mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Msieve (https://www.mersenneforum.org/forumdisplay.php?f=83)
-   -   Polynomial Request Thread (https://www.mersenneforum.org/showthread.php?t=18368)

firejuggler 2013-08-17 09:07

I will start with a leading coef of 12M.....

swellman 2013-08-17 12:33

Thanks to all for running the poly. I really appreciate it. Hoping to dip my toe in the GPU pond next year.


[QUOTE=debrouxl;349907]Aiming at setting new GNFS records for XYYXF, Sean ? :smile:

My GT540M is weak, others can help more efficiently...[/QUOTE]

Nah, I'm just trying to eliminate all GNFS composites from the xyyxf project. C169 is about as large as I dare try solo.

It's fun to be on the record list, but I think it's more the result of Moore's Law than anything else. A C165 fell to GNFS on my i7 in a month. That was probably unimaginable 3-4 years ago. I can see in the not so distant future where a C175 GNFS is a fire-and-forget hobbyist-friendly effort on a home rig. How about GPU sieving. Is it theoretically possible?

If anyone wants to set the high water mark for xyyxf, there's [url=http://www.factordb.com/index.php?query=147%5E136%2B136%5E147]C197_147_136[/url]. It's the largest remaining GNFS. NFS@Home interested?:smile:

wombatman 2013-08-17 14:17

I'll go ahead and run from 80M on the C169.

firejuggler 2013-08-17 19:30

So far, no good on the C169.
I encourage everyone to post their poly. should they be bad or not. I want competition, if only by name. Best score for now is 3.615e-013 (with a leading coef of 12M and a skew of 6.7M) while the excepted range is between 4.05e-013 and 4.66e-013. I'll dig around to see if I can do any better.

VBCurtis 2013-08-18 02:07

The last C169 I ran for Swellman in Mid-July produced a best score of 3.32 from 100kb of -nps hits 4e22 or better. I don't recall the winning score- the 3.32 was my best.

That 3.61 looks pretty strong! Don't believe the forecast for every size.

I'm running at coeff=44M.
-Curtis

wombatman 2013-08-18 14:21

Here's the best I've gotten so far on the C169:

[CODE]
expecting poly E from 4.05e-013 to > 4.66e-013
R0: -108814122533272592708956055223543
R1: 35795218528797959
A0: 5392088451186473967300858158426027843200
A1: 3334511055750972840213938071872634
A2: -1143844152923197626854541213
A3: -2857037486590150968874
A4: -33274016097472
A5: 80442960
skew 3448026.18, size 1.578e-016, alpha -7.624, combined = 3.730e-013 rroots = 3[/CODE]

fivemack 2013-08-18 17:21

Thanks for all your work on the C176.

I've done test sieving on all eight of the polynomials presented; for scoring, I estimated some parameters from one of the polynomials, ran 100kQ (160-190k relations, about 18 CPU-hours) on each polynomial, listed the times to get each successive 10,000 relations for each polynomial, listed them in rank order, and worked out the average of the time-ranks of each polynomial.

VBCurtis's polynomial of [url]http://www.mersenneforum.org/showpost.php?p=347257&postcount=69[/url] was the clear winner, with the two polynomials of [url]http://www.mersenneforum.org/showpost.php?p=347455&postcount=79[/url] neck and neck in second place. The ranking's been pretty stable since 30kQ, so probably you don't need to test-sieve much beyond that - say fifty thousand relations per polynomial.

firejuggler 2013-08-18 17:43

my best on the last C169
[code]
polynomial selection complete
R0: -158901476678054196574790648790253
R1: 273171512097277613
A0: 119772796240168711849603868646487794265920
A1: 34101690350238626462893755265184824
A2: -19888694759843446875301033386
A3: -3020714507634859404531
A4: 574986327565814
A5: 12113640
skew 6738936.87, size 1.541e-016, alpha -8.050, combined = 3.615e-013 rroots = 5
[/code]
And just for the sake of it, I'll look around 700M

lorgix 2013-08-18 18:54

I'm taking a whack at C169_141_71 as well. Starting from 60M, don't know how long I'll let it run.

Did anyone find a poly for (5591^61-1)/((5591-1)*16556099215542617537*743213379283195327995487*11686924821525596917649777)
?

I have polys for the other candidates I posted.

I also have a bunch of GNFS-candidates (p^q-1) in the 160s. I'm working on pretesting them.

firejuggler 2013-08-18 19:04

Not that I've seen lorgix.

wombatman 2013-08-18 21:06

Lorgix, I went ahead and add your number to my document of polynomial requests. If anybody wants at it, just post and I'll update it.

[url]https://docs.google.com/spreadsheet/ccc?key=0AlFp2DvBLxsUdEtUMFE0bmk3blRQQlJhS2NkcEF2b0E&usp=sharing[/url]

Also, new best on the C169:

[CODE]expecting poly E from 4.05e-013 to > 4.66e-013
R0: -108714164971617773602170358979662
R1: 448108841571013439
A0: -295061681144065041864936334197085476805
A1: 1065699402893288703728377652011076
A2: 1620565059022284300728763742
A3: -197667938698960378456
A4: -420679333037017
A5: 80813460
skew 1916604.19, size 1.876e-016, alpha -6.662, combined = 4.106e-013 rroots = 5[/CODE]

wombatman 2013-08-18 21:22

Lorgix, I'll take a run on the number you just posted (the 157 digit one) starting at 20M.

