Bruce AllenForum moderator
Project administrator
Project developer
Project scientist
Send message
Joined: Oct 15 04
Posts: 1094
Credit: 171,442,692
RAC: 2,294
Many thanks to the DC Russian Union for their ""Pulsar Rampage" computing challenge which helped to push Einstein@Home past the 800 Teraflop computing-power level for the first time in the history of the project!
Bruce Allen
Director, Einstein@Home
____________
I would suggest thanking SETI@home as well. They are having major database problems and have restricted users to 100 WU per GPU. There are also bandwidth problems making connection to the SETI servers difficult. As a result, many SETI GPU crunchers including myself have moved (temporarily?) to Einstein to keep their crunchers busy.
____________
My Team!
How is the computing power/floating point speed computed? Does every work unit have a fixed and known number of floating point operations? Doesn't this depend on the system that is used? And why are there different numbers on different BOINC web pages?
Is that a lot?
With POEM ceasing WU production, expect it to get even busier !
dunx
Bruce AllenForum moderator
Project administrator
Project developer
Project scientist
Send message
Joined: Oct 15 04
Posts: 1094
Credit: 171,442,692
RAC: 2,294
I would suggest thanking SETI@home as well. They are having major database problems and have restricted users to 100 WU per GPU. There are also bandwidth problems making connection to the SETI servers difficult. As a result, many SETI GPU crunchers including myself have moved (temporarily?) to Einstein to keep their crunchers busy.
This is a good point, but the fact that SETI@Home is having database and bandwidth problems isn't something I can feel happy about. Over the years the people behind SETI@Home have given us a lot of technical and intellectual support, and I am sorry that they are having these issues.
In any case, I am very grateful to the SETI@Home community for supporting Einstein@Home, and I hope that many of you continue to provide that support even after SETI@Home resolves its technical problems!
____________
Bruce AllenForum moderator
Project administrator
Project developer
Project scientist
Send message
Joined: Oct 15 04
Posts: 1094
Credit: 171,442,692
RAC: 2,294
Is that a lot?
Yes! To put this in perspective, it was only four years ago (June 2008) that a computer broke the 1 Petaflop (1000 Teraflop) barrier. Einstein@Home is now very close to that computing speed. What's more, Einstein@Home is delivering this performance on a real-world problem, not on an artificial "Linpack" test code.
____________
I would suggest thanking SETI@home as well. They are having major database problems and have restricted users to 100 WU per GPU. There are also bandwidth problems making connection to the SETI servers difficult. As a result, many SETI GPU crunchers including myself have moved (temporarily?) to Einstein to keep their crunchers busy.
This is a good point, but the fact that SETI@Home is having database and bandwidth problems isn't something I can feel happy about. Over the years the people behind SETI@Home have given us a lot of technical and intellectual support, and I am sorry that they are having these issues.
In any case, I am very grateful to the SETI@Home community for supporting Einstein@Home, and I hope that many of you continue to provide that support even after SETI@Home resolves its technical problems!
I fully agree and my post was not meant to be taken too literally, just to point out that a lot of seti crunchers use Einstein as a backup project and that means that over the last 2-3 weeks they have been significantly increasing the amount of Einstein they crunch. I hope that seti resolves at least some of its problems soon so that I can take my main cruncher back to seti which is where I intended it to be. In the meantime it is hard at work crunching Einstein on CPU and GPU for you.
Hello, Bruce.
You could, to add to your 800 petaflops, have 75% of my true quadcore i7 when I'm not using it, and 25% when I am--if only someone will reply to my post on the Problems page.
Pieter Watson
Hello again.
Never mind my previous post: Jord has replied to me. I assumed that replies would appear when I viewed my messages, but evidently they don't.
Sorry for hassling you.
Pieter
With Seti@home off-line until next tuesday, we can make it to the 1Petaflop mark!
____________
In any case, I am very grateful to the SETI@Home community for supporting Einstein@Home, and I hope that many of you continue to provide that support even after SETI@Home resolves its technical problems!
