Astropulse units too big?

Message boards : Number crunching : Astropulse units too big?
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
jethot

Send message
Joined: 9 Sep 02
Posts: 1
Credit: 283,277
RAC: 0
Portugal
Message 800379 - Posted: 21 Aug 2008, 9:50:35 UTC

Hi all
My AMD x2 machine takes like 80 hours to process a single Astropulse unit. Shouldn't the units be made a bit smaller? It would make me a bit less anxious ;)
Regards
Rui
ID: 800379 · Report as offensive
wayne0308

Send message
Joined: 7 Sep 07
Posts: 3
Credit: 286,359
RAC: 0
Ireland
Message 800383 - Posted: 21 Aug 2008, 10:18:42 UTC - in response to Message 800379.  

Hi all
My AMD x2 machine takes like 80 hours to process a single Astropulse unit. Shouldn't the units be made a bit smaller? It would make me a bit less anxious ;)
Regards
Rui


I have to agree, I am just finishing my third AP workunit. The first two AP work units I did have been completed for quite some time now, and the other person (two people need to complete the same WU) has not completed their side of things yet and theres only a few days left.

I have a feeling that people are just deleting them or ignoring them and that will mean that I've wasted a good 8 days processing time.

I've subsequently stopped any more AP work units coming to my computers until these three WU's have been verified completed by the other people involved.

I think if they were smaller then people wouldn't be so inclined to delete/ignore them. Because as it stands there's no benefit to anyone to compute these AP WU's, if they are simply ignored.
ID: 800383 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 800393 - Posted: 21 Aug 2008, 11:56:00 UTC

If you consider the increase in performance of computers and the history of Seti over the years, you will probably come to the conclusion they are about the right length.
A good computer when Classic started took ~20 hrs to complete a unit, and unless you were rich you only had a single core. Now on AP there a common computers that can do four units in under 30 hrs at the same time and therefore turn out a unit every 8 hrs. So in a couple of years it may be common for a computer to upload a unit every 2 hours, less if the 'brains' can optimise the app.

Unfortunately for AMD users the processing time is over twice as long as an Intel core2.
ID: 800393 · Report as offensive
PhonAcq

Send message
Joined: 14 Apr 01
Posts: 1656
Credit: 30,658,217
RAC: 1
United States
Message 800398 - Posted: 21 Aug 2008, 12:25:05 UTC

AMD 2x slower than Core? Is that due to some Intel specific libraries or something?

I think the number crunchers generally would like to see more action than one or fewer completed wu's per day. Have the developers considered using multiple threads in the science app for ap? Most number crunchers probably have a core 2 duo or quad by now and would be able to offer up parallel processing to the project.
ID: 800398 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 800431 - Posted: 21 Aug 2008, 15:44:57 UTC
Last modified: 21 Aug 2008, 15:48:59 UTC

Intel being faster than AMD in these kinds of scenarios are due to Intel's large L2, and that the L2 is shared amongst the cores. About 90% of the crunch time can be decided with the built-in benchmark of BOINC. Integer math doesn't do anything for crunching, it's the floating point benchmark that matters.

AMD CPUs are more efficient in their design, but Intel's large L2 and extremely high FSB are what make the difference here.

I posted about this back in...earlier this year, a pretty detailed breakdown of the difference between the two.

Found it.. http://setiathome.berkeley.edu/forum_thread.php?id=45616&nowrap=true#721263

Hope this answers your question.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 800431 · Report as offensive
Profile AlphaLaser
Volunteer tester

Send message
Joined: 6 Jul 03
Posts: 262
Credit: 4,430,487
RAC: 0
United States
Message 800436 - Posted: 21 Aug 2008, 15:55:30 UTC - in response to Message 800398.  

AMD 2x slower than Core? Is that due to some Intel specific libraries or something?

I think the number crunchers generally would like to see more action than one or fewer completed wu's per day. Have the developers considered using multiple threads in the science app for ap? Most number crunchers probably have a core 2 duo or quad by now and would be able to offer up parallel processing to the project.


AFAIK, BOINC itself does not support multithreading within a single app. It will still try to run n tasks to run on n cores, even if a single task uses two or more threads of processing. As a result, a project could monopolize CPU cores and BOINC's debts mechanism would go haywire. Multithreading would require several changes to the CPU scheduler of the BOINC client.
ID: 800436 · Report as offensive
Profile SATAN
Avatar

Send message
Joined: 27 Aug 06
Posts: 835
Credit: 2,129,006
RAC: 0
United Kingdom
Message 800456 - Posted: 21 Aug 2008, 17:09:04 UTC

I'd just like to get a unit. then I can join in.