VBCurtis 2013-08-18 21:36

After 1 day, by best on Swellman's new C169 is just 3.68. That 4.10 looks like a winner! I have another 12 hrs of GPU data to process, then I'll move on to Lorgix' 5591^61-1 posting.

-Curtis

VBCurtis 2013-08-18 21:58

[QUOTE=lorgix;350043]I'm taking a whack at C169_141_71 as well. Starting from 60M, don't know how long I'll let it run.

Did anyone find a poly for (5591^61-1)/((5591-1)*16556099215542617537*743213379283195327995487*11686924821525596917649777)
?

I have polys for the other candidates I posted.

I also have a bunch of GNFS-candidates (p^q-1) in the 160s. I'm working on pretesting them.[/QUOTE]

I ran a half-day on this number back in July, finding a 1.71e-12 score; I don't know why I aborted the search, nor why I didn't post even this weak poly. We'll get something good for it this week.

firejuggler 2013-08-19 04:37

another poly for swellman C169 :
[code]
R0: -70574329567623593855850350397599
R1: 319115713422217939
A0: 209723292287328832289802943774706861280
A1: 598429061904895552641271174461560
A2: -4062997052989133800994001071
A3: -4271517584968935687318
A4: 5799959550341004
A5: 700939512
skew 899896.26, size 1.358e-016, alpha -7.181, combined = 3.349e-013 rroots = 5
[/code]
Not very good,I know.

wombatman 2013-08-19 13:18

Here's an initial hit for Lorgix's C157:

[CODE]expecting poly E from 1.99e-012 to > 2.29e-012
R0: -754734928404993486930393168259
R1: 43285768764248873
A0: 4802110579359003727584901710197368440
A1: -85570950573115301637711491765378
A2: -99931221296503445803925939
A3: 135339915234481955812
A4: 31283765611140
A5: 20077200
skew 1126185.69, size 2.219e-015, alpha -7.919, combined = 1.782e-012 rroots = 3[/CODE]

firejuggler 2013-08-19 21:12

Lorgix C157
[code]
R0: -1103437414754847615838977449859
R1: 55750041186014801
A0: 873086791320246604576954052331292964
A1: -6423816204639220626317306638268
A2: 15482888406301730540743393
A3: 18367616204405016272
A4: -8257072567026
A5: 3005640
skew 1118845.74, size 2.482e-015, alpha -6.748, combined = 1.942e-012 rroots = 1
[/code]

swellman 2013-08-19 22:17

Many thanks for all the GPU brought to bear on this latest C169. This stuff is really impressive.

I'll run some test sieving this weekend, but until then please keep those polys coming (if you're still searching that is).

And if anyone needs assistance on a non-GPU task, just drop me a line. I'm here to help. :grin:

firejuggler 2013-08-19 22:23

Run 2014's NY marathon and give me all the glory swellman, that's all I ask.. :lol:

swellman 2013-08-19 22:36

Sure, no problem, as long as I can use a Segway...

lorgix 2013-08-20 07:57

[QUOTE=swellman;350168]Many thanks for all the GPU brought to bear on this latest C169. This stuff is really impressive.

I'll run some test sieving this weekend, but until then please keep those polys coming (if you're still searching that is).

And if anyone needs assistance on a non-GPU task, just drop me a line. I'm here to help. :grin:[/QUOTE]
I ran stage1 over leading coeff 60M to 61.434M. That took 21hrs.
Size optimization is moving (leading coeff) at about 10000/hour. I'm using stage2_norm=4e22.
If I let it finish I should have almost 3000 candidates to sort before running -npr on some of them.

VBCurtis 2013-08-21 00:51

My best for the last C157 is 1.84:
[CODE] n: (5591^61-1)/((5591-1)*16556099215542617537*743213379283195327995487*11686924821525596917649777)
R0: -942530367041324355849999977433
R1: 66386936784413669
A0: -11627136426338468347417478381938863430
A1: 31537335140191737468233407352821
A2: 55719258620543025550987018
A3: -58361329266411186886
A4: -34306254782868
A5: 6609960
skew 1395583.72, size 2.337e-015, alpha -6.965, combined = 1.845e-012 rroots = 5 [/CODE]
This was 2 days on the GPU, and ends my search on this number.

wombatman 2013-08-21 13:34

My last best for the C157 is:

[CODE]R0: -1102225542723473838562347398326
R1: 46997859610529773
A0: -58978494067576752313653893703695131059
A1: 69333563221774646199876658034895
A2: 169011075619257748542422093
A3: -2025152247490881835
A4: -20641467903774
A5: 3022200
skew 2469855.09, size 2.244e-015, alpha -7.219, combined = 1.797e-012 rroots = 3[/CODE]

Looks like the 1.942e-12 is the best.

swellman 2013-08-22 21:30

[QUOTE=lorgix;350203]I ran stage1 over leading coeff 60M to 61.434M. That took 21hrs.
Size optimization is moving (leading coeff) at about 10000/hour. I'm using stage2_norm=4e22.
If I let it finish I should have almost 3000 candidates to sort before running -npr on some of them.[/QUOTE]

Did any good polys fall out? Again my thanks for the help.

lorgix 2013-08-23 11:19

[QUOTE=swellman;350554]Did any good polys fall out? Again my thanks for the help.[/QUOTE]
I stopped -nps early and started -npr. I can finish -nps if you have time.

-npr is now running on the best candidates from the range of leading coeff 60M to 60.9M.