Hi Bruce and thanks for all who remembers us, we are very happy to be here and give a hand to the E@H cause. Of course we are expecting the SETI problems will fix ASAP but until then you could count with all of our "firepower". And even some more from few new mates that are comming soon.
I just curious, and probabily that is a well know fact in the E@H comunitiy and i know could find the answer in some place, just don´t find where, there is any place that i could easely find whow many FLOPS (TERA i don´t expect but maybe in the range of GIGA) our team is doing for the E@H cause?
And congrats for the 800Mark, but that was part of the past, the last number show by the server status page is: 868.1 TERAFLOPS (maybe a new record?)
I expect our team will break the 2MM daily credit mark today (credit x FLOPS i have no ideia if have any conversion rate), hope that will add some more FLOPS to that mark. And that number could be better, only I have Pending credit: 10,176.39 in allmost 1900 WU, and i imagine the others are in the same range.
Congratulations on breaking 800 Teraflops!
And the kitties helped!!!
Meows.
And the kitties helped!!!
Meows.
With some more help from the kitties, E@H reaches 872.9 TFLOPS (and rissing) showed now in the servers status screen!
I realy start to think E@H likes our hungry hosts...
I've a question, what would it cost roughly to operate an 800 TF supercomputer?
I've a question, what would it cost roughly to operate an 800 TF supercomputer?
Megabucks per year in power costs ( including cooling ) for a start.
[ We're at 892.5 as I speak .... :-) ]
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
Although significant, I don't imagine power is the 'biggie' in the mix. You need infrastructure to house it, people to manage/operate it, maintenance contracts to keep it all going, a hardware/software budget to keep it up-to-date and consultants fees to make sure you are able to convince those who supply all the funds that the money is being spent efficiently :-). I'm sure there are other 'things in the mix' as well.
This is what makes the concept of BOINC so brilliant. It turns out there are a whole host of people out there willing to work all sorts of hours, just to provide ALL the above for ... a bunch of 'cobblestones' :-).
Now who woulda thunk that!! :-).
____________
Cheers,
Gary.
Hope we could help in the final push tomorrow and break the 900 TeraFlops mark before SETI returns on next tuesday. We still trying to do our best for the E@H cause, today the production of our "litte" team reaches the 2,852,288 daily production mark and 2 more friends from our SETI main team join us now at the end of the night, so i hope we could reach the 3MM mark tomorrow.
Of course there is a list of members here since the beginning and will continue to be here 24/7 as long as this project exists regardless what other projects are doing.
____________
Of course there is a list of members here since the beginning and will continue to be here 24/7 as long as this project exists regardless what other projects are doing.
I am sure that is quite true.
But some of us are also quite happy to have chosen Einstein as our backup when our preferred project is down.
I suppose it really is a "fair" deal. We give them widgets, they give us cobblestones/kibbles/popcorn..
Oh who am I kidding, but as long as it is useful why not? Hope to see you to 900 tomorrow.
____________
Please don´t take my words wrong, i realy like the time we pass here, and realy hope we gave E@H a little help, going to sleep now, hope to hear the good news of the new record when i wake-up, we are just 3.8 TF to break the 900TF mark!
Well one teraflop is as good as any other! :-)
898.1 as I write .....
Cheers, Mike.
[ edit ] 898.7
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
According to the info here (date 24.02.2012), germans fastest supercomputer makes 831 teraflops with initial cost of 22,5 million euros and running costs of 2 million euros annual.
Don't know how excact the flops calculation based on RAC is, but in the end, a cool performance for some "Home-Computer" :-).
(at the moment 899.0 TeraFlops on status page)
____________
Bikeman (Heinz-Bernd Eggenstein)Forum moderator
Project administrator
Project developer
Send message
Joined: Aug 28 06
Posts: 3019
Credit: 46,645,001
RAC: 59,781
Hi!
Just for the records: since then, the supercomputer described in the article above has lost its throne as Germany's Top Supercomputer to this impressive installation:
http://en.wikipedia.org/wiki/SuperMUC
Cheers
HB
____________
899.5
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
Hi!
