Don't wait. It looks as if the PSU could not take the extra load. It won't even turn the fans any more. I'll have to take the PC to a repair shop tomorrow for a new PSU (if that's what failed).

I've already had to buy a better CPU cooler to replace the stock one. Adding the cost of a new PSU to that and it'll about double the cost of the GPU.

Chris

[QUOTE=chris2be8;365553]Don't wait. It looks as if the PSU could not take the extra load. It won't even turn the fans any more. I'll have to take the PC to a repair shop tomorrow for a new PSU (if that's what failed).

I've already had to buy a better CPU cooler to replace the stock one. Adding the cost of a new PSU to that and it'll about double the cost of the GPU.

Chris[/QUOTE]Sorry to hear that. I'll go ahead and run a test today then to see which of the 3 polys sieves faster.

[QUOTE=VBCurtis;365457]If mine sieves close to yours, I'm happy to give it another 24-36hr on the GPU to perhaps find improvement. But when my first effort scored 10% worse than yours, I admit to losing interest.
-Curtis[/QUOTE]My next jobs are going to be larger, so I assume they will benefit more from an intensive GPU search. As it stands, my highest poly and your poly sieved about the same speed, with yours faster by a hair, but the yield for yours was ~13% less..... I should have my answer in ~10 days or so.

OK, thanks for the info. I'll proceed to RichD's C160 once I finish a job for my own composite; I should have a first poly Thursday evening, and will let it run until Sunday. A C160 should have 10+ GPU days of poly search, but I don't have the resources available to do that.

for the C184, I reverted back to quintic, and got
[code]
R0: -189739683797999317529400126331832722
R1: 693895955161095043
A0: 32453832495498028059546140602061149196755545
A1: 1293761121414808264524462998111743706
A2: -228832774841744122536924096105
A3: -20926589075998572306546
A4: -345246593021584
A5: 35193264
skew 19710555.98, size 3.593e-018, alpha -6.391, combined = 3.765e-014 rroots =
3
[/code]

For the C160 from 3408:1385, the first day's search turned up a score 1.19e-12. The window msieve "expects" is 1.33e-12 to 1.52e-12.

I will post any poly that improves on this 1.19 score.
-Curtis

For what it's worth I've run -npr against the best 100 (of 2669441) entries in msieve.dat.ms from my run on Tuesday. I got: [code]
R0: -135858192791394956439864945410
R1: 570849297736093
A0: 14301565586950358401384714963946613561
A1: 16152650324808439814192471692756
A2: 3567237672198900849530610
A3: -879298234954072455
A4: -49781921806
A5: 3864
skew 7818935.60, size 2.043e-14, alpha -7.261, combined = 6.747e-12 rroots = 3
elapsed time 00:09:43
[/code] That's probably not very good since the GPU didn't run for very long before the PSU died. Elapsed time above is for npr only.

Chris

For the C160 from 3408:1385
[code]expecting poly E from 1.33e-012 to > 1.52e-012
R0 -4033130224608614852170208876887
R1 180460901903302283
A0 -52357711339572497186925087177217184640
A1 428629807184883242734908451572136
A2 -111194134081525777552853666
A3 -93001930969062599969
A4 9212558892474
A5 3828720
skew 3102530.12, size 1.335e-015, alpha -7.118, combined = 1.333e-012 rroots = 5[/code]
I have no polys above 1.19 except this one. I'll give it one more day to improve, but this is a usable candidate in my opinion.

I haven't decided whether I want to tackle my next largest composite or go for a "more wanted", but if someone can find a really nice poly for this one, [URL="http://factordb.com/sequences.php?se=1&aq=7044&action=last20&fr=0&to=100"]7044[/URL]:i3426.c159:[code]132849801556061831314309873822050317788619462490949245156896198056485583983802960449476380694570581312461543888860167332690420786216424942137403018401845810663[/code]I'll do the more wanted.

I'll have a go at schickel's C160. I'll only scan a limited range of HLQs, but enough to test my scripts, and the new PSU.

Chris

I didn't get far. After about 27 minutes the system crashed again. This time the PSU survived so I could power the system back up. Running npr on the best 100 entries in msieve.dat.ms found: [code]
R0: -15595511478358973801530755477526
R1: 6316360635713693
A0: 2096409315900123991624591017294407684793
A1: -21208160729383271288378383776159
A2: -79287985949944664147956451
A3: 2064769571596254311
A4: 249346723154
A5: 144
skew 19821182.17, size 1.713e-15, alpha -6.031, combined = 1.539e-12 rroots = 5
[/code]
I think it needs more fans to pump the head from the GPU out. But that will take some time to set up,

Chris

[QUOTE=chris2be8;365932]I didn't get far. After about 27 minutes the system crashed again. This time the PSU survived so I could power the system back up. Running npr on the best 100 entries in msieve.dat.ms found: [code]
R0: -15595511478358973801530755477526
R1: 6316360635713693
A0: 2096409315900123991624591017294407684793
A1: -21208160729383271288378383776159
A2: -79287985949944664147956451
A3: 2064769571596254311
A4: 249346723154
A5: 144
skew 19821182.17, size 1.713e-15, alpha -6.031, combined = 1.539e-12 rroots = 5
[/code]
I think it needs more fans to pump the hea[t] from the GPU out. But that will take some time to set up,

Chris[/QUOTE]That's a pretty nice score. I'll run some searching locally and see if I can find anything better. If not, I might go ahead and tackle this one next (5 days or so to go on my current job).

