mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > Operation Kibibit

Reply
 
Thread Tools
Old 2011-04-23, 14:24   #23
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

102410 Posts
Default

Quote:
Originally Posted by bdodson View Post
... So the GPU's are running 1.48, while the CPUs are running 1.49.
Here's how the current outfiles look
Code:
  1050 Apr 23 09:26 msieve.dat5b.m
  3220 Apr 23 09:26 msieve.dat0b.m
  3409 Apr 23 09:26 msieve.dat4b.m
  3767 Apr 23 09:26 msieve.dat1b.m
 19162 Apr 23 09:26 msieve.dat.m
  4830 Apr 23 09:26 msieve.dat3b.m
  5619 Apr 23 09:26 msieve.dat2b.m
with both cards writing to msieve.dat.m and
Code:
wc -l msieve.dat.m

272 msieve.dat.m
from 12 hours, and ... -Bruce
Wow, 1.49 on the CPUs is really picking up hits; perhaps I ought
to recompile the GPU binary as well.
Code:
16310 Apr 23 10:20 msieve.dat0b.m
20645 Apr 23 10:20 msieve.dat.m
12950 Apr 23 10:20 msieve.dat5b.m
26312 Apr 23 10:20 msieve.dat4b.m
30940 Apr 23 10:20 msieve.dat3b.m
35924 Apr 23 10:20 msieve.dat2b.m
23396 Apr 23 10:20 msieve.dat1b.m
with
Code:
wc -l msieve.dat*b.m
   235 msieve.dat0b.m
   332 msieve.dat1b.m
   509 msieve.dat2b.m
   447 msieve.dat3b.m
   373 msieve.dat4b.m
   188 msieve.dat5b.m
  2084 total
from just a bit over 1 hour. -BD
bdodson is offline   Reply With Quote
Old 2011-04-24, 14:28   #24
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

20008 Posts
Default

Quote:
Originally Posted by bdodson View Post
Code:
wc -l msieve.dat*b.m
   235 msieve.dat0b.m
   332 msieve.dat1b.m
   509 msieve.dat2b.m
   447 msieve.dat3b.m
   373 msieve.dat4b.m
   188 msieve.dat5b.m
  2084 total
from just a bit over 1 hour. -BD
Looks like the cards will stay with 1.48. One is at/past 2M+27K, the
other 10M+56K, and reports
Code:
p = 72.79 bits, sieve = 104.53 bits
coeff 10056432-10074900 specialq 16428161 - 16533911
coeff 10056432-10074900 specialq 16428161 - 16533911 p1 18407303 - 20248034 p2 20248035 - 22272839
crap
The output from the CPUs with 1.49 is even more interesting, with
juicy info on the search ranges and random choices. Very nice!
Code:
randomizing rational coefficient: using piece #139 of 307
p = 72.40 bits, sieve = 103.17 bits
coeff 2005692 specialq 340381603 - 342170728 other 3483506 - 5225260
aprogs: 40757 entries, 139404 roots
hashtable: 22491 entries,  2.69 MB
randomizing rational coefficient: using piece #122 of 307
p = 72.40 bits, sieve = 103.17 bits
coeff 2005944 specialq 311374924 - 313011583 other 3642188 - 5463283
aprogs: 42170 entries, 142060 roots
hashtable: 66472 entries,  2.69 MB
randomizing rational coefficient: using piece #137 of 307
p = 72.40 bits, sieve = 103.17 bits
coeff 2005968 specialq 336850680 - 338621246 other 3501756 - 5252635
aprogs: 41201 entries, 139960 roots
---

randomizing rational coefficient: using piece #16 of 360
p = 72.50 bits, sieve = 103.51 bits
coeff 3008280 specialq 209050246 - 209986931 other 4599526 - 6899290
aprogs: 52508 entries, 175374 roots
hashtable: 32975 entries,  2.94 MB
randomizing rational coefficient: using piece #76 of 360
p = 72.50 bits, sieve = 103.51 bits
coeff 3008460 specialq 273373561 - 274598458 other 4022188 - 6033282
aprogs: 47565 entries, 168910 roots
---