[CODE]C169_141_71

R0: -115274200224198355694855896328260
R1: 302768509259685271
A0: -201242861993156913967388884915385225277
A1: 882968486254771001548376947426776
A2: 1629011842745093689627719511
A3: -592286320350946094372
A4: -450873229496418
A5: 60291180
skew 1893745.33, size 1.561e-016, alpha -6.434, combined = 3.707e-013 rroots = 5[/CODE]Maybe I can find a better one. Let me know if I should stop.

swellman 2013-08-23 12:20

I can wait for as long as it takes. Not on any timetable.

Thanks again.

lorgix 2013-08-23 16:35

[QUOTE=swellman;350625]I can wait for as long as it takes. Not on any timetable.

Thanks again.[/QUOTE]
OK, I'll give it some more effort then.

What norm did the rest of you use for -npr?

Posting this one because of the crazy alpha-value;
[CODE]C169_141_71
R0: -115352069094744576726520403778903
R1: 686495812988035681
A0: -163446241091275491921780732361558730649640
A1: -130930414646634523589695517244566348
A2: 16371793575226784454320614810
A3: -2136772890666075261961
A4: -943961642526546
A5: 60087960
skew 7293821.18, size 1.487e-016, alpha -8.930, combined = 3.535e-013 rroots = 3[/CODE]

VBCurtis 2013-08-23 17:00

I use the default norms for -npr. I set a bound on the e-value just to keep output brief- that does not alter the polys found.

I did set stage2norm of 3.5e22 while doing -nps, and then ran -npr on the entire output.
-Curtis

wombatman 2013-08-23 20:10

I normally reduce the stage 2 norm by one order of magnitude from the default choice and pull the top 100 or 200 (usually 200) results from -nps with sort -g -k 10 file.ms | head -200 > newfile.ms. Then I'll run -npr, using a minimum e-value if I already have something and am only looking to better it.

frmky 2013-08-26 23:46

I've just started the polynomial search for 3,766+. Any help you wish to provide would be appreciated.

Just to set a baseline, here's the best so far:

[code]# norm 9.900789e-16 alpha -9.251101 e 4.224e-16 rroots 2
skew: 9073870.85
c0: 139221485278577320652164834978699370706236308160
c1: -1295658612562855727156808735812558835809336
c2: 163771486239390907909286755592113710
c3: 43656819592951246110772775111
c4: -3890115263418711871319
c5: -494769166057413
c6: 8200236
Y0: -58027728060921253760709506018956107
Y1: 72038266426478147627

# norm 9.808488e-16 alpha -8.780447 e 4.199e-16 rroots 6
skew: 7050029.50
c0: -220945970580033192482207350438562369695523503600
c1: -366418378394057555130756972699810943184540
c2: 65127417239769823903620404282746080
c3: 51919154391212047912008400199
c4: -2144689123321499659289
c5: -528344113932981
c6: 8200236
Y0: -58027728110080022693605341003325653
Y1: 72038266426478147627
[/code]

firejuggler 2013-08-27 00:02

edit :oops, C6.. nevermind

wombatman 2013-08-27 13:25

I'll start at 10M and work toward 12M or so on 3,766+.

Edit: Or at least I will once I figure out how to get msieve to work on numbers > 311 digits...

firejuggler 2013-08-27 13:37

hmm? get to factordb, type 3^766+1; find the C216...
and don't forget to add "polydegree=6" in your parameters
[code]
313068751706172934164029607615444196674154606234575242211641355472713959762783712390895701060952639425447396356557581585630115258626457263968597368593309998959766117756941753233822835800989497155176060903761800503381[/code]

wombatman 2013-08-27 14:41

I should have known it was a smaller composite. My bad. I'll start my run tonight.

VBCurtis 2013-08-28 05:58

I estimate 7-10 GPU-months to find a poly for this? wowsers.

I'll give it a GPU-week or two, after the heat in the IE subsides a little.

frmky 2013-08-28 06:48

I'm giving it about 6 GPU-months over 2 weeks total. Sieving will start in about a week or so.

bdodson 2013-08-28 16:10

[QUOTE=firejuggler;351022]hmm? get to factordb, type 3^766+1; find the C216...
and don't forget to add "polydegree=6" in your parameters
[code]
313068751706172934164029607615444196674154606234575242211641355472713959762783712390895701060952639425447396356557581585630115258626457263968597368593309998959766117756941753233822835800989497155176060903761800503381[/code][/QUOTE]

Yes; a smaller composite, C216. Perhaps it is worth recalling that
the two previous gnfs from the 3+ extension, C202 (745+) and
C207 (706+) both had p62 factors, left after tests to p60; perhaps
2t60 if Sam finished an initial t60. Since there's already an extensive
polyn search in progress, perhaps it is too late for "more" ECM.

Ah; looks like I already finished my t60, so this number is no more
under-tested than the previous two. Seems that there's a sharp
difference between these and the numbers from pre-extension
Cunninghams that ryanp is sieving; no p6x's, hardly any p7x's.
Looks like his smallest snfs factor is a p68 from 7,334+ C205 ---
just slightly smaller than his spectacular ECMs p70, p69 and p68
from the oldest Cunninghams, 2^N-1, 1000 < N < 1200;
N = 1069, 1051 and 1067, from the Mersenne list. Maybe Bob's
right that lots of curves were run, by lots of people, to remove
most p<70s.

