Msieve 1.52 feedback

chris2be8 · 2014-01-29, 17:23

Perhaps we should have a msieve 1.52 feedback thread, but I'll post this here.

I've tested the improved LA in msieve 1.52 SVN 956. It took 11:01:48 to solve the 5174395 x 5174622 matrix for 15^208+1 on my Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz stepping 09.

By contrast msieve 1.50 took 19:07:27 to solve the slightly smaller 5115047 x 5115204 matrix from 47^131+1 on the same system. So it's a nice speedup.

Will it speed up LA if I'm only using 1 core for LA (they will be small jobs so I get best throughput with 1 core per number)?

Chris

chris2be8 · 2014-02-01, 17:24

Hello,

There was a post a year or two ago about how to stop msieve's polynomial search producing so much screen output by commenting out some printfs. I can remember it but I can't find it.Can anyone point me to it.

I would like to cut it down to just saying which highest coefficient it's processing, without listing all the stage 1 hits. As if I was running:

Code:

 msieve -np1 -nps (other parameters) | grep '[a-zA-Z]'

To skip all the lines that are just numbers.

Chris

PS. Does the faster LA in 1.52 help if you are only using 1 thread?

EdH · 2014-02-02, 04:35

Quote:

Originally Posted by chris2be8

Hello,

There was a post a year or two ago about how to stop msieve's polynomial search producing so much screen output by commenting out some printfs. I can remember it but I can't find it.Can anyone point me to it.

I would like to cut it down to just saying which highest coefficient it's processing, without listing all the stage 1 hits. As if I was running:

Code:

 msieve -np1 -nps (other parameters) | grep '[a-zA-Z]'

To skip all the lines that are just numbers.

Chris

PS. Does the faster LA in 1.52 help if you are only using 1 thread?

I think this was discussed in the YAFU Feature Request thread a couple times, but I don't recall a resolution there. Sorry I'm not more help...

jasonp · 2014-02-02, 15:25

Quote:

Originally Posted by chris2be8

PS. Does the faster LA in 1.52 help if you are only using 1 thread?

Yes, the new code makes much better use of large caches. On an Ivy Bridge CPU the new code is 30% faster than the old with 1 thread (50% faster with 4 threads).

chris2be8 · 2014-02-08, 16:59

Hello,

How does the poly_deadline parameter work. I've written a script to run -np1 -nps, sort the hits in msieve.dat.ms, then run -npr with the deadline set to 10% of the time needed for the first stage. But msieve ignores the deadline.

I was testing it on rsa100 so the deadline was 30s. I called it with:
/home/chris/msieve-svn/trunk/msieve" -npr 'poly_deadline=30' -v -t 1 -l ggnfs.log -s rsa100.msieve.dat -nf rsa100.msieve.fb -i rsa100.ini >rsa100.msieve.npr.sysout

And the log from that run (until I hit control-C) is:

Code:

Sat Feb  8 16:00:00 2014
Sat Feb  8 16:00:00 2014  Msieve v. 1.52 (SVN 956)
Sat Feb  8 16:00:00 2014  random seeds: abf100f6 339d840b
Sat Feb  8 16:00:00 2014  factoring 1522605027922533360535618378132637429718068114961380688657908494580122963258952897654000350692006139 (100 digits)
Sat Feb  8 16:00:00 2014  searching for 15-digit factors
Sat Feb  8 16:00:00 2014  commencing number field sieve (100-digit input)
Sat Feb  8 16:00:00 2014  commencing number field sieve polynomial selection
Sat Feb  8 16:00:00 2014  polynomial degree: 4
Sat Feb  8 16:00:00 2014  max stage 1 norm: 1.36e+17
Sat Feb  8 16:00:00 2014  max stage 2 norm: 3.19e+15
Sat Feb  8 16:00:00 2014  min E-value: 9.14e-09
Sat Feb  8 16:00:00 2014  poly select deadline: 30
Sat Feb  8 16:01:51 2014  polynomial selection complete
Sat Feb  8 16:01:51 2014  R0: -1140493609401007544237814
Sat Feb  8 16:01:51 2014  R1: 7040576769931
Sat Feb  8 16:01:51 2014  A0: 1109762429222849242076992255
Sat Feb  8 16:01:51 2014  A1: 2580903068981216992378
Sat Feb  8 16:01:51 2014  A2: -2965715880152720
Sat Feb  8 16:01:51 2014  A3: -8954093337
Sat Feb  8 16:01:51 2014  A4: 900
Sat Feb  8 16:01:51 2014  skew 1264463.83, size 1.029e-13, alpha -4.838, combined = 1.222e-08 rroots = 2
Sat Feb  8 16:01:51 2014  elapsed time 00:01:51

Does it apply to running -npr only?