No fair.
ID: 800456 · Report as offensive
Profile arkayn
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4438
Credit: 55,006,323
RAC: 0
United States
Message 800457 - Posted: 21 Aug 2008, 17:17:34 UTC - in response to Message 800456.  

I'd just like to get a unit. then I can join in.

No fair.


Now that he has the correct software installed to work on one.

ID: 800457 · Report as offensive
PhonAcq

Send message
Joined: 14 Apr 01
Posts: 1656
Credit: 30,658,217
RAC: 1
United States
Message 800484 - Posted: 21 Aug 2008, 18:54:53 UTC
Last modified: 21 Aug 2008, 18:55:07 UTC

Be careful of what you ask for...
ID: 800484 · Report as offensive
Profile Edward Lee Michau
Avatar

Send message
Joined: 31 Jul 06
Posts: 138
Credit: 9,640,846
RAC: 0
United States
Message 800497 - Posted: 21 Aug 2008, 19:33:19 UTC

My 2003 Compac got an Astropulse unit said 67hours to completion. It took 89+ hours to actually run. Someone else must have run it, because it is not still
in my pending units. I dont mind running them, but I wish the estimated time was closer to the actual time.
Do not live your life for anyone else. Do not allow anyone to live their lives for you. Ayn Rand
ID: 800497 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 800505 - Posted: 21 Aug 2008, 20:20:45 UTC

My Opteron 2210s show that an AP unit should take ~68 hours, and it takes just over twice that long in actuality. The four that I've run so far have been 128-154 hours.

My Celeron 1.8 says 130 hours and completes in about 165, so at least that's a little closer.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 800505 · Report as offensive
Profile Richard Walton

Send message
Joined: 21 May 99
Posts: 87
Credit: 192,735
RAC: 0
United States
Message 800513 - Posted: 21 Aug 2008, 20:37:19 UTC - in response to Message 800505.  

My Opteron 2210s show that an AP unit should take ~68 hours, and it takes just over twice that long in actuality. The four that I've run so far have been 128-154 hours.

My Celeron 1.8 says 130 hours and completes in about 165, so at least that's a little closer.


My first completed AP was predicted for around 62 hours and took 67. That is not too bad.
My computer is a Toshiba satellite running Intel(R) Core(TM)2 CPU T5500 @ 1.66GHz [x86 Family 6 Model 15 Stepping 6] with Vista, uing BOINC 6.2.16. It is a long grind for a laptop though. . .
ID: 800513 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 800525 - Posted: 21 Aug 2008, 21:25:53 UTC - in response to Message 800513.  

It is a long grind for a laptop though. . .

A long grind is the work I make that Celeron go through. Toshiba Satellite 2410-S185. Yeah, 2003 era. I run SETI 24/7 on it at 100%. It's been going about three years non-stop now.

It takes long enough that when it does get an AP unit, I don't have any other work until about 75%.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 800525 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 800531 - Posted: 21 Aug 2008, 21:50:05 UTC - in response to Message 800379.  

Hi all
My AMD x2 machine takes like 80 hours to process a single Astropulse unit. Shouldn't the units be made a bit smaller? It would make me a bit less anxious ;)
Regards
Rui

It's not possible with out loosing some algorithm sensitivity.
But it's possible to cut processing times itself by means of algorithm/app optimisation (or OCing/replacing hardware ;) )
ID: 800531 · Report as offensive
Mats Nilsson

Send message
Joined: 22 Oct 05
Posts: 26
Credit: 111,298
RAC: 0
Sweden
Message 800708 - Posted: 22 Aug 2008, 6:34:21 UTC

Got my first ap the computer said about 30hrs when it started but looks more like 120 on my AMD 3500+. This meens that this will be the only ap unit I will run because I split my time 50/50 between Seti and Einstein. I will probably have too run the ap unit att panic in the end to complete it in time. So I will stop running ap until a faster app becomes availeble and without knowledge how too do it I have too wait.
ID: 800708 · Report as offensive
1mp0£173
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 8423
Credit: 356,897
RAC: 0
United States
Message 800712 - Posted: 22 Aug 2008, 6:47:09 UTC - in response to Message 800708.  

Got my first ap the computer said about 30hrs when it started but looks more like 120 on my AMD 3500+. This meens that this will be the only ap unit I will run because I split my time 50/50 between Seti and Einstein. I will probably have too run the ap unit att panic in the end to complete it in time. So I will stop running ap until a faster app becomes availeble and without knowledge how too do it I have too wait.

