mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > NFS@Home

Reply
 
Thread Tools
Old 2019-02-05, 23:40   #1
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

117E16 Posts
Default Matrix data for GNFS 190+ on NFS@home

I took advantage of all the pastebin'ed logs to compile a list of results for all the 15e jobs that were GNFS 190 or bigger (and one 189 because I ran it and had the data handy)
First, the 32LP jobs:
Code:
Size	name		poly score	unique rels	matrix 	density
189	149_128	xyyx	2.47e-14	360M		22.3M	128
190	HP2_4496_296	2.02e-14	343M		34.1M	110
190	3270_698	1.70e-14	320M		23.9M	120
191	A3408_1668	1.73e-14	385M		29.0M	130
191	L5355A		1.41e-14	324M		31.4M	124
192	L1438		1.31e-14	425M		27.8M	146
							28.5M	134
192	L3655A		1.31e-14	382M		25.0M	150
							26.5M	134
193	2340_742	1.50e-14	320M		31.5M	120
193	L5895B		9.36e-15	346M		38.5M	126
I listed two matrices on some of fivemack's jobs because I don't typically try densities larger than 136, so I wanted comparisons with TD 134 even though he built and solved denser matrices. His logs include his filtering runs at TD 134, so I list that data.

33LP jobs from the 15e queue:
Code:
Size	name		poly score	unique rels	matrix 	density
193	143_93 xyyx	1.27e-14	680M		38.3M	128
194	A4788_5236	1.22e-14?*	659M		34.4M	116.7
195	148_83 xyyx	7.62e-15	707M		49.1M	136
195	A3366_2180	1.02e-14	695M		37.4M	134
196	135_124 xyyx	7.95e-15	687M		44.4M	151.6
							45.8M	134
197	A4788_12515	7.57e-15	615M		51.2M	118
198	Euclid-Mullin	7.11e-15	786M		38.4M	120
* : The log for this job did not include the skew; skew 1.00 made the poly score x.xxe-17. There is a best-poly-score for this size from that era, so I guessed it might be from this job and used that score.
Decimal target densities reflect a requested TD higher than what filtering produced, indicating the input was oversieved and that higher TDs would not produce different matrices (but limiting number of relations might!)

This data gives us an idea of how large a matrix might be expected, and can guide us in deciding whether to sieve more. With the benefit of hindsight and this comparative data, I would say the C190 HP2 should have been sieved a little more, likewise Aliquot 4788_12515. Also, the 148_83 xyyx input was a nasty, nasty matrix and I would have asked for more sieving even though 707M unique relations is more than nearly any other number here received.

Later I'll do the same for frmky's 16e logs.

Last fiddled with by VBCurtis on 2019-02-06 at 00:02
VBCurtis is offline   Reply With Quote
Old 2019-02-05, 23:56   #2
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2×2,239 Posts
Default

Some conclusions:
15e tasks below GNFS-190 can be solved on 16GB machines, as the matrices should be 26M or smaller. Memory footprint is around 10% larger than the reported size of the matrix in msieve, and 25-26M matrices are generally reported to be 13GB or smaller by msieve.

GNFS-190 to 193 should use 32LP to produce a matrix smaller than 33M. If it doesn't, request some more sieving.

GNFS-194 to 197 should produce matrices 35M to 45M. Greg's run with Euclid-Mullin indicates that more relations than we've otherwise used may be helpful in keeping matrices under 45M in size.

GNFS 193 and 194 should consider using 32LP, as the 33LP job using a 25% better-scoring poly produced the same size matrix as a 32LP job, while the 32LP job needed just over half the unique relations of the 33LP. If yield is good, go 32LP here.
More data comparing the two would help, and NFS@home is just the place to create that data! Perhaps a 32/33 hybrid for poly scores just below 1e-14?
VBCurtis is offline   Reply With Quote
Old 2019-02-06, 00:42   #3
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

2·2,239 Posts
Default

Data from the 16e queue, before 1/1/2017:
Code:
Size	name		poly score	unique rels	matrix 	density
190	7,401-		2.22e-14	835M		18.0M	130
191	3,790+		2.03e-14	755M		20.5M	130
192	5,485+		1.70e-14	948M		18.7M	130
194	2.2186M		1.06e-14	737M		29.1M	130
196	5,1085L		8.16e-15	824M		32.8M	130
196	2.2530M		skew1		866M		26.2M	130
197	10,325+		7.24e-15	830M		28.0M	130
197	147_136 xyyx	7.22e-15	833M		27.6M	130
202	6,460+		3.44e-15	749M		41.9M	130
202	11,671M		3.16e-15	851M		35.9M	130
204	7,373-		2.47e-15	709M		46.0M	130
208	3,703+		1.21e-15	765M		53.0M	130		
208	10,359-		1.12e-15	631M		72.5M	120
212	6,490+		6.43e-16(deg5)	661M		79.1M	120
218	2.1285-		2.64e-16(deg5)	674M		108.5M	120
Seems frmky has a different idea of properly sieved than we do! Looks like 800M unique relations keeps matrices below 35M even up to GNFS-198. Even GNFS-202 is manageable at 36M matrix, provided we can find 850M unique relations!

I'll post 2017's data later.
VBCurtis is offline   Reply With Quote
Old 2019-02-07, 00:19   #4
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

105768 Posts
Default

Data from 16e queue after 1/1/17:
Code:
Size	name		poly score	unique rels	matrix 	density
193	E162		1.27e-14	839M		24.8M	130
198	B228		6.18e-15	844M		28.0M	130
200	E192		3.99e-15	771M		41.4M	130
200	Repunit507	4.53e-15	735M		40.1M	130	KurtB
202	E148		2.88e-15	804M		44.8M	130
206	XYYX139_123	2.11e-15	792M		43.6M	134
207	HP2_4496_310	1.63e-15	730M		51.6M	130
208	B248		1.15e-15	723M		58.1M	130
If a mod wishes to merge this into the previous post, be my guest.

Last fiddled with by VBCurtis on 2019-04-28 at 12:35
VBCurtis is offline   Reply With Quote
Old 2019-04-27, 23:47   #5
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

117E16 Posts
Default

Added: The C206 from XYYX project. I've just finished filtering; 1165M raw 33LP relations yielded 792M unique. TD=134 produced a matrix just under 44M in size.
VBCurtis is offline   Reply With Quote
Old 2019-04-28, 05:59   #6
kurtb
 
"Beschorner Kurt"
Jul 2016
Germany

238 Posts
Default

for comparison/as an addition:

our team is currently working on the repunit-number R507,GNFS, c200 with lpb.= 33,

combined = 4.526e-15, TD=130, 735.4M uniques, 625.7 unique ideals andmatrix-size = 40.094968.

closer: www.kurtbeschorner.de
Kurt

VBCurtis: Added to chart. Thanks!

Last fiddled with by VBCurtis on 2019-04-28 at 12:36
kurtb is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Connectivity Matrix Xyzzy Lounge 13 2017-02-21 18:29
GNFS matrix stage on GPU fivemack GPU Computing 0 2016-07-14 15:44
12+256 matrix job fivemack Factoring 11 2009-08-18 17:39
GF(2) Matrix request oslik Factoring 22 2008-11-02 12:53
[Need help] about Matrix Polynomial buan Homework Help 3 2007-07-17 15:07

All times are UTC. The time now is 05:28.

Fri Nov 27 05:28:47 UTC 2020 up 78 days, 2:39, 4 users, load averages: 1.06, 1.26, 1.38

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.