| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 9
|
|
| Author |
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
I used the benchmark function in the Boinc agent on my desktop and here is what I get:
----------------------------------------System is: CPU is Intel I7 975 3.42 Ghz DRAM Memory DD3 tri-channel at 2000 MHZ Latency 9 W7 64bit Ultimate Benchmark values 8 CPU's 3'310 MFLOPS / CPU 8'370 Integer MIPS / CPU From this result I extrapolate that the CPU's global performance is: 26.48 GFLOPS So to reach the 1 TERAFLOP mythical value which is Supercomputers basic unit value, my little CPU must crunch non stop for 37.76 seconds. For one PETAFLOP then he must work for 10.5 hours, that is less then a day. Not bad at all. Let's say that with WCG Cobblestone metric I should get 100'000 points worth of computation every 37.76 seconds. In a full day of computation I should be rewarded with 3'813'559 points. If I look at the INTEL Export Compliance Table for processors where official GFLOP rating for each processor is published I have for the 975 the value of 55.34 GFLOPS at the stock rated 3.42 GHz speed, which is double the value of the Boinc performance test. I am aware that an official value must be a kind of theoretical maximum. But still we should be at 80% not at 50% of that value. Am I wrong somewhere? I would like to get your opinions on these values and calculations. ![]() |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Hi Hypernova,
----------------------------------------I'll ask the admin to move this to Chat Room. It's not a BOINC support issue. GridRepublic momentarily reports WCG as 531 Tfl http://www.gridrepublic.org/index.php?page=stats
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Am I wrong somewhere? Yes, at least twice (after only a quick reading of your post): 1. the equivalent of one teraflop is not 100,000 WCG points but 100,000 BOINC credits, i.e. 700,000 WCG points. 2. as already reported several times in other threads the BOINC benchmark for a hyperthreaded machine is far above what it should be with regards to the actual throughput of such processors; in other terms it is too close to the benchmark when running only one task per physical core. Cheers. Jean. |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Jean,
----------------------------------------if I understand you then let's take only 4 physical cores and forget the HT. Then we are at about 12 Gflops CPU performance. So let's halve everything in two. I would need 74 seconds to produce a Tflop, and 21 hrs to produce a Petaflop. If I apply your metric then the CPU that runs 7/7 24/24 should generate 700'000 WCG points per Tflop, in short about 14'000'000 points per day. We are very far from that. This means that when WCG units run on the CPU the throughput in terms of calculation is then very much lower than that. It's a pity that a theoretical performace of more then 50Gflops ends up into only a small fraction. I wonder what would be the reason for that. ![]() ![]() |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7850 Status: Offline Project Badges:
|
Then we are at about 12 Gflops CPU performance. Not 12 Gflops but 1.2 Gflops. Recalculate please. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Jean, if I understand you then let's take only 4 physical cores and forget the HT. Then we are at about 12 Gflops CPU performance. So let's halve everything in two. I would need 74 seconds to produce a Tflop, and 21 hrs to produce a Petaflop. If I apply your metric then the CPU that runs 7/7 24/24 should generate 700'000 WCG points per Tflop, in short about 14'000'000 points per day. We are very far from that. Of course we're very long from this, since you've made a big mistake... It's not 100k Cobblestones for 1 Tflop, but it's 100k Cobblestones for 86400 Tflop. Hmm, how did you get 74 seconds for 1 Tflop... But, if you're using the 74 seconds, you per day will get... 100k / 74 = 1351 Cobblestones/day, or 9500 WCG-points per day. More exactly, if WCG isn't weigting flops over iops, your computer would claim... (8 * 3.31G/1e9 + 8 * 8.37G/1e9) * 100 / 2 = 4672 Cobblestones/day. The granted will likely be somewhat lower than this in WCG... ![]() "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Yawn,
----------------------------------------Which project IS weighting iops / fpops that computes the credits straight from the combined whet/dhry benchmark, because that is what the credit claims are. Someone confusing flops with flop still.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Dec 19, 2009 8:18:13 AM] |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Thank's for all your comments, but I am completely lost now.
---------------------------------------- One thing is sure, there must be a way to relate flop+iop to WCG points. If Intel rates it's processor i7 975 at 55 Gflops, I see it difficult that the CPU running fully and exclusively on WCG units would produce such a low flop/iop rate. I will give it a long thought after all you said and will be back. ![]() ![]() |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Whetstone (fpops) and Dhrystone (iops) are just dumbly summed and divided by about
----------------------------------------1.26 quintillion floating-point operations Though I'm fairly sure the largest portion of their work is integer and lets my CPU run hot. And to celebrate TFlop, a bubble chart superimposed with the Daily CPU years printed on them. Each day they hang in a different place. (Someone hopes they all move far right to topple the tree. (He forgot the scale resizes itself) :D Have a Merry One, ![]() Edit: The more CPU years, the bigger the balls become.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 2 times, last edit by Sekerob at Dec 19, 2009 9:13:49 AM] |
||
|
|
|