In any case, this gnfs216 is a large step up from the current
gnfs212; probably still twice as difficult for the extra 4-digits?
With some advance notice, we might have run a larger proportion
of a t65. Speaking of which, looks like Greg has reserved two more
[code]
7,394+ c197 NFS@Home gnfs
10,770M c212 NFS@Home gnfs
3,766+ c216 NFS@Home gnfs
11,323+ c221 NFS@Home gnfs [/code]

-Bruce

PS - Ah; a nice p69 from 2, 2186L C227.

frmky 2013-08-28 19:19

Yes, this one sneaked up on me. It's been a really crazy summer, and it just dawned on me last week how quickly the current c197 will finish. I quickly put all my gpus on it. :smile:

There is time to give 11,323+ some more ECM if it's needed.

Thanks!

bdodson 2013-08-28 21:15

[QUOTE=frmky;351166]Yes, this one sneaked up on me. It's been a really crazy summer, and it just dawned on me last week how quickly the current c197 will finish. I quickly put all my gpus on it. :smile:

There is time to give 11,323+ some more ECM if it's needed.

Thanks![/QUOTE]

Yes, I see that we've switched to the c197, now that you
mention it. This one had a bunch of B1=900M curves,
23834-of-13061=t60, just on the dual-8core machine, during
burn-in.

For 11,323+ I have 3t55; I'll plan on adding 2-or-3 t60's,
depending how they go.

Looking forward to hear progress on postprocessing of 10, 770M. -Bruce

VBCurtis 2013-08-29 03:07

Starting a week on the c216 today- glad you pointed out it's going to begin sieving quickly.

This will stretch the 16e siever quite a bit, yes? Can you compare the parameters for this vs the c212? Are you confident the c221 will be possible with 16e (perhaps that depends on how the 216 goes?).

henryzz 2013-08-29 18:48

It should be possible to raise the large primes limit(a recompile after removing the check will be necessary). I have checked that the siever and msieve work with larger large primes(I think I checked 35 bits. It will have been at least 34. I think jasonp reckoned that >35 would be too many relations for msieve to handle). I haven't checked that it doesn't miss some of the relations with larger large primes but if that is an issue it will be discovered during parameter selection anyway.
I would imagine after raising the large prime limit there is a fair amount more room in 16e. You could also try sieving with >3 lp using the lasieve5 siever.

frmky 2013-08-29 21:15

Here's a somewhat better one for 3,766+:
[CODE]# norm 1.088509e-15 alpha -10.727690 e 4.501e-16 rroots 6
skew: 10476738.56
c0: -5369554424004030826402973140282841011572519398275
c1: -3414866216469788096538063790563208721226714
c2: 854150826092935212357938287070651342
c3: 341347932610455553798257621420
c4: -19812274695016079944507
c5: -1277018966659986
c6: 9218160
Y0: -56907029561383080746868850208970432
Y1: 20291767539399957581
[/CODE]

Gimarel 2013-08-30 13:54

[QUOTE=frmky;351287]Here's a somewhat better one for 3,766+:
[/QUOTE]

Although the skew is very high, this degree 5 poly seems to sieve 10%-20% better:

[CODE]# norm 2.316186e-21 alpha -8.836349 e 3.282e-16 rroots 3
skew: 1161620713.03
c0: -150197586295000094627583800604174270758750606052640635
c1: -972901144347733530674190364546696103274696609
c2: -588214638592090440436049221701783597
c3: 2579751407152116188605093717
c4: -122719450234339820
c5: 102342240
Y0: -314135458405637114606970259658568784545142
Y1: 131759766057910809619
[/CODE]

But that probably needs to be verified with the actual sieving parameters and for the Q-range to be sieved.

lorgix 2013-08-30 20:04

[QUOTE=swellman;350625]I can wait for as long as it takes. Not on any timetable.

Thanks again.[/QUOTE]
I'm done with C169_141_71. I didn't find anything better than the 3.7 I posted.

swellman 2013-08-30 21:09

[QUOTE=lorgix;351404]I'm done with C169_141_71. I didn't find anything better than the 3.7 I posted.[/QUOTE]

Thanks for all your efforts. I'll do some test sieving next week.

wombatman 2013-08-31 15:34

It's not better than the polys posted, but here's my best so far on the C216:

[CODE]#skew 4318719.56, size 6.192e-016, alpha -9.299, combined = 3.282e-016 rroots = 6
N 313068751706172934164029607615444196674154606234575242211641355472713959762783712390895701060952639425447396356557581585630115258626457263968597368593309998959766117756941753233822835800989497155176060903761800503381
SKEW 4318719.56
R0 -56118476053099371919661459200054844
R1 3376910376775132711
A0 422506237083294610548658191220217155380023123625
A1 838036371757770978178729323549824699235795
A2 202841735403297089138679493763867919
A3 -220430066293747906637149137727
A4 -15892839097395096736892
A5 7154030015548284
A6 10023156[/CODE]

firejuggler 2013-08-31 20:09

I do not have any luck either
[code]
R0: -43691089618675121990463269690614331
R1: 8280875056331246431
A0: 882667634292843493128109607284600675569970860000
A1: 1045365354762879660059156791608455967589860
A2: -262046219683142179195810002182539336
A3: -232917903025869392497781403589
A4: 37574151948391299096992
A5: 4597605901709826
A6: 45007248
skew 5344180.09, size 7.638e-016, alpha -9.814, combined = 3.854e-016 rroots = 6
[/code]

lorgix 2013-08-31 20:13

Well, that's fourth place among the ones posted in this thread.