randomizing rational coefficient: using piece #271 of 404
p = 72.57 bits, sieve = 103.75 bits
coeff 4007784 specialq 645908807 - 648487083 other 2680654 - 4020981
aprogs: 32793 entries, 114558 roots
hashtable: 26688 entries,  1.47 MB
randomizing rational coefficient: using piece #58 of 404
p = 72.57 bits, sieve = 103.75 bits
coeff 4007988 specialq 276480211 - 277583838 other 4097285 - 6145928
aprogs: 47812 entries, 164798 roots
----

and

randomizing rational coefficient: using piece #160 of 580
p = 72.79 bits, sieve = 104.53 bits
coeff 10011564 specialq 501553948 - 502947639 other 3285210 - 4927815
aprogs: 38739 entries, 135384 roots
hashtable: 37284 entries,  2.69 MB
randomizing rational coefficient: using piece #59 of 580
p = 72.79 bits, sieve = 104.53 bits
coeff 10011708 specialq 378968459 - 380021516 other 3779384 - 5669076
aprogs: 43783 entries, 150930 roots
---

randomizing rational coefficient: using piece #436 of 603
p = 72.81 bits, sieve = 104.61 bits
coeff 11011440 specialq 1071917762 - 1074782587 other 2265214 - 3397822
aprogs: 28304 entries, 100502 roots
hashtable: 30066 entries,  1.47 MB
randomizing rational coefficient: using piece #294 of 603
p = 72.81 bits, sieve = 104.61 bits
coeff 11011572 specialq 733776392 - 735737495 other 2737844 - 4106766
aprogs: 33223 entries, 115534 roots
---

randomizing rational coefficient: using piece #220 of 624
p = 72.83 bits, sieve = 104.68 bits
coeff 12014640 specialq 612376804 - 613958302 other 3018949 - 4528424
aprogs: 36878 entries, 134740 roots
hashtable: 83968 entries,  2.69 MB
randomizing rational coefficient: using piece #127 of 624
p = 72.83 bits, sieve = 104.68 bits
coeff 12015600 specialq 481792212 - 483036468 other 3403598 - 5105397
aprogs: 40127 entries, 142482 roots
The cards/1.48 have
Code:
    567 10M

    298 2M
while the CPUs/1.49 have
Code:
 wc -l msieve.dat*b.m

   6169 msieve.dat0b.m   2M
   7190 msieve.dat1b.m  10M
   9007 msieve.dat2b.m  11M
   6750 msieve.dat3b.m   3M
   8401 msieve.dat4b.m  12M
   6220 msieve.dat5b.m   4M
     ----
  43737 total
If what matters most is a productive flare, and the search parameters
are apples-to-oranges between 1.48 and 1.49, hard to draw any
conclusion as to relative GPU/CPU performance. A few more days
of this? -Bruce
bdodson is offline   Reply With Quote
Old 2011-04-24, 20:18   #25
jrk
 
jrk's Avatar
 
May 2008

44516 Posts
Default

Quote:
Originally Posted by bdodson View Post
Code:
p = 72.79 bits, sieve = 104.53 bits
coeff 10056432-10074900 specialq 16428161 - 16533911
coeff 10056432-10074900 specialq 16428161 - 16533911 p1 18407303 - 20248034 p2 20248035 - 22272839
crap
There should have been a "poly..." line printed right before that "crap". Can you find it and print it here? Thanks.
jrk is offline   Reply With Quote
Old 2011-04-25, 00:50   #26
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

102410 Posts
Default

Quote:
Originally Posted by jrk View Post
There should have been a "poly..." line printed right before that "crap". Can you find it and print it here? Thanks.
That's
Code:
...skipping
poly 39 16493909 18497239 20289103 22273495589109453572813443470021100589

poly 1 16532809 18527963 20755981 1344409181677691008652746587903351020900243880

crap

poly  2 16527841 18530881 20419897 22280024062844895271365237084380991633
from the C2050 tesla card, searching in 10M with 1.48. -Bruce

Ooops, here's another, loc. cit.
Code:
...skipping
poly 17 511855391 3319039 3707107 22269606992654849528684178539454344108
poly 36 510302171 3337199 3673937 1348682944076260343695021723909596962402744069
crap
poly  1 511616627 3314963 3974611 22273270022396631287163377150661153420

