| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: The Road to 600 TeraFLOPS: Need More Powah to stay Smarter Than Watson. |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 86
|
|
| Author |
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
I made a small test. Benchmarked with Sandra one of my 980X running at 4 Ghz with 12 GB DDR3 RAM running at 1833 Mhz.
----------------------------------------I got a Whetstone count of 127 GFlops. This machine burns 280 Watts. Result is a ratio of 445 MFlops/Watt. The top of the green list an IBM Blue Gene stands at 2026/Watt. Hokiespeed is at 926 MFlops/Watt. The average of the Top 500 Green list with CPU based machines is around 320 MFlops/Watt. So I am well above average. But if we take GPU based systems then the average is more than double at 788 MFlops/watt. There I am far behind. This is another argument in favor of GPU computing for WCG. If we want WCG to be green enough then we must push GPU computing.   |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
A little jaw dropper, but today we could be hitting the 700 TFL mark... the morning numbers suggest so. This would lift the 7 day average to 647 TFL.
Forward On --//-- |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
A little jaw dropper, but today we could be hitting the 700 TFL mark... the morning numbers suggest so. This would lift the 7 day average to 647 TFL. Forward On --//-- Have I been a little too conservative in my predictions. At that speed the 1 PFlops may be reached well before.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Seems it's not allowed to be... could someone please kneecap Murphy... another big validator slip of about 25 CPU years worth, this time from Monday to Tuesday. Still, a member from Sveden wanting to know how we're comparing to supercomputers, and doing now 640 TFL as a 21 day average and 648 TFL average for the last 7 days.
Crunch On. --//-- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
After the slight hick-up, really no stopping this machine for long, riding hi and on the half way mark to 700 TFL. The 7 day
All together we're doing now a rounded 405 CPU years as 21 day average. Same time last year that was 358 for 503 TFL. This gives the interesting comparison of then doing 1.41 TFL per contributed CPU year where now we're doing 1.59, a power per unit growth of 12.76% in a year. To think that my I7 at 75 W/H does 28 percent more work than the ol Q6600 at 170 W/H, there's a bunch more room for improvement and cut on the electricity bill in big digits [and CO2 footprint] Pretty cool numbers as we are in the 3rd month of the meteorological winter on the Northern Hemisphere. Spring starts in 3 weeks... time then to once again clean out those dust-bunnies. Crunch On. --//-- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Too good to be true... knew there was some spillage from Friday due a 7+ hour outage and validator halting on GFAM, but as much as the Saturday morning stats imply, I'd not thought.
Last 3 Saturday mornings: 186:174:04:30:52 Jan.21 182:033:01:08:11 Feb.04 181:006:11:20:45 Jan.28 This Saturday morning: 213:176:01:03:39 Feb.11 Will we be doing 400 CPU year today? Needing 187 in the afternoon and that number we've met and exceeded in the last 3 weeks too... on a lazy Saturday afternoon, winter 2012 on the Northern Hemisphere. Whilst, last week broke many overall records... ran at an average of 653 TFL... we're in FF mode.  Crunching On --//-- |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
I was crunching exclusively CEP2 an all my devices. My daily output fell to about 150 days runtime per day and 650'000 ppd. I have now switched fully to C4CW and my daily output raise to 260 days runtime per day and around a 1 million ppd.
---------------------------------------- I wonder if many crunchers left CEP2 for other projects like C4CW but also GFAM and DFSL which also have a much better performance in terms of output, than that could have a sizeable effect on the overall performance of WCG. Very cold temps in europe are also helping, as we have switched on to WCG heating system.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Maybe you build up a large number of PV? CEP2 is truly best run at 50% of cores [easiest achieved with a small cache setting] and last night ran this set up again [on Windows for alpha testing purposes] and had 99% efficiency as before, with GFAM/DSFL on the side.
Saturday ended up totaling 408 CPU years for 654 TFL, a second half of 195 CPU years, which is as they say in Amsterdam 'Vet' (Fat aka extra cool or something to that effect). --//-- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I was crunching exclusively CEP2 an all my devices. My daily output fell to about 150 days runtime per day and 650'000 ppd. I have now switched fully to C4CW and my daily output raise to 260 days runtime per day and around a 1 million ppd. Your observation is yet another drop in the vast pool of testimonies from crunchers confirming the weakness in the current-form of CEP2 WUs. If there is no other way to do solarEnergy-materials research on the grid other than what the current-form of CEP2 WUs is already doing, then I'm afraid that CEP2 is fighting a losing battle on the grid and dragging WCG's over-all output with it.I wonder if many crunchers left CEP2 for other projects like C4CW but also GFAM and DFSL which also have a much better performance in terms of output, than that could have a sizeable effect on the overall performance of WCG. As it happens currently, the crunching-resources (time, effort, money) consumed by CEP2 would have gone instead to other WCG projects that are equally worthy, equally important, and equally needed worldwide, but definitely more deserving of a cruncher's crunching-resources and those of WCG's.My bet is that WCG would have reached at least 700 TeraFLOPS by now had WCG had an efficient and a processor-intensive CEP2 WU (CPU-aware now, and GPU-aware later) rather than the ineffective/inefficient and I/O-intensive CEP2 WU that WCG has currently. ; |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
It's OPT-IN andzgrid, OPT-IN and very tunable to keep it down to below the half-core count mark, because *it is very well known* for the high hardware demands [not really very hi in practice]. Ran another test last night [with another alpha client], 4 concurrent and started staggered with 7 minute interval to get them passed the 2nd checkpoint, and same as the old 6.10.58 and 6.12.33 they ran 99.2% efficient on W7-64, and this is just a simple HT laptop doing GFAM on the other cores. Same is observed with 5+years old Q6600 on W7-64! 99+% efficiency!
----------------------------------------Moving on from [old] lament, we broke yesterday through the 7 day rolling 2900 CPU year mark, good for a 659 TFL mean. Will we be passing the 7 day 3000 CPU years before the Easter egg is laid? Let's Crunch On. --//-- edit; btw, those who read it all, will have seen that Q-Chem released a much faster version as per note in the cleanenergy log (and I'm sure it was the WCG programmer/techs who set them on the right track) To add, the credit per hour is huge on CEP2, substantially offsetting the time that the CPU cant work in waiting on disk IO! [Edit 1 times, last edit by Former Member at Feb 13, 2012 1:54:55 PM] |
||
|
|
|