If you can't finish AP on time, BOINC will give it priority.

While AP is running "priority" Einstein will build debt.

Because of the debt, BOINC won't download another AP unit for a while, and when the debt is paid back, you'll be back at your 50/50 resource share.

... and BOINC will handle all of that just fine.
ID: 800712 · Report as offensive
Profile Francois Piednoel
Avatar

Send message
Joined: 14 Jun 00
Posts: 898
Credit: 5,969,361
RAC: 0
United States
Message 800718 - Posted: 22 Aug 2008, 7:28:18 UTC - in response to Message 800431.  
Last modified: 22 Aug 2008, 7:29:44 UTC

Intel being faster than AMD in these kinds of scenarios are due to Intel's large L2, and that the L2 is shared amongst the cores. About 90% of the crunch time can be decided with the built-in benchmark of BOINC. Integer math doesn't do anything for crunching, it's the floating point benchmark that matters.

AMD CPUs are more efficient in their design, but Intel's large L2 and extremely high FSB are what make the difference here.

I posted about this back in...earlier this year, a pretty detailed breakdown of the difference between the two.

Found it.. http://setiathome.berkeley.edu/forum_thread.php?id=45616&nowrap=true#721263

Hope this answers your question.


This is so innacurate, i can't let it be.

Core 2 has much more resources to process float that K8 or K10, the pipeline is 4 large on Core 2 while it is only 3 large on K10 and previous.
Macrofusion allow Core 2 to decode up to 5 instructions, in the case of Compare and jump for example ...
This allow Core 2 to process more in parallel, and increase float Instruction per clock ratio.

Memory deambiguation in core 2 accelerate SETI quite a bit too, it allow the processor to keep the memory streams linear, while other processor need to use the CAS of their memory controler to seek several times.

The cache of Core 2 is only a good resources for the large 4 wide pipeline. If you take time to vTune seti, you ll see that the cache miss ratio is low, and the prefetch success is almost 90%.

Grrrrrrr......

who?
ID: 800718 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 800737 - Posted: 22 Aug 2008, 9:19:00 UTC

Have a QX9650 @ 3500MHz, FSB 1400, running UBUNTU 8.04 (x86), for a change.
An AP WU takes about 20-28 hours, to complete.
Does anyone know, why LINUX is showing the real amount off L2 cache ½ per 2 core's off total L2, while WINDOWS only shows 244KB instead off (¼ or ½ off total cache amount)?
I don't run an optimized app. yet, on this one.


ID: 800737 · Report as offensive
pete1939

Send message
Joined: 30 Aug 99
Posts: 4
Credit: 251,809
RAC: 0
United States
Message 800830 - Posted: 22 Aug 2008, 15:33:48 UTC

Hi, one of my systems has been crunching an AP WU for several weeks now.

It shows 69 1/2 hours of CPU time to this point, and 1007 hours to completion.

Its Status is "Running, high priority."

Its report deadline is 9/10/2008.

It'll never make it at this rate.

I'd hate to have all this CPU time wasted.

My system is a Intel Dual Core D805 running at 2.66GHz with 2Gig of RAM installed.

Normal Seti@Home WU's take about 6 hours to complete on these CPU's.

Any suggestions?

Thanks.
ID: 800830 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 800836 - Posted: 22 Aug 2008, 15:44:27 UTC - in response to Message 800830.  

Hi, one of my systems has been crunching an AP WU for several weeks now.

It shows 69 1/2 hours of CPU time to this point, and 1007 hours to completion.

Its Status is "Running, high priority."

Its report deadline is 9/10/2008.

It'll never make it at this rate.

I'd hate to have all this CPU time wasted.

My system is a Intel Dual Core D805 running at 2.66GHz with 2Gig of RAM installed.

Normal Seti@Home WU's take about 6 hours to complete on these CPU's.

Any suggestions?

Thanks.

The 1007 hours to completion isn't really an important figure - that's only an estimate, and all it tells us is that BOINC has got the basis for its estimation wrong - no AP unit should take anything like that long on a machine like yours.

The useful figure that could help us advise you is the "Progress" - the %age complete shown in BOINC Manager.

One thing that worries me is that you've only done 69.5 hours (under three days) of work in the 'several weeks' it's been on your machine. Either that means your computer isn't turned on for very many hours each day, or that the CPU spends a high proportion of its time on other BOINC tasks - a low SETI resource share. Neither is good news for completing AP tasks.

But let's wait for that %age figure before rushing to conclusions.
ID: 800836 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Astropulse units too big?


 
©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.