Just for the records: since then, the supercomputer described in the article above has lost its throne as Germany's Top Supercomputer to this impressive installation:
http://en.wikipedia.org/wiki/SuperMUC
Cheers
HB
Out of curiosity, how does MPI-AEI's own Atlas compare on that scale?
We just reached it! =D
Floating point speed
(from recent average credit of all users) 900.3 TFLOPS
____________
Team Belgium
Congrats friends from E@H the 900.8 TFLOPS mark was breaked!
I´m happy we are here with you all to celebrate!
Just too early here to open the celebration beer! But later that is guaranted!
Somebody needs to change the thread title...!!!
Out of curiosity, how does MPI-AEI's own Atlas compare on that scale?
Just go to the http://www.top500.org/ and find it in the Top 500 lists. They show the Flops, Power Hungry, and some more interesting features.
BTW a 900 TERAFLOPS is almost equivalent to the Rmax (TFlop/s) of the #26 world biggest supercomputer by that list and uses about 2800.00 kW to do it´s job.
Hallo!
Amazing!!!
Today at about 11:35(UTC) (interpeolated) we reached 900 TFLOPS and it´s still increasing at a rate of impressive 23,7 +/- 0,6 TFLOPS/day. If this holds, we will have 1 PFLOPS to the end of this week. Hopefully !!!!!!!
Kind regards and happy crunching
Martin
____________
On that basis, we should be past Russia's fastest TOP500 entry shortly (901.9 TFLOPs).
Yes, but you can´t realy compare two systems just by the TERAFLOPS, but that´s gives us "simple mortals" a good way to do that.
BTW i ask before but no one answer, is there any place i could see how many MEGA or GIGA Flops my hosts or team are realy doing? Something like a credit x flops conversion table? Thanks in advance for any help.
BTW i ask before but no one answer, is there any place i could see whow many MEGA or GIGA Flops my hosts or team are realy doing? Something like a credit x flops conversion table? Thanks in advance for any help.
The BOINC credit is the Cobblestone, whereby a 1,000 MIPS/1 GFlOp machine doing 24 hours of computation will earn 200 credits. This works out to 432 billion ops. = 1 BOINC credit. Computing is so cheap these days, no?
So you can get your GFlOps by dividing your RAC by 200... You're doing just over a teraflop.... nice. :^)
Happy to have brought my piddly fifth of a teraflop to this project while SAH is down. Seems to be the go-to secondary for the SAH crunchers (so similar... even uses Arecibo data.) We'll break a petaflop yet.
BTW i ask before but no one answer, is there any place i could see how many MEGA or GIGA Flops my hosts or team are realy doing? Something like a credit x flops conversion table? Thanks in advance for any help.
E@H seem to use RAC / 100000:
http://einstein.phys.uwm.edu/forum_thread.php?id=9639
Thanks... I'll have a look at that. Thought it was universal <<shrug>>.
____________
BTW i ask before but no one answer, is there any place i could see how many MEGA or GIGA Flops my hosts or team are realy doing? Something like a credit x flops conversion table? Thanks in advance for any help.
E@H seem to use RAC / 100000:
http://einstein.phys.uwm.edu/forum_thread.php?id=9639
So you can get your GFlOps by dividing your RAC by 200... You're doing just over a teraflop.... nice. :^)
Thanks... so i´m happy i almost made my first TERAFLOP for the E@H cause!.
Still thinking somebody needs to change the thread title... 800 is past!
Congrats for 903.4 TFLOPS !!!
Bruce AllenForum moderator
Project administrator
Project developer
Project scientist
Send message
Joined: Oct 15 04
Posts: 1094
Credit: 171,442,692
RAC: 2,294
So you can get your GFlOps by dividing your RAC by 200... You're doing just over a teraflop.... nice. :^)
Thanks... so i´m happy i almost made my first TERAFLOP for the E@H cause!.
Still thinking somebody needs to change the thread title... 800 is past!
I'm quite excited about this -- I am thinking about adding another news items for > 900 Tflops. I hope we can pass one Petaflop (1000 Teraflops) before the end of the year!
____________
Out of curiosity, how does MPI-AEI's own Atlas compare on that scale?