What parameters are you using?

firejuggler 2013-08-31 20:34

beside the polydegree 6? stage2_norm=1e27

wombatman 2013-08-31 20:40

And a slightly better one:

[CODE]#skew 4046323.72, size 6.980e-016, alpha -10.086, combined = 3.539e-016 rroots = 6
N 313068751706172934164029607615444196674154606234575242211641355472713959762783712390895701060952639425447396356557581585630115258626457263968597368593309998959766117756941753233822835800989497155176060903761800503381
SKEW 4046323.72
R0 -56080138255852003624871619361635946
R1 9812788240400569969
A0 -217428344690088236980111921227628323587881810755
A1 -304896515339816524240109075958463638965304
A2 653295607884514249440865324339558838
A3 11482264525447064228014465568
A4 -155933079432204874141015
A5 -4263623035233592
A6 10064340[/CODE]

RichD 2013-09-01 04:37

c148 poly?
 
This might be too little of a request. Aliquot Sequence 3408 is approaching GNFS readiness. ECM to nearly t50.

The last term is [URL="http://factordb.com/sequences.php?se=1&aq=3408&action=last&fr=0&to=100"]here[/URL].

The C148 number is [URL="http://factordb.com/index.php?id=1100000000630792634"]here[/URL].

Let us know if we should just do a CPU poly search.

VBCurtis 2013-09-01 07:32

C148 is definitely enough for one of the GPUers to tackle it. I imagine one or more of us will post polys for you Monday.

My best for the C216 is 3.87, with alpha 9.40. I used a variety of stage 2 settings, settling on 1.1e27 as plenty wide enough, and only two -nps hits below 3e26 (one 1.2e26!). I left stage 1 default.

If I'd had any hits above 4.00, I would have messed with -npr settings; my best hit had norm 9.2e-16.
-Curtis

wombatman 2013-09-01 13:47

Still working on the C216, but I'll run the C148 as well. I'll run from 7M headed to 10M.

firejuggler 2013-09-01 13:57

Stopped working on the C216, will work on the C148

VBCurtis 2013-09-01 20:51

First hit on c148 is low: 5.16e-12. This is the second time around c150 that I get no hits from npr at the default msieve settings! Seems we still have some parameter-tweaking on the default settings to do.
-Curtis

frmky 2013-09-01 22:00

A small improvement on the C216:

[CODE]# norm 1.151728e-15 alpha -9.599450 e 4.614e-16 rroots 6
skew: 2050199.35
c0: -1579353278635676715631290757256478191243559744
c1: -3429903439852403145526446933945902857792
c2: 36880018074606679524160256525283636
c3: -13697458855764052177232835104
c4: -47728830331085590774493
c5: 190187915747686
c6: 3219216
Y0: -67813308479544618717746782002380725
Y1: 11562712847052532201
[/CODE]

frmky 2013-09-01 22:19

If you want to play with -npr settings, the 4 best polys come from these two hits from -nps:

[CODE]3219216 188896147381798 -47792211291020524851873 -920936794704352912477068289 106160083630569026815528289307099010 801811354690321198689852105715175167948 -30607416062921875004876714127680707377837022015 11562712847052532201 -67813308480317909827531961250919203 -2.10 4.076245e+26
9218160 -1113893264248146 -37441391074710636811327 1260310938060465689277990112 19301218289127125020178978701622320 -254628563593934339925002254936864479487 -1291403264521702360220048668647081873991457257 20291767539399957581 -56907029501535474987469427717617758 -2.55 9.851720e+25
[/CODE]

Interestingly, the best poly so far comes from first of these, which comes close to not making the top-1000 from -nps.

wombatman 2013-09-02 00:17

Interesting. I normally take the top 200 or so for the -npr step.

VBCurtis 2013-09-02 06:37

[QUOTE=wombatman;351595]Interesting. I normally take the top 200 or so for the -npr step.[/QUOTE]

Yes, but we take 200 from a week's work; he's taking 1000 from 5-6 GPU-months. Note the size is well within any bound we would set for this- I had only 12-15 hits better than this "bad" one in 5 days.

wombatman 2013-09-02 12:38

Good point. It's a bit mind-boggling to consider multiple months worth of hits :smile:

On a related note, I've still not found anything really even close to the best polys thus far, but I'll keep on searching for a bit.

firejuggler 2013-09-02 16:00

for the C148
[code]
expecting poly E from 7.18e-012 to > 8.25e-012
R0: -14103494893784526519356005989
R1: 14231751083558387
A0: -1082563197314027058117447692104240
A1: 6763855833638605837425179132
A2: 683918654337887371011610
A3: -1178635426885569059
A4: -25388067192588
A5: 6020820
skew 174228.78, size 2.420e-014, alpha -6.975, combined = 7.292e-012 rroots = 5
[/code]

VBCurtis 2013-09-02 17:45

Nice find! My best was 6.70, with no other hit above 6.10.
-Curtis

EdH 2013-09-02 18:03

re: c148
 
@firejuggler: Thanks!

If I could ask for some more guidance (or, a source to review) in supplying the rest of the necessary elements for the file to be used for sieving, I'll proceed with gathering some relations.

fivemack 2013-09-02 19:18

Thanks for 3270.677 work
 
Hello everyone

3270.677 is now slain.

A slightly larger and older dragon is 8352.1755

[code]4501201576453331166199832789791297593369658735642360757062244642928667372333520186228057700129947196448599558976151228285312199225640499870800004359537419188757959878137081775709[/code]

I've run 7855 curves at 3e8, which seems to have taken around a CPU-year; if anyone has spare computrons and would like to try for a polynomial, I would appreciate it greatly.

lorgix 2013-09-02 19:26

Two days on the C216 didn't produce anything better than this:
[CODE]R0: -55254695430285730547103940791165762
R1: 312119293759135507
A0: -30293185997072202023126406251938943320224112555
A1: 39022686988409701912269268856413080821476
A2: 308597010425180289043844970668583500
A3: 95335518703809504663944362
A4: -187151833690387205781241
A5: 2008030111355554
A6: 11000808
skew 2896044.67, size 6.043e-016, alpha -9.361, combined = 3.125e-016 rroots = 6[/CODE]The rest of the top:
3.089
2.954
2.943
2.910

wombatman 2013-09-02 19:34

[QUOTE=fivemack;351648]Hello everyone

3270.677 is now slain.