Last fiddled with by bdodson on 2011-04-25 at 00:53 Reason: 2nd one
bdodson is offline   Reply With Quote
Old 2011-04-25, 01:14   #27
jrk
 
jrk's Avatar
 
May 2008

21058 Posts
Default

From that output it looks like you're using an old SVN. Please use 1.48 or the latest SVN revision and see if you still get "crap" and post here if you do. Thanks.
jrk is offline   Reply With Quote
Old 2011-04-25, 01:54   #28
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

210 Posts
Default

Quote:
Originally Posted by jrk View Post
From that output it looks like you're using an old SVN. Please use 1.48 or the latest SVN revision and see if you still get "crap" and post here if you do. Thanks.
Try again, friend. As I've mentioned 4-5 times, this is 1.48:
Code:
Msieve v. 1.48
Fri Apr 22 21:37:10 2011
random seeds: 57802ace 45d126b2
factoring 123018668453011775513049495838496272077285356959533479219732245215172640050726365751874520219978
6469389956474942774063845925192557326303453731548268507917026122142913461670429214311602221240479274737794
080665351419597459856902143413 (232 digits)
...
Early 1.48, here's the date
Code:
2063872 Jan  5 22:27 jan2011/msieve-trunk.tar
Any chance you would know why the CUDA=1 compile that worked
for several earlier msieve versions no longer works for 1.49? Greg
tried a time-or-two, and the new version doesn't locate files that the
previous ones did; but even after sorting through cuda.h, lib64 and
the nvcc locations, 1.49 gives
Code:
...
ge1_sieve_gpu_sq.no
ar: creating libmsieve.a
ranlib libmsieve.a
gcc -D_FILE_OFFSET_BITS=64 -O3 -fomit-frame-pointer -march=core2 -DNDEBUG -D_LARGEFILE64_SOURCE  -Wall -W -I. -Iinclude -Ignfs -Ignfs/poly -Ignfs/poly/stage1 -I"/usr/include/cuda" -DHAVE_CUDA demo.c -o msieve  \
                        libmsieve.a -lcuda -lz -lgmp -lm -lpthread
libmsieve.a(stage1_sieve.no): In function `sieve_lattice':
stage1_sieve.c:(.text+0x127): undefined reference to `sieve_lattice_cpu'
stage1_sieve.c:(.text+0x1be): undefined reference to `sieve_lattice_cpu'
collect2: ld returned 1 exit status
make: *** [x86_64] Error 1
or I would otherwise have switched to 1.49. -Bruce

PS --- Are you saying that you don't want data from this Jan-2011 version
of 1.48? I'll retire the cards until there's a version of 1.49 that runs, if you
prefer?

Last fiddled with by bdodson on 2011-04-25 at 01:58 Reason: data wanted from old 1.48?
bdodson is offline   Reply With Quote
Old 2011-04-25, 02:47   #29
jrk
 
jrk's Avatar
 
May 2008

1,093 Posts
Post

Quote:
Originally Posted by bdodson View Post
Try again, friend. As I've mentioned 4-5 times, this is 1.48:
Quote:
Originally Posted by bdodson View Post
Early 1.48, here's the date
Code:
2063872 Jan  5 22:27 jan2011/msieve-trunk.tar
SVN revisions between 1.47 and 1.48 would have all said v1.48 in the logs, even though they weren't final. The final 1.48 was released on Jan 8th. Does that make sense? Sorry for the confusion.

(I knew you weren't using the final 1.48 because the format of the "poly..." lines changed a little before 1.48 was released, and yours is the old format.)

Quote:
Originally Posted by bdodson View Post
Any chance you would know why the CUDA=1 compile that worked
for several earlier msieve versions no longer works for 1.49? Greg
tried a time-or-two, and the new version doesn't locate files that the
previous ones did; but even after sorting through cuda.h, lib64 and
the nvcc locations, 1.49 gives
Code:
...
ge1_sieve_gpu_sq.no
ar: creating libmsieve.a
ranlib libmsieve.a
gcc -D_FILE_OFFSET_BITS=64 -O3 -fomit-frame-pointer -march=core2 -DNDEBUG -D_LARGEFILE64_SOURCE  -Wall -W -I. -Iinclude -Ignfs -Ignfs/poly -Ignfs/poly/stage1 -I"/usr/include/cuda" -DHAVE_CUDA demo.c -o msieve  \
                        libmsieve.a -lcuda -lz -lgmp -lm -lpthread
libmsieve.a(stage1_sieve.no): In function `sieve_lattice':
stage1_sieve.c:(.text+0x127): undefined reference to `sieve_lattice_cpu'
stage1_sieve.c:(.text+0x1be): undefined reference to `sieve_lattice_cpu'
collect2: ld returned 1 exit status
make: *** [x86_64] Error 1
or I would otherwise have switched to 1.49. -Bruce
Do a "make clean" first before rebuilding the code with CUDA=1. Right there you have some files built without CUDA=1 and the linking fails.

