Let's do another nice big GNFS job!

fivemack · 2010-11-12, 22:23

Various people - bdodson, batalov, jrk - have done a GPU-assisted polynomial search for the 187-digit cofactor of 2^956+1, and we have a decent polynomial, which should produce a matrix no harder than SNFS matrices that I've already done, in no more than ten to fifteen CPU-years of sieving.

This is a 16e job (with three large primes), so you will want about a gigabyte per core/thread - 15e doesn't seem to give enough yield at the larger Q values. Start off with Q=20 to 160 million. Won't be done by Christmas, but we could probably get the relations in by the end of winter and factors by Easter.

Polynomial:

Code:

n: 1332203900608590809938480645779084229713406802212647383719972524986911760145731133592887032687895490700549453758482415452796437240091528083264268919532584204888670812466960924609380670193
type: gnfs
# norm 2.589534e-18 alpha -8.321781 e 2.991e-14 rroots 5
skew: 320811150.54
c0:  405063155184101117008321502924944026097244672704
c1:  3339583019941934259279259002396544515176
c2: -47189249406761789351128506210198
c3: -265660627688613695069141
c4:  586429385543294
c5:  182280
Y0: -1488559874631410107110585286834796715
Y1:  149978515624994424157
alim: 150000000
rlim: 150000000
lpba: 32
lpbr: 31
mfba: 94
mfbr: 62
alambda: 3.6
rlambda: 2.6

Command line:
gnfs-lasieve4I16e -v -M 1 -a t.poly -f <starting value> -c 10000

This is not a quick job, each relation takes about one CPU-second on a K8/2500 using the 64-bit assembly siever, and we need about four hundred million of them. The yield is about 3.5 million relations per million Q (less for smaller Q), so a range of one million will take about four to five weeks on a single core; take short ranges unless you have multi-core machines, and you might wish to consider other projects if your machine is not 64-bit.

Upload by ftp, host ssh.fivemack.org, directory 2+956, as .gz

Reservations:

gimarel 20M-21M done
ET_ 21M-22M done
gimarel 22M-25M done
fivemack 25M-26M done
bsquared 26M-40M done
fivemack 40M-43M done
andi47 43M-44M done
gimarel 44M-50M done
bsquared 50M-55M done
fivemack 55M-60M done
bdodson 60M-67M done
batalov 67M-68M done
fivemack 68M-70M done
bdodson 70M-130M done
bsquared 130M-134M done
batalov 134M-135M done
bdodson 135M-150M done
fivemack 150M-151M done
bdodson 151M-160M done
bdodson 160M-180M done
bdodson 180M-190M done

Results:

Code:

 20M- 21M 17/11  2250928 gimarel
 21M- 22M 15/12  2300458 ET_
 22M- 25M 30/11  7071161 gimarel
 25M- 26M 25/11  2478091 fivemack
 26M- 40M 06/12 37088160 bsquared
 40M- 43M 03/12  8924404 fivemack
 43M- 44M 19/12  3014307 andi47
 44M- 50M 22/12 17975701 gimarel
 50M- 55M 14/12 15809882 bsquared
 55M- 60M 23/12 16096105 fivemack
 60M- 67M 06/01 22967376 bdodson
 67M- 68M 31/12  3283937 batalov
 68M- 70M 08/01  6600716 fivemack
 70M- 75M 25/12 16624144 bdodson
 75M- 80M 25/12 16661056 bdodson
 80M- 85M 22/12 16757155 bdodson
 85M- 90M 22/12 16789593 bdodson
 90M- 95M 24/12 16839477 bdodson
 95M-100M 24/12 16746582 bdodson
100M-105M 17/12 16849538 bdodson
105M-110M 17/12 16858220 bdodson
110M-115M 19/12 16682276 bdodson
115M-120M 19/12 16718242 bdodson
120M-125M 20/12 16678274 bdodson
125M-130M 20/12 16619360 bdodson
130M-134M 27/12 13226962 bsquared
134M-135M 27/11  3315231 batalov
135M-150M 01/12 49267416 bdodson
150M-151M 20/11  3263769 fivemack
151M-160M 03/12 29117682 bdodson
160M-165M 05/01 16012568 bdodson
165M-170M 05/01 15878307 bdodson
170M-175M 08/01 15727747 bdodson
175M-180M 08/01 15569140 bdodson
180M-185M 10/01 15507504 bdodson
185M-190M 10/01 15374119 bdodson