A slightly larger and older dragon is 8352.1755

[code]4501201576453331166199832789791297593369658735642360757062244642928667372333520186228057700129947196448599558976151228285312199225640499870800004359537419188757959878137081775709[/code]

I've run 7855 curves at 3e8, which seems to have taken around a CPU-year; if anyone has spare computrons and would like to try for a polynomial, I would appreciate it greatly.[/QUOTE]

Added to the list ([url]https://docs.google.com/spreadsheet/ccc?key=0AlFp2DvBLxsUdEtUMFE0bmk3blRQQlJhS2NkcEF2b0E&usp=sharing[/url]), and I'll take a whack at it in a day or so.

firejuggler 2013-09-02 19:41

Working on fivemack's C178. Leading coef of 26M to 27M for now. stage 2 bound is normal one/10 +1 rounded down

edit : ok, the skew is still too high, 106-107M and stage2_norm dropped to 1e24

lorgix 2013-09-02 20:02

[QUOTE=fivemack;351648]Hello everyone

3270.677 is now slain.

A slightly larger and older dragon is 8352.1755

[code]4501201576453331166199832789791297593369658735642360757062244642928667372333520186228057700129947196448599558976151228285312199225640499870800004359537419188757959878137081775709[/code]I've run 7855 curves at 3e8, which seems to have taken around a CPU-year; if anyone has spare computrons and would like to try for a polynomial, I would appreciate it greatly.[/QUOTE]
Here's something to improve upon:
[CODE]R0: -13510300690714233128962473784787911
R1: 6033666436834932973
A0: -7034748373882411992025425942353529267330539568
A1: 71338558289372431896711534793022720038
A2: 14748583117555536229325143815075
A3: 26578335541599568634856
A4: -4478227283242628
A5: 10000080
skew 62604670.38, size 1.049e-017, alpha -8.504, combined = 6.838e-014 rroots = 5[/CODE]And one with lower skew:
[CODE]# norm 1.268524e-017 alpha -7.678027 e 6.640e-014 rroots 5
skew: 17111756.25
c0: 210599198530084596051506780338272197545928
c1: 2527411309030685160782799540300491498
c2: 1065181840417068104646006801855
c3: -364038192920007149521688
c4: -3207651518717828
c5: 10000080
Y0: -13510300537390852798621699508797335
Y1: 6033666436834932973[/CODE]

firejuggler 2013-09-02 20:25

hmmm, sorry lorgix ^^
edit : and as soon as I post, I get something better...
edit2: and one again....
[code]
R0: -8424609947188589022080303049239527
R1: 2883303214296225487
A0: -1025957941551236123340742630353362456769120
A1: 23139567493487836324707827126723976
A2: 336879047668564417814061593030
A3: 25774464945517024330291
A4: -11994763228641670
A5: 106066800
skew 5914607.03, size 1.586e-017, alpha -7.883, combined = 9.118e-014 rroots = 5
[/code]

wombatman 2013-09-02 21:03

firejuggler, what's the expected score on this one?

firejuggler 2013-09-02 21:07

expecting poly E from 1.06e-013 to > 1.22e-013

firejuggler 2013-09-03 04:18

one slightly better
[code]
R0: -8407143352811808814320719097477087
R1: 1093231323027035599
A0: -1033043932332288447826927595522335408816144
Tue Sep 03 06:14:22 2013 A1: -17897682125008254951716401225227540
Tue Sep 03 06:14:22 2013 A2: 87833614330134391302128104568
A3: -10121373174071281106151
A4: 544593508546310
A5: 107173200
skew 7910402.55, size 1.751e-017, alpha -7.557, combined = 9.911e-014 rroots = 3
[/code]
tried to mess with npr stage to no avail

wombatman 2013-09-03 13:22

Doc is updated with the newest high score. I'm taking a quick break from the C216 and running a day or so on the C178 starting at 200M. Still nothing to top the best on the C216 or C148, but I'm still running the C148 as well and will return to the C216 soon.

bdodson 2013-09-03 14:41

[QUOTE=bdodson;351151] ...
In any case, this gnfs216 is a large step up from the current
gnfs212; probably still twice as difficult for the extra 4-digits?
With some advance notice, we might have run a larger proportion
of a t65. Speaking of which, looks like Greg has reserved two more
[code]
7,394+ c197 NFS@Home gnfs
10,770M c212 NFS@Home gnfs
3,766+ c216 NFS@Home gnfs
11,323+ c221 NFS@Home gnfs [/code]
...[/QUOTE]

OK; 11,323+ is at 3t60 =.c 60% of a t65. That's
the same amount of ecm pretesting as 10,770M; while
gnfs 221 is a lot harder, four times at least. Probably
enough to have found a p62; but there's still lots of
space in [p63,p79] for factors in ecm-range. I'm
taking a break; pending 10,770M. -Bruce