Just go to the http://www.top500.org/ and find it in the Top 500 lists. They show the Flops, Power Hungry, and some more interesting features.
Well, I think I was asking because even by the time of the Open Day and guided tours in July last year, Atlas had already outgrown its 2008 build specification of 30 TeraFlops. They were adding Tesla nodes to supplement the original CPU-only design around the time we were there, and were planning to double the cluster's size by taking over the empty basement room next door.
So my curiosity was more to do with the status of those expansion plans.
I'm quite excited about this -- I am thinking about adding another news items for > 900 Tflops. I hope we can pass one Petaflop (1000 Teraflops) before the end of the year!
Don't forget that tomorrow (Dec. 4th ETA) SETI@Home will be reopening and a lot of the big crunchers will be returning to there.
Sorry to be a downer. :^)
____________
I'm quite excited about this -- I am thinking about adding another news items for > 900 Tflops. I hope we can pass one Petaflop (1000 Teraflops) before the end of the year!
Don't forget that tomorrow (Dec. 4th ETA) SETI@Home will be reopening and a lot of the big crunchers will be returning to there.
Sorry to be a downer. :^)
But i´m sure our "pendings" (i already have more than 10k pendings in more than 2K WU > 1MM in pendigs credits) will continue to add some more TERAFLOPS to the E@H cause for at least few days... and with the 100WU limit, our battleships will sure continue to crunch a lot of more E@H WU... helping the E@H cause!
I'm quite excited about this -- I am thinking about adding another news items for > 900 Tflops. I hope we can pass one Petaflop (1000 Teraflops) before the end of the year!
Maybe a new thread for the 900 Tflops is a better ideia.
But i´m sure our "pendings" will continue to add some more TERAFLOPS to the E@H cause for few days...
I have enough work from here for about a week (due to the 100 work share you indicated... heh) and I'll let it all complete.
Ideally, SETI@Home and Einstein@Home would be semi-unified as they are both using Arecibo data. It would be nice if every work unit on both projects was checked for everything (ie both narrowband ETI signals, pulsars and whatever else.) It's all good discovery!
____________
I'm quite excited about this -- I am thinking about adding another news items for > 900 Tflops. I hope we can pass one Petaflop (1000 Teraflops) before the end of the year!
Don't forget that tomorrow (Dec. 4th ETA) SETI@Home will be reopening and a lot of the big crunchers will be returning to there.
Sorry to be a downer. :^)
Well, that is an estimate. And when things do come up, I would not be surprised at all to see some difficulties arise.
The servers and the bandwidth are going to be very congested with all the empty hosts clamoring for work.
All my rigs are on auto...and I suspect they shall still be doing some EAH for a week or two if they can't get enough Seti to keep them occupied.
All my rigs are on auto...and I suspect they shall still be doing some EAH for a week or two if they can't get enough Seti to keep them occupied.
Now that I finally know about the zero workshare for secondary projects -- thanks Juan... only took me ten years to find that out. ;^) -- I'll be doing the same and leaving both projects active so I don't have to keep manually switching back and forth, oh, and running out of work!
Yeah that first couple of days back will be rough so it helps having a week's work here to keep busy with.
____________
Out of curiosity, how does MPI-AEI's own Atlas compare on that scale?
Just go to the http://www.top500.org/ and find it in the Top 500 lists. They show the Flops, Power Hungry, and some more interesting features.
Well, I think I was asking because even by the time of the Open Day and guided tours in July last year, Atlas had already outgrown its 2008 build specification of 30 TeraFlops. They were adding Tesla nodes to supplement the original CPU-only design around the time we were there, and were planning to double the cluster's size by taking over the empty basement room next door.
So my curiosity was more to do with the status of those expansion plans.
I'd forgotten about that! Nearby the cage with the tape robot inside. Yeah, what's the go with that?
Can anyone remember what E@H's TFlops was back then? I think we've at least tripled since.
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
I'd forgotten about that! Nearby the cage with the tape robot inside. Yeah, what's the go with that?
Can anyone remember what E@H's TFlops was back then? I think we've at least tripled since.
Cheers, Mike.
