning for our furry friends since Jan 2002 |
|
AdamYusko Send message Joined: 29 Jun 12 Posts: 26 Credit: 21,540,800 RAC: 0 Level  Scientific publications
|
Awesome! I guess there are far fewer GPU Grid cruchers, but with the abundance of top notch GPU's on this project I would have thought GPU grid has been above it for quite some time. Glad I could help, though I only have a GT 640 and GT 440 OEM crunching now. Can not wait for Short runs to return so I can put my 440 on those. 
|
|
Send message Joined: 9 Dec 08 Posts: 1006 Credit: 5,068,599 RAC: 0 Level  Scientific publications |
In fact we have so much incoming data that it's becoming hard for the server to keep up. We are working around the clock to keep the pipes clear. |
|
Send message Joined: 19 Dec 10 Posts: 1 Credit: 421,995,530 RAC: 0 Level  Scientific publications
|
Congratulations for all crunchers, such a big milestone! |
|
Send message Joined: 5 Dec 11 Posts: 147 Credit: 69,970,684 RAC: 0 Level  Scientific publications
|
In fact we have so much incoming data that it's becoming hard for the server to keep up. We are working around the clock to keep the pipes clear. Generally one of the nicer problems to have :) I imagine that the new(ish) 4.2App had a fair bit to do with getting this project over the 1PF line with it's massive increase in efficiency. |
|
Send message Joined: 17 Feb 13 Posts: 181 Credit: 144,871,276 RAC: 0 Level  Scientific publications
|
Hi, Folks: Congratulations to all on this achievement. I am impressed, in particular by GPUGrid's showing the scientific papers to which crunchers have contributed. This really makes one feel part of something useful. John |
|
Send message Joined: 4 Oct 12 Posts: 53 Credit: 333,467,496 RAC: 0 Level  Scientific publications
|
Woohoo - Congrats GPUGrid - broken the Petaflop barrier again! Clearly a result of the recent: Application stability improvements. Smaller work unit sizes. Improved communication - keeping crunchers motivated. Good work all - roll on 2 Petaflops! |
|
Send message Joined: 5 Jan 09 Posts: 670 Credit: 2,498,095,550 RAC: 0 Level  Scientific publications 
|
WooHoo indeed. I hope we can make it to 2 Peta Flops with the Titan and 780's now fully on board. I would respectfully suggest some aggressive PR and a comprehensive FAQ (constantly updated) which can hold the hands of new users who are not as computer savvy as crunchers at GPU Grid. If that's done I see no reason why GPU Grid should not hit 2 Peta Flops and more... PS What is the speed of a current super computer anyway? |
skgiven Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level  Scientific publications  |
33.86 petaFLOPS is the present fastest non-distributed supercomputer. Ref http://en.wikipedia.org/wiki/Supercomputer Folding@Home is around 18 petaFLOPS. Ref. http://en.wikipedia.org/wiki/Folding@home Boinc combined is presently 7.6 petaFLOPS. Ref. http://boincstats.com/en/stats/-1/project/detail FAQ's HOW TO: - Opt out of Beta Tests - Ask for Help |
|
Send message Joined: 5 Jan 09 Posts: 670 Credit: 2,498,095,550 RAC: 0 Level Scientific publications
|
33.86 petaFLOPS is the present fastest non-distributed supercomputer. Thanks SK. We're on a par with the IBM RoadRunner (2008) not bad! |
Chilean Send message Joined: 8 Oct 12 Posts: 98 Credit: 385,652,461 RAC: 0 Level Scientific publications
|
*Group hug* Now, back to crunching. 