EdH 2013-09-03 18:04

[QUOTE=wombatman;351725]Doc is updated with the newest high score. I'm taking a quick break from the C216 and running a day or so on the C178 starting at 200M. Still nothing to top the best on the C216 or C148, but I'm still running the C148 as well and will return to the C216 soon.[/QUOTE]
We are using the c148 poly provided by firejuggler yesterday. Thanks to all who have assisted.

wombatman 2013-09-03 18:14

Good deal!

firejuggler 2013-09-03 18:17

half joking, but should I refrain from posting my poly until someone else has posted something?

wombatman 2013-09-03 18:21

Definitely not! Post away. It helps give a good benchmark to aim for so we have an idea of what can be reached, score-wise.

wombatman 2013-09-04 00:30

For the C178:

[CODE]# norm 2.336437e-017 alpha -8.750154 e 1.042e-013 rroots 1
skew: 8101608.89
c0: -1124802423237697198907119489783480071149680
c1: -974204032868244810143592284577125356
c2: -186990917123318951862940976624
c3: 8439029065470934693477
c4: -4981313244457890
c5: 200514600
Y0: -7417140276099365940084623920414947
Y1: 820410665804958487[/CODE]

Gimarel 2013-09-04 19:15

Here's a much better degree 5 poly for 3,766+

[CODE]# norm 2.775079e-21 alpha -8.960946 e 3.612e-16 rroots 5
skew: 63077872.55
c0: 1887020519362839888115657135050861565577526397568
c1: 1063754686294298940104013857297655722687416
c2: -487067776713547205197156932678978146
c3: 10205037559243136766266083631
c4: 110430678512311392366
c5: 100503683280
Y0: -79193866781948524407733628428745627292279
Y1: 8357656079342536381
[/CODE]

wombatman 2013-09-04 19:27

Is the expected score different for the degree 5 polynomial? The best degree 6 score is 4.614e-16

frmky 2013-09-04 22:01

They aren't directly comparable. I will test this degree 5 poly and the best degree 6 I find to see which sieves better. Thanks!

Gimarel 2013-09-05 15:36

[QUOTE=frmky;351949]They aren't directly comparable. I will test this degree 5 poly and the best degree 6 I find to see which sieves better. Thanks![/QUOTE]

My first degree 5 poly seems to sieve better for higher special-Q, you should consider it too.

What range do you plan to sieve?

lorgix 2013-09-06 19:26

Here are the best two I've found so far for the C178:

[CODE]# norm 2.074604e-017 alpha -8.370658 e 9.554e-014 rroots 5
skew: 27407813.51
c0: -34479147319693175161599349315340052834699584
c1: 4429282378256539747970358196795900328
c2: 1434881867303190777125708901302
c3: 33628945964909798531803
c4: -1929401141291154
c5: 10000080
Y0: -13510300181730257396835568376043747
Y1: 814102016366565773
# norm 2.003266e-017 alpha -7.203611 e 9.529e-014 rroots 3
skew: 13010929.25
c0: 976352382201425037872835478818751759346434
c1: -274755295225894962489202667192849721
c2: -115707756918283545938338351718
c3: -5529842672220826821354
c4: 1055486076049144
c5: 10029720
Y0: -13502305455454092085413920152946925
Y1: 3084936245691597311[/CODE]I've found eight ones >9e-14.
Maybe the one wombatman found is a winner.

fivemack 2013-09-07 08:26

From trial sieving it looks like firejuggler's [url]http://www.mersenneforum.org/showpost.php?p=351690&postcount=169[/url] (c5=107173200) is the best of the C178 so far, with wombatman's [url]http://www.mersenneforum.org/showpost.php?p=351781&postcount=176[/url] in second place.

I think we're ready to think about sieving that one; may I propose

4788.5154

[code]
17285154910805941577069464828335617544658066950627644021728302169526833018711670895092479561808160256160945139573800969912234390238908363042669550995167201537635764747005337
[/code]

as a next polynomial-selection target? It's received an enormous amount of ECM from yoyo@home - I suspect twice as many cycles as will be required for the sieving.

wombatman 2013-09-07 13:34

Sure. I'll start at 100M. I've updated the document with it as well.

firejuggler 2013-09-07 13:45

ok, i'll start at 150 M, then.
Since I get away from home Tuesday afternoon, I won't be able to do much.
expecting poly E from 2.27e-013 to > 2.61e-013

firejuggler 2013-09-07 14:22

and here is a first look at it
[code]
R0: -649106117267591846867343899276661
R1: 1476313446490191469
A0: 10782780581991004974082092246843346818160832
A1: 9872003751360519183436215391887794588
A2: -1198519317595998643602562010588
A3: -107539073258193732676277
A4: 4960166814080970
A5: 150001200
skew 15362972.91, size 4.139e-017, alpha -8.994, combined = 1.602e-013 rroots = 5
[/code]
Getting in the excepted poly range should be easy.

firejuggler 2013-09-07 16:31

a better one :
[code]
R0: -649106121065203735403440768762520
R1: 829423026522415277
A0: 11840833659022501956266274797457992656701
A1: 5416933874751647390737163098891614
A2: -28064405985398111945180071556
A3: -1010129301483599319579
A4: 5394748664135520
A5: 150001200
skew 2484106.52, size 5.271e-017, alpha -7.341, combined = 1.895e-013 rroots = 5
[/code]
messing with npr sometime work : "natural" poly was ( same NPS hit)
[code]
R0: -649106121436268520739959348140949
R1: 829423026522415277
A0: 5795439107163174921644154112805635510096
A1: 5300282288863527562360450565041500
A2: -20364562350473250910946757532
A3: -10363853521995763895739
A4: 5059213229873520
A5: 150001200
skew 2183932.81, size 4.821e-017, alpha -6.973, combined = 1.784e-013 rroots = 5
[/code]

wombatman 2013-09-07 19:12

Here's a nice jump that puts us closer to the expected:

[CODE]R0: -703882715400154767239221030785027
R1: 109256907036035759
A0: -32458290986930814614288309447638552942448
A1: 103153855552643435982140704058765640
A2: 30534525659025455419966454893
A3: -8313312278444358634242
A4: -994692832026932
A5: 100039560
skew 5450899.35, size 5.768e-017, alpha -7.498, combined = 2.022e-013 rroots = 5[/CODE]

firejuggler 2013-09-07 22:11

[code]
R0: -697013904459768580991097397641011
R1: 5606793472754501
A0: -1847971520796502134306506190386798968500
A1: 40821386607537415162180472039027730
A2: 65031139921790296595480109002
A3: 11389536971489005146365
A4: -2922818243778546
A5: 105066936
skew 4418742.30, size 5.992e-017, alpha -7.710, combined = 2.022e-013 rroots = 5
[/code]
exact same score

wombatman 2013-09-07 23:39

Hahahaha, nice!

frmky 2013-09-08 03:10

[QUOTE=frmky;351949]They aren't directly comparable. I will test this degree 5 poly and the best degree 6 I find to see which sieves better. Thanks![/QUOTE]

For the C216, Gimeral's e 3.612e-16 degree 5 poly sieves about 15% better than a degree 6 poly with e 4.662e-16.

VBCurtis 2013-09-08 06:25

[QUOTE=frmky;352366]For the C216, Gimeral's e 3.612e-16 degree 5 poly sieves about 15% better than a degree 6 poly with e 4.662e-16.[/QUOTE]

Does this suggest that perhaps msieve's root-opt for deg 6 is rough enough that the theoretical "line" for deg 5 vs deg 6 is lower than real-world work?

Are the CADO root-opt tools better for deg 6? Perhaps someone could collect the best few hundred hits for a CADO root-opt run?

firejuggler 2013-09-08 08:46

after running for the night, I got
[code]
R0: -696455968474482254849269323572607
R1: 348684257076673921
A0: 214663815990857477983089978209004516580928
A1: 177424600277632306484616564575041074
A2: -34342254092265056219064954433
A3: -8565004414637263688679
A4: 546856311911345
A5: 105488460
skew 6382404.47, size 6.264e-017, alpha -7.953, combined = 2.130e-013 rroots = 3
[/code]
for the latest composite

jasonp 2013-09-09 13:23

[QUOTE=VBCurtis;352382]Does this suggest that perhaps msieve's root-opt for deg 6 is rough enough that the theoretical "line" for deg 5 vs deg 6 is lower than real-world work?

Are the CADO root-opt tools better for deg 6? Perhaps someone could collect the best few hundred hits for a CADO root-opt run?[/QUOTE]
I suspect it's less the crossover point between degree 5 and 6 than the scaling factor between degree-5 E-values and degree-6 E-values. Hopefully everyone knows but it's worth repeating: the E-value algorithm was invented to compare polynomials of like degree only.

My anecdotal experience is that the best output from the CADO root-opt tools is noticeably better than what you get with Msieve, but on average they work equivalently well. Even among same-degree polynomials, the definition of E-value would make one think that a polynomial with X percent better E-value would sieve X percent more efficiently, but experience with RSA768 shows the difference is much smaller than that. i.e. we found lots of degree-6 polynomials with half the E-value score of the one actually used for RSA768, that sieved maybe 15-20% slower.

bdodson 2013-09-09 16:34

2nd effort
 
[QUOTE=bdodson;351728]OK; 11,323+ is at 3t60 =.c 60% of a t65. That's
the same amount of ecm pretesting as 10,770M; while
gnfs 221 is a lot harder, four times at least. Probably
enough to have found a p62; but there's still lots of
space in [p63,p79] for factors in ecm-range. I'm
taking a break; pending 10,770M. -Bruce[/QUOTE]

OK, never mind the polyn search for 11,323+ C221:
[code]
Input number is 47684588221623639056961705608173079138153779302378955848869404
42648307317037627130989942393222503668627639186419734789253370
97393792502082005378750936019951281361678639264962841236093010
58482691872665246622667910987254487 (221 digits)
Using B1=400000000, B2=15892277350966, polynomial Dickson(30), sigma=4180268258
Step 1 took 3510272ms
Step 2 took 1859204ms
********** Factor found in step 2: 482632031053134403896770035981249734273506307638396123269389497006923
Found probable prime factor of 69 digits: 482632031053134403896770035981249734273506307638396123269389497006923
Probable prime cofactor 98801126227724201982197175315480431013249163755336329531650444
70656245838965370494907444192757868193236504714266221744759846
9879406843807465214724901669 has 152 digits [/code]
on one of the pc's running 8 curves on an i7. This near the end
of a new 7t55. Ah, there was another 1.8t55 of left-overs from
last weekend's count. Looks like my total curve count is c. 23.3t55*,
out of 25t55 for t65. Wow! a new highest count for me; not so
much luck, as sustained effort.

Please apply this to the account with the two p62's "not found"
for not having run enough on the NFS@Home 3+ gnfs's! -Bruce

*PS - OK, that's 127,215 curves with B1=400M, default B2, to
be precise. This p69 is just short of my previous, the current
10th-of-the-top10, 482... -vs- 563...


All times are UTC. The time now is 12:54.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.