Chris

jasonp · 2014-02-08, 20:10

Yes, the time limit only applies to stage 1; stage 2 does not have limits.

chris2be8 · 2014-02-09, 17:05

Quote:

Originally Posted by jasonp

Yes, the time limit only applies to stage 1; stage 2 does not have limits.

Thanks, I can work round that now I know why it was ignored.

chris2be8 · 2014-02-09, 17:08

Quote:

Originally Posted by EdH

I think this was discussed in the YAFU Feature Request thread a couple times, but I don't recall a resolution there. Sorry I'm not more help...

I've found it in the Yafu on Linux thread. Posts 12 and 15.

Chris

EdH · 2014-02-09, 20:12

Quote:

Originally Posted by chris2be8

I've found it in the Yafu on Linux thread. Posts 12 and 15.

Chris

Ah, that looks familiar. Glad you found them...

wombatman · 2014-03-15, 22:03

Does anybody have a 64-bit CUDA enabled version of the latest (or close to it) version of Msieve for windows/MinGW-64? I know there's the SVN 942 version available, but it has issues with the matrix steps. I'd like to have everything in one exe if possible.

xilman · 2014-03-31, 15:52

Since upgrading the Linux driver to 334.21 just now, msieve polynomial searching has been printing vast amount of crap (literally) and failing with "error (line 734): unexpected error" after running for anywhere between a second and ten minutes. This on a Tesla C1060 (cc=1.3) which is running nice and cool.

According to the nvidia forum there are other CUDA problems with this driver.

Luckily a laptop with a GT240M has filled the breach.

Paul

2014-01-29, 17:23	#1
chris2be8 Sep 2009 3⁴×5² Posts	Msieve 1.52 feedback Perhaps we should have a msieve 1.52 feedback thread, but I'll post this here. I've tested the improved LA in msieve 1.52 SVN 956. It took 11:01:48 to solve the 5174395 x 5174622 matrix for 15^208+1 on my Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz stepping 09. By contrast msieve 1.50 took 19:07:27 to solve the slightly smaller 5115047 x 5115204 matrix from 47^131+1 on the same system. So it's a nice speedup. Will it speed up LA if I'm only using 1 core for LA (they will be small jobs so I get best throughput with 1 core per number)? Chris

2014-02-01, 17:24	#2
chris2be8 Sep 2009 3⁴×5² Posts	Hello, There was a post a year or two ago about how to stop msieve's polynomial search producing so much screen output by commenting out some printfs. I can remember it but I can't find it.Can anyone point me to it. I would like to cut it down to just saying which highest coefficient it's processing, without listing all the stage 1 hits. As if I was running: Code: msieve -np1 -nps (other parameters) \| grep '[a-zA-Z]' To skip all the lines that are just numbers. Chris PS. Does the faster LA in 1.52 help if you are only using 1 thread?

2014-03-31, 15:52	#11
xilman Bamboozled! "𒉺𒌌𒇷𒆷𒀭" May 2003 Down not across 29×367 Posts	It's hitting the fan again. Since upgrading the Linux driver to 334.21 just now, msieve polynomial searching has been printing vast amount of crap (literally) and failing with "error (line 734): unexpected error" after running for anywhere between a second and ten minutes. This on a Tesla C1060 (cc=1.3) which is running nice and cool. According to the nvidia forum there are other CUDA problems with this driver. Luckily a laptop with a GT240M has filled the breach. Paul

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Msieve 1.53 feedback	xilman	Msieve	149	2018-11-12 06:37
Msieve 1.50 feedback	firejuggler	Msieve	99	2013-02-17 11:53
Msieve 1.43 feedback	Jeff Gilchrist	Msieve	47	2009-11-24 15:53
Msieve 1.42 feedback	Andi47	Msieve	167	2009-10-18 19:37
Msieve 1.41 Feedback	Batalov	Msieve	130	2009-06-09 16:01

2014-02-08, 20:10	#6
jasonp Tribal Bullet Oct 2004 2⁴×13×17 Posts	Yes, the time limit only applies to stage 1; stage 2 does not have limits.

2014-03-15, 22:03	#10
wombatman I moo ablest echo power! May 2013 1,741 Posts	Does anybody have a 64-bit CUDA enabled version of the latest (or close to it) version of Msieve for windows/MinGW-64? I know there's the SVN 942 version available, but it has issues with the matrix steps. I'd like to have everything in one exe if possible.