Andi47 · 2010-11-13, 08:05

What exactly does the "-M 1" in the command line do?
(I guess, "-v" stands for "verbouse"?)

Batalov · 2010-11-13, 08:17

Here. It is the side which is going to MPQS'd first.

ET_ · 2010-11-13, 14:46

I'd like to offer some help. I have an i5-750 @ 2.66 GHz and 8 GB of memory on Ubuntu 64-bit, but I'm unsure if I got the correct executable (1055887 bytes).

1 - Is it the right one or is there a link where I can get it?
2 - the t.poly is the polynomial presented in the first post, correct?
3 - What does the -c switch mean?

I'll try first with 1 core, then (when other jobs will finish) I'll add some more power.

4 - Should I insert 21000000 on the <starting value> position?
5 - Should I restart the computer, what is the correct procedure to maintain the work done?

Thank you!

Luigi

Andi47 · 2010-11-13, 16:58

1. - have not checked if it's the right one....
2. - yes
3. - this is the switch to adjust the size of the range. When you start at 21M and type "-c 10000", then it runs up to 21010000
4. - yes (if you want to start at 21M)
5. - just add "-R" to the command line you have typed the first time. In this case GGNFS reads the output file, restarts the job where it has been interrupted and runs up to your desired end of the range.

ET_ · 2010-11-13, 17:28

Quote:

Originally Posted by Andi47

1. - have not checked if it's the right one....
2. - yes
3. - this is the switch to adjust the size of the range. When you start at 21M and type "-c 10000", then it runs up to 21010000
4. - yes (if you want to start at 21M)
5. - just add "-R" to the command line you have typed the first time. In this case GGNFS reads the output file, restarts the job where it has been interrupted and runs up to your desired end of the range.

Thank you Andy!

I will start 21M-22M as soon as question 1 will get an answer.

Luigi

fivemack · 2010-11-13, 18:35

Quote:

Originally Posted by ET_

I'd like to offer some help. I have an i5-750 @ 2.66 GHz and 8 GB of memory on Ubuntu 64-bit, but I'm unsure if I got the correct executable (1055887 bytes).

There are a fair number of versions of gnfs-lasieve4I16e floating about; the one I'm using is 958368 bytes long with a date 2009-04-15 but I'm sure there are newer compiles. Where did you get yours?

I would just run yours on a range of 1000 or so; if you get about 3000 relations and a time-per-relation of about 0.7 seconds it's likely to be right.

Tom

ET_ · 2010-11-13, 19:36

Quote:

Originally Posted by fivemack

There are a fair number of versions of gnfs-lasieve4I16e floating about; the one I'm using is 958368 bytes long with a date 2009-04-15 but I'm sure there are newer compiles. Where did you get yours?

I would just run yours on a range of 1000 or so; if you get about 3000 relations and a time-per-relation of about 0.7 seconds it's likely to be right.

Tom

Hm... I just reached 0.85 sec/rel. I'm afraid I have the wrong executable.

Hints on where to get the good one?

Luigi

Batalov · 2010-11-14, 00:05

I get 0.55s/rel - but that's a K10 at 3640MHz (and using L1_BITS=16).
Use a L1_BITS=15 binary for Intel CPUs (they all have only 32Kb cache).