Make sure you have the correct cuda library in the Makefile.

Code:
        # Also, the CUDA driver library has a different name in linux
        # LIBS += "$(CUDA_LIB_DIR)/cuda.lib"
        LIBS += -lcuda
fix as necessary for Linux, and you will also need to set CUDA_PATH to the location of the cuda installation. e.g.:

CUDA=1 CUDA_PATH=/usr/local/cuda/ make x86_64

or wherever your cuda is installed.

(trailing slash is necessary, don't omit it)


Quote:
Originally Posted by bdodson View Post
PS --- Are you saying that you don't want data from this Jan-2011 version
of 1.48? I'll retire the cards until there's a version of 1.49 that runs, if you
prefer?
Who said that? The data (e.g. msieve.dat.m files) is fine. I only wanted you to try upgrading to see if the occasional "crap" was fixed in 1.48 and later, as it should be. That's all.

But its up to you.

Thanks.

Last fiddled with by jrk on 2011-04-25 at 03:10
jrk is offline   Reply With Quote
Old 2011-04-25, 13:29   #30
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

210 Posts
Default

Quote:
Originally Posted by jrk View Post
SVN revisions between 1.47 and 1.48 would have all said v1.48 in the logs, even though they weren't final. The final 1.48 was released on Jan 8th. Does that make sense? Sorry for the confusion.
...
Who said that? The data (e.g. msieve.dat.m files) is fine. I only wanted you to try upgrading to see if the occasional "crap" was fixed in 1.48 and later, as it should be. That's all.

But its up to you.

Thanks.
Good points, all, I'll have another look --- definitely missed that make-clean
between the compile without CUDA=1, and then the one with. -Bruce
bdodson is offline   Reply With Quote
Old 2011-04-25, 18:29   #31
bdodson
 
bdodson's Avatar
 
Jun 2005
lehigh.edu

102410 Posts
Default

Quote:
Originally Posted by jrk View Post
... I only wanted you to try upgrading to see if the occasional "crap" was fixed in 1.48 and later, as it should be. ...
Actually, I think that's one of the reasons I was mentioning it; I believe
I do recall reading that it was fixed.

Anyway, 1.49_gpu is running now; I shipped of the last of the old
1.47 data to Jason. -Bruce
bdodson is offline   Reply With Quote
Old 2011-04-25, 20:39   #32
jasonp
Tribal Bullet
 
jasonp's Avatar
 
Oct 2004

3·1,163 Posts
Default

Got it, thanks.

Latter-day v1.48 really has a lot of improvements in this neighborhood, hopefully your throughput will go up now. I think I really do need to print out the SVN version, keeping the same version number just confuses too many people.
jasonp is offline   Reply With Quote
Old 2011-04-26, 12:08   #33
jasonp
Tribal Bullet
 
jasonp's Avatar
 
Oct 2004

3×1,163 Posts
Default

I'm boiling down the first batch of 33000 stage 1 hits now, and it looks like the first part of stage 2 is going to need a *lot* of improving before it can deal with polynomials that are so highly skewed.

polynomial deleted...one machine here was doing a poly search for RSA232, not RSA768

Last fiddled with by jasonp on 2011-04-27 at 11:38
jasonp is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Polynomial selection Max0526 NFS@Home 9 2017-05-20 08:57
Improved NFS polynomial selection jasonp Operation Kibibit 5 2014-09-07 11:02
Call for volunteers: RSA896 jasonp Operation Kibibit 134 2013-09-03 22:08
2^877-1 polynomial selection fivemack Factoring 47 2009-06-16 00:24
Polynomial selection CRGreathouse Factoring 2 2009-05-25 07:55

All times are UTC. The time now is 03:02.

Tue Oct 27 03:02:59 UTC 2020 up 47 days, 13 mins, 1 user, load averages: 1.51, 1.54, 1.56

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.