|
|
Send message Joined: 8 Mar 12 Posts: 411 Credit: 2,083,882,218 RAC: 0 Level Scientific publications
|
So, just wondering: How much GigaFLOPS does say a 780 put out for this website's app? Meaning, if a 780 were left running at stock 24/7, how much would one increase the sites performance? |
|
Send message Joined: 5 Jan 09 Posts: 670 Credit: 2,498,095,550 RAC: 0 Level Scientific publications
|
So, just wondering: How much GigaFLOPS does say a 780 put out for this website's app? Meaning, if a 780 were left running at stock 24/7, how much would one increase the sites performance? With nothing holding it back it should be 4 Tera Flops (or just under) of single precision FPO which is what GPU Grid uses. |
|
skgiven Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level Scientific publications |
The theoretical single precision GFlops of the main Reference spec cards are shown below. GeForce Card GFLOPS (SP) GTX 650 813 GTX 650 Ti 1421 GTX 650 TiBoost 1505 GTX 660 1882 GTX 660 Ti 2460 GTX 670 2460 GTX 680 3090 GTX 690 5622 GTX 760 2258 GTX 770 3213 GTX 780 3977 GTX Titan 4500 For this project most of these GFlops values reflect the relative GPU performances reasonably accurately, but the accuracy of the Ti cards especially isn't great (some higher, some lower). The accuracy of non-listed entry level cards with miniscule bus widths are even worse. Application and system constraints would reduce the acutal GFlops. The power targets/GPU Usage might better reflect the actual amount of GFlops being used. So a card using 90% of the GPU and running at a power target of 90% might only be using up to 90% of the available GFlops. In my experience mid-range reference cards have a high GPU usage compared to higher end cards and would therefore come closer to reaching their reference GFlops - perhaps 90% of the theoretical GFlops would be used (on a good setup). High end cards tend to use less of what's available (say 75 to 85% GPU usage, at least on Windows). The % usage of the TDP's tends to be lower too. Values vary depending on task type. Another consideration is that many cards are non-reference with higher TDP's and clocks. - Good to see GPUGrid back over the 1Peta Flops mark, but I'm not sure where this is taken from or of it's accuracy; Boinc All Project Stats has the value at 2,810.281 TeraFLOPS (2.8 PFlops), and the project RAC at 214M. http://www.allprojectstats.com/po.php?projekt=48 Boinc Stats has the current GFlops at 968.609 TeraFLOPS. FAQ's HOW TO: - Opt out of Beta Tests - Ask for Help |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level Scientific publications |
In addition to what SK said: for GP-GPU it's very good to achieve even 50% of the theoretical maximum throughput. I don't know any more specific numbers for GPU-Grid. MrS Scanning for our furry friends since Jan 2002 |
|
Send message Joined: 11 Jul 09 Posts: 1639 Credit: 10,159,968,649 RAC: 326,008 Level  Scientific publications
|
- Good to see GPUGrid back over the 1Peta Flops mark, but I'm not sure where this is taken from or of it's accuracy; Boinc All Project Stats has the value at 2,810.281 TeraFLOPS (2.8 PFlops), and the project RAC at 214M. http://www.allprojectstats.com/po.php?projekt=48 No information about FLOPS (peak or actual) is fed out from BOINC projects to the stats aggregation sites. Every FLOPS figure you see on a stats site has been derived in some way from the credit awarded by the project. For example, the current figures at Boinc Stats are: Recent average credit RAC 193,721,774 Average floating point operations per second 968,608.9 GigaFLOPS The RAC figure is 199.9999938 times the GigaFLOPS figure: that's not coincidence, that's arithmetic (the definition is 200x, with a tiny rounding error because of limited precision). So, according to BOINC statistics, all a project has to do to increase its G/T/PFlops production rate is to increase the amount of credit awarded per task - and there are some projects which have done that. That's the crazy world of statistics for you. |
|
Chilean Send message Joined: 8 Oct 12 Posts: 98 Credit: 385,652,461 RAC: 0 Level Scientific publications
|
I wonder how much "science" can be done, say per month, at this speed.
|
|
Send message Joined: 16 Mar 11 Posts: 509 Credit: 179,005,236 RAC: 0 Level  Scientific publications
|
So, according to BOINC statistics, all a project has to do to increase its G/T/PFlops production rate is to increase the amount of credit awarded per task - and there are some projects which have done that. That's the crazy world of statistics for you. It's not statistics that are at fault, it's the way some people do statistics and what they do with the numbers they come up with. I once had respect for BOINC stats sites but that quickly ended when they started doing crazy stuff like deriving performance (FLOPS) from credits. They pander to the naive, like many politicians do, thereby reinforcing beliefs that have no foundation in reality. BOINC <<--- credit whores, pedants, alien hunters |
|
Send message Joined: 11 Jul 09 Posts: 1639 Credit: 10,159,968,649 RAC: 326,008 Level Scientific publications
|
So, according to BOINC statistics, all a project has to do to increase its G/T/PFlops production rate is to increase the amount of credit awarded per task - and there are some projects which have done that. That's the crazy world of statistics for you. And the real crime against scientific method is that BOINC themselves commit the same sin, in the top-right corner of the home page, and more dramatically on http://boinc.berkeley.edu/chart_list.php. How many of those GFLOPS statements would pass a floating-point benchmark audit? |
©2025 Universitat Pompeu Fabra