It also depends on the range. Tom probably has the data from sims - some graphs (just like those in earlier projects) could be helpful for participants to adjust their expectations.

fivemack · 2010-11-14, 11:42

ET_: 0.85s/rel probably isn't too bad, but you might want to try

ftp://ssh.fivemack.org/gnfs-lasieve4I16e

and see if it's any faster for you.

ET_ · 2010-11-14, 11:56

Quote:

Originally Posted by fivemack

ET_: 0.85s/rel probably isn't too bad, but you might want to try

ftp://ssh.fivemack.org/gnfs-lasieve4I16e

and see if it's any faster for you.

Response: 550 Failed to open file. :-(

May I start working on with my executable and eventually exchange it with a newer version during the run?

Luigi

2010-11-12, 22:23	#1
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1100100110110₂ Posts	Let's do another nice big GNFS job! Various people - bdodson, batalov, jrk - have done a GPU-assisted polynomial search for the 187-digit cofactor of 2^956+1, and we have a decent polynomial, which should produce a matrix no harder than SNFS matrices that I've already done, in no more than ten to fifteen CPU-years of sieving. This is a 16e job (with three large primes), so you will want about a gigabyte per core/thread - 15e doesn't seem to give enough yield at the larger Q values. Start off with Q=20 to 160 million. Won't be done by Christmas, but we could probably get the relations in by the end of winter and factors by Easter. Polynomial: Code: n: 1332203900608590809938480645779084229713406802212647383719972524986911760145731133592887032687895490700549453758482415452796437240091528083264268919532584204888670812466960924609380670193 type: gnfs # norm 2.589534e-18 alpha -8.321781 e 2.991e-14 rroots 5 skew: 320811150.54 c0: 405063155184101117008321502924944026097244672704 c1: 3339583019941934259279259002396544515176 c2: -47189249406761789351128506210198 c3: -265660627688613695069141 c4: 586429385543294 c5: 182280 Y0: -1488559874631410107110585286834796715 Y1: 149978515624994424157 alim: 150000000 rlim: 150000000 lpba: 32 lpbr: 31 mfba: 94 mfbr: 62 alambda: 3.6 rlambda: 2.6 Command line: gnfs-lasieve4I16e -v -M 1 -a t.poly -f <starting value> -c 10000 This is not a quick job, each relation takes about one CPU-second on a K8/2500 using the 64-bit assembly siever, and we need about four hundred million of them. The yield is about 3.5 million relations per million Q (less for smaller Q), so a range of one million will take about four to five weeks on a single core; take short ranges unless you have multi-core machines, and you might wish to consider other projects if your machine is not 64-bit. Upload by ftp, host ssh.fivemack.org, directory 2+956, as .gz Reservations: gimarel 20M-21M done ET_ 21M-22M done gimarel 22M-25M done fivemack 25M-26M done bsquared 26M-40M done fivemack 40M-43M done andi47 43M-44M done gimarel 44M-50M done bsquared 50M-55M done fivemack 55M-60M done bdodson 60M-67M done batalov 67M-68M done fivemack 68M-70M done bdodson 70M-130M done bsquared 130M-134M done batalov 134M-135M done bdodson 135M-150M done fivemack 150M-151M done bdodson 151M-160M done bdodson 160M-180M done bdodson 180M-190M done Results: Code: 20M- 21M 17/11 2250928 gimarel 21M- 22M 15/12 2300458 ET_ 22M- 25M 30/11 7071161 gimarel 25M- 26M 25/11 2478091 fivemack 26M- 40M 06/12 37088160 bsquared 40M- 43M 03/12 8924404 fivemack 43M- 44M 19/12 3014307 andi47 44M- 50M 22/12 17975701 gimarel 50M- 55M 14/12 15809882 bsquared 55M- 60M 23/12 16096105 fivemack 60M- 67M 06/01 22967376 bdodson 67M- 68M 31/12 3283937 batalov 68M- 70M 08/01 6600716 fivemack 70M- 75M 25/12 16624144 bdodson 75M- 80M 25/12 16661056 bdodson 80M- 85M 22/12 16757155 bdodson 85M- 90M 22/12 16789593 bdodson 90M- 95M 24/12 16839477 bdodson 95M-100M 24/12 16746582 bdodson 100M-105M 17/12 16849538 bdodson 105M-110M 17/12 16858220 bdodson 110M-115M 19/12 16682276 bdodson 115M-120M 19/12 16718242 bdodson 120M-125M 20/12 16678274 bdodson 125M-130M 20/12 16619360 bdodson 130M-134M 27/12 13226962 bsquared 134M-135M 27/11 3315231 batalov 135M-150M 01/12 49267416 bdodson 150M-151M 20/11 3263769 fivemack 151M-160M 03/12 29117682 bdodson 160M-165M 05/01 16012568 bdodson 165M-170M 05/01 15878307 bdodson 170M-175M 08/01 15727747 bdodson 175M-180M 08/01 15569140 bdodson 180M-185M 10/01 15507504 bdodson 185M-190M 10/01 15374119 bdodson Last fiddled with by fivemack on 2011-01-10 at 09:24 Reason: and it's done