The Wayback machine says:
Edit - and to answer the other part of the question, Atlas was rated at #58 in the top 500 in June 2008, when it had 5,420 CPU cores. The Linpack benchmark hasn't been updated since, and Atlas last appeared at #415 in the November 2010 list.
So a tad over doubled ( since mid 2011 ) then, thanks Richard. :-)
Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
I personally believe an official news post would be very appropriate to mark your 900T threshold.
____________
Floating point speed 915.9 TFLOPS and still rissing!
Floating point speed 915.9 TFLOPS and still rissing!
Translation, Seti@home is still down ...
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2971
Credit: 76,517,893
RAC: 34,003
For the past few days we have been granting enough credit to correspond to a Petaflop/s. It's just a matter of days for the gliding average to show this on the server status page.
Regarding ATLAS:
- The tape library has been moved to the basement in the office building; in fact it should be sitting right under my chair one floor down (we moved to a larger office in the corner of the building to accommodate Heinz-Bernd, too).
- The Linpac tests for the TOP500 list were made with 5368 compute cores. Atlas currently has 6720 cores in 1680 nodes
- Add 66 GPU nodes with 4 CPU cores and 4 NVIDIA Tesla cards (C1060 and C2050) each. Due to various software issues the GPUs are not used to full extent yet. In addition these nodes are the first to be shut down in case of problems with the cooling, which we had a lot this year.
- The extension ("Bauabschnitt #2") currently consists of empty racks. The cooling facility originally planned for this extension is currently cooling the older part, because the original cooling facility never fully worked. A "final" replacement is planned for 2Q next year (IIRC).
BM
Edit: by far the most increase in computing power comes from GPUs on the BRP search. It has more than doubled in the past ~2 months.
Regarding ATLAS:
With the updates do you know what is the actual capacity in TF?
Just as curiosity too, 1,000,000 E@H in credits is equivalent to about 5TF?
Congrats for the upcoming PenthaFlops mark break, realy amazing, and congrats, your servers realy have the capacity to feed ours hungry hosts. You guys in the lab make an excelent work too.
Floating point speed 915.9 TFLOPS and still rissing!
Translation, Seti@home is still down ...
It's back up now.
____________
Our small team brake the mark of 4MM of daily credit production, about 20TFlops to help our E@H friends. Hope that makes the 1PFlops mark a little closer...
____________
946.8 TFLOPS :-)
Cheers, Mike
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
First time we hit the border of 950 TFLOPS at 13:46 (UTC) (interpolated) today at a rate of 0,36[TFLOPS/h]. But tonight it will drop below this level, because of this peculiar daily sawtooth like behaviour of the total crunching power. About 1/2h before midnight (UTC) it drops by up to 5%. (I would like to know why this is.)
Let´s see when we will pass the 1 PFLOS! At the next weekend? Let´s hope!!!
Kind regards and happy crunching
Martin
____________
First time we hit the border of 950 TFLOPS at 13:46 (UTC) (interpolated) today at a rate of 0,36[TFLOPS/h]. But tonight it will drop below this level, because of this peculiar daily sawtooth like behaviour of the total crunching power. About 1/2h before midnight (UTC) it drops by up to 5%. (I would like to know why this is.)
Let´s see when we will pass the 1 PFLOS! At the next weekend? Let´s hope!!!
Kind regards and happy crunching
Martin
We will pass it - by Tuesday
Our small team reaches almost another 4MM of daily credit production today, about 20TFlops more adeed to the E@H cause. The 1PFlops mark is a little closer... Hope our E@H friends apreciate that and our enemies hate! (if we have any of course, i don´t hope so, who knows?)
____________
I made a graph yesterday, but the status page was out of date most of the time tonight:
http://crna.zapto.org/rrd/einstein/
I could do graphs of any other parameter as well if desired.
Of course there is a list of members here since the beginning and will continue to be here 24/7 as long as this project exists regardless what other projects are doing.
I am sure that is quite true.
But some of us are also quite happy to have chosen Einstein as our backup when our preferred project is down.
What is so brilliant about BOINC is that even those who have only completed ONE workunit has added to the cause!