BOINC NFS sieving - NFS@Home - Page 144

VBCurtis · 2014-09-17, 04:26

I leave HT on, but use 3 threads for LA on my i7 laptop. I find that less memory-intensive items can run on a couple threads without meaningfully lengthening the matrix solve time; so I run t45-level ECM on 2-4 threads during LA phases. 3 ECM threads adds something like 40-50% to the matrix solve time.

Even one thread of LLR adds more than 33% to the matrix solve time, a net loss of productivity- more evidence that LA is memory bound.

wombatman · 2014-09-17, 13:11

Good to know. For what it's worth, on my desktop (i7-4930K with DDR3-2133), running with 8 threads (out of 12 max) gave a shorter matrix time than running with 5.

fivemack · 2014-09-17, 15:38

Five seems a rather odd number to run with; are 8 threads faster than 6?

wombatman · 2014-09-17, 20:37

With 8 threads, at 93.4% left, estimated time was 6 hours, 6 minutes. After stopping and restarting with 6 threads, at 94.2% left, estimated time is 6 hours, 7 minutes. So not a huge difference at this point, but 8 threads does seem to be slightly faster.

pinhodecarlos · 2014-09-17, 21:19

Quote:

Originally Posted by wombatman

With 8 threads, at 93.4% left, estimated time was 6 hours, 6 minutes. After stopping and restarting with 6 threads, at 94.2% left, estimated time is 6 hours, 7 minutes. So not a huge difference at this point, but 8 threads does seem to be slightly faster.

Did you make the restart after a checkpoint write? If not your estimates are wrong....lol...Trust me, use only 4 threads on an i7.

wombatman · 2014-09-17, 22:00

Quote:

Originally Posted by pinhodecarlos

Did you make the restart after a checkpoint write? If not your estimates are wrong....lol...Trust me, use only 4 threads on an i7.

This was on the 4930K, which is a hexcore. So I assume you would suggest using 6 then? As for the checkpoint write, it restarted at whatever the most recent checkpoint was. I allowed it to run for a little while (2 hours maybe?) before I made the comparison.

wombatman · 2014-09-18, 07:40

GC_2_804 factors as:

Code:

prp110 factor: 33171467545913528394901826342849435175076320098356616463849806938179530846476177045145467298416719242338777543
prp128 factor: 66337207356412000743645957542928569346131357963000529884803640096419852671381210310677453787545453308135827524459731428312304629

Used 8 threads for the most part, but not sure on exact time due to stopping and restarting.

xilman · 2014-09-18, 10:20

Quote:

Originally Posted by wombatman

GC_2_804 factors as:

Code:

prp110 factor: 33171467545913528394901826342849435175076320098356616463849806938179530846476177045145467298416719242338777543
prp128 factor: 66337207356412000743645957542928569346131357963000529884803640096419852671381210310677453787545453308135827524459731428312304629

Used 8 threads for the most part, but not sure on exact time due to stopping and restarting.

Nice one. Most emphatically not an ECM miss.

fivemack · 2014-09-18, 11:07

Code:

Thu Sep 18 12:01:49 2014  prp55 factor: 5582432252872869236521104018238102090268889308241072791
Thu Sep 18 12:01:49 2014  prp57 factor: 787070341411058525069805258917368057681579098025714593919
Thu Sep 18 12:01:49 2014  prp113 factor: 13258758255385801515762418489024142874342924955295765687599155762757473553398787872568751402338595721038333998723

90 hours for 9.0M matrix on i7/2600 -t3

Log at http://pastebin.com/kXQS6ALs

I see we've now pretty much caught up with the sievers.

PS the P55 was found about two months ago by Rob Hooft

wombatman · 2014-09-22, 01:05

About 88 hours at 120 density for HomePrimes C169.

fivemack · 2014-09-22, 22:31

Code:

Mon Sep 22 12:56:33 2014  prp68 factor: 11796685657411858681673189671529474078015358355745635406322085184849
Mon Sep 22 12:56:33 2014  prp201 factor: 167275321270004823113094376167500206492953755551507388838876379729575927948920882695267133105652523997625793124565907114456691515023037317492383141127893428298359878471498659572083339084478279782404701

202 hours for 18.7M matrix on i7/4930K -t6

Log at http://pastebin.com/U7581X1v

2014-09-18, 11:07	#1582
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2³·11·73 Posts	C803 done Code: Thu Sep 18 12:01:49 2014 prp55 factor: 5582432252872869236521104018238102090268889308241072791 Thu Sep 18 12:01:49 2014 prp57 factor: 787070341411058525069805258917368057681579098025714593919 Thu Sep 18 12:01:49 2014 prp113 factor: 13258758255385801515762418489024142874342924955295765687599155762757473553398787872568751402338595721038333998723 90 hours for 9.0M matrix on i7/2600 -t3 Log at http://pastebin.com/kXQS6ALs I see we've now pretty much caught up with the sievers. PS the P55 was found about two months ago by Rob Hooft Last fiddled with by fivemack on 2014-09-18 at 14:02

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Boinc Statistics for NFS@Home borked ?	thomasn	NFS@Home	1	2013-10-02 15:31
BOINC NFS sieving - RSALS	debrouxl	NFS@Home	621	2012-12-14 23:44
BOINC?	masser	Sierpinski/Riesel Base 5	1	2009-02-09 01:10
BOINC?	KEP	Twin Prime Search	212	2007-04-25 10:29
BOINC	bebarce	Software	3	2005-12-15 18:35

2014-09-17, 04:26	#1574
VBCurtis "Curtis" Feb 2005 Riverside, CA 4873₁₀ Posts	I leave HT on, but use 3 threads for LA on my i7 laptop. I find that less memory-intensive items can run on a couple threads without meaningfully lengthening the matrix solve time; so I run t45-level ECM on 2-4 threads during LA phases. 3 ECM threads adds something like 40-50% to the matrix solve time. Even one thread of LLR adds more than 33% to the matrix solve time, a net loss of productivity- more evidence that LA is memory bound.

2014-09-17, 13:11	#1575
wombatman I moo ablest echo power! May 2013 13·137 Posts	Good to know. For what it's worth, on my desktop (i7-4930K with DDR3-2133), running with 8 threads (out of 12 max) gave a shorter matrix time than running with 5.

2014-09-17, 15:38	#1576
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2³·11·73 Posts	Five seems a rather odd number to run with; are 8 threads faster than 6?

2014-09-17, 20:37	#1577
wombatman I moo ablest echo power! May 2013 13·137 Posts	With 8 threads, at 93.4% left, estimated time was 6 hours, 6 minutes. After stopping and restarting with 6 threads, at 94.2% left, estimated time is 6 hours, 7 minutes. So not a huge difference at this point, but 8 threads does seem to be slightly faster.

2014-09-22, 01:05	#1583
wombatman I moo ablest echo power! May 2013 13·137 Posts	About 88 hours at 120 density for HomePrimes C169.