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
The "Hello - I am... - Nice to meet you" thread....	Prime Monster	Lounge	29	2023-01-02 21:46
Nice progress!	schickel	FactorDB	29	2012-07-18 17:03
Nice pic	Dubslow	Forum Feedback	0	2012-05-02 02:13
Very nice strategical puzzle	Raman	Puzzles	2	2009-11-01 10:23
Nice link...	Xyzzy	Lounge	4	2003-06-28 13:37

2010-11-13, 08:05	#2
Andi47 Oct 2004 Austria 2·17·73 Posts	What exactly does the "-M 1" in the command line do? (I guess, "-v" stands for "verbouse"?)

2010-11-13, 08:17	#3
Batalov "Serge" Mar 2008 San Diego, Calif. 2×3×1,733 Posts	Here. It is the side which is going to MPQS'd first.

2010-11-13, 14:46	#4
ET_ Banned "Luigi" Aug 2002 Team Italia 4,871 Posts	I'd like to offer some help. I have an i5-750 @ 2.66 GHz and 8 GB of memory on Ubuntu 64-bit, but I'm unsure if I got the correct executable (1055887 bytes). 1 - Is it the right one or is there a link where I can get it? 2 - the t.poly is the polynomial presented in the first post, correct? 3 - What does the -c switch mean? I'll try first with 1 core, then (when other jobs will finish) I'll add some more power. 4 - Should I insert 21000000 on the <starting value> position? 5 - Should I restart the computer, what is the correct procedure to maintain the work done? Thank you! Luigi

2010-11-13, 16:58	#5
Andi47 Oct 2004 Austria 9B2₁₆ Posts	1. - have not checked if it's the right one.... 2. - yes 3. - this is the switch to adjust the size of the range. When you start at 21M and type "-c 10000", then it runs up to 21010000 4. - yes (if you want to start at 21M) 5. - just add "-R" to the command line you have typed the first time. In this case GGNFS reads the output file, restarts the job where it has been interrupted and runs up to your desired end of the range.

2010-11-14, 00:05	#9
Batalov "Serge" Mar 2008 San Diego, Calif. 2×3×1,733 Posts	I get 0.55s/rel - but that's a K10 at 3640MHz (and using L1_BITS=16). Use a L1_BITS=15 binary for Intel CPUs (they all have only 32Kb cache). It also depends on the range. Tom probably has the data from sims - some graphs (just like those in earlier projects) could be helpful for participants to adjust their expectations.

2010-11-14, 11:42	#10
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2·7·461 Posts	ET_: 0.85s/rel probably isn't too bad, but you might want to try ftp://ssh.fivemack.org/gnfs-lasieve4I16e and see if it's any faster for you.