My upcoming jobs above c150 are c154, [c159], and c164. (And I'm not real sure I'm up to tackling a c164 again....

I've concluded that EVO Labs PSUs don't always provide their full rated power. Another try killed the 700W model I'd replaced the original 500W PSU with. The GPU uses 170W max, but mostly 12v power which the PSU may not be able to provide enough of.

I've replaced it with a Corsair CX750 PSU. That's rated to provide up to 62 amps of 12v power, which should be plenty. But it cost nearly as much as the GPU (that was second hand though).

I'm running another soak test against your C160. If the new PSU survives that it should survive anything. I'll post the best poly it finds.

This has definitely reached "frustrating". PSUs are supposed to either provide their rated power or have overload protection to stop them being damaged.

Chris

[QUOTE=chris2be8;366027]I've concluded that EVO Labs PSUs don't always provide their full rated power. Another try killed the 700W model I'd replaced the original 500W PSU with. The GPU uses 170W max, but mostly 12v power which the PSU may not be able to provide enough of.

I've replaced it with a Corsair CX750 PSU. That's rated to provide up to 62 amps of 12v power, which should be plenty. But it cost nearly as much as the GPU (that was second hand though).

I'm running another soak test against your C160. If the new PSU survives that it should survive anything. I'll post the best poly it finds.

This has definitely reached "frustrating". PSUs are supposed to either provide their rated power or have overload protection to stop them being damaged.

Chris[/QUOTE]Wow, something to keep in mind for the next system I build. I had a PS go in my main system; luckily, when I ordered a replacement, they had one with a higher rating on sale cheaper than the rating I replaced.

some PSU are certified. no-name PSU and low quality can deliver between 70- to 75% of their nominal power. Certified PSU have better rate. Also a small link.

[url]http://www.tomshardware.co.uk/forum/267343-28-explain-bronze-silver-certified-silvers-best[/url]

Run finished. Best poly was: [code]
R0: -6364655919201796079867797099408
R1: 295615254460640777
A0: -179435110610878523647136682144480120081
A1: -75945269806551158820034767590206
A2: 38100445529483857085112451
A3: 7411823563051778744
A4: -537302487308
A5: 12720
skew 7801224.82, size 2.517e-15, alpha -7.009, combined = 1.985e-12 rroots = 3
[/code]
That was searching leading coefficients from 1 to 20000 (in two stages) and running npr on the best 1000 stage 1 hits.

I'm now happy that the new PSU can cope. "All" I have to do is learn how to tune polynomial selection for best results on various size inputs.

Chris

[QUOTE=chris2be8;366105]Run finished. Best poly was: [code]
R0: -6364655919201796079867797099408
R1: 295615254460640777
A0: -179435110610878523647136682144480120081
A1: -75945269806551158820034767590206
A2: 38100445529483857085112451
A3: 7411823563051778744
A4: -537302487308
A5: 12720
skew 7801224.82, size 2.517e-15, alpha -7.009, combined = 1.985e-12 rroots = 3
[/code]
That was searching leading coefficients from 1 to 20000 (in two stages) and running npr on the best 1000 stage 1 hits.

I'm now happy that the new PSU can cope. "All" I have to do is learn how to tune polynomial selection for best results on various size inputs.

Chris[/QUOTE]Thanks. I could not find a higher one after a couple of days of CPU search, so I'm going to start working with this one in the next week or so.

AS4788:i5212 - C158
 

Looks like Aliquot Sequence 4788 is ready for GNFS with a C158.

The last term is [URL="http://www.factordb.com/sequences.php?se=1&aq=4788&action=range&fr=5212&to=5212"]here[/URL].

And the number is:
[CODE]76869571906355491906959871188888092471250708894766586952516503475089113869342318053568967454050390052064136129541897584223796694069753683292240241556441311641[/CODE]A nice poly is requested.

I'll give the C152 from 4788 a 2-GPU-day shot. Coeff 3.6M and 4.7M start points.
This size number deserves about a GPU-week, so someone else should have a run also.
-Curtis

I'll run from 100k.

Top result so far:

[CODE]expecting poly E from 1.69e-012 to > 1.94e-012

R0: -2689280107307295526801513157343
R1: 2931167726039959
A0: 99040350235836961822906826562013900000
A1: 95781172805689829891986809259050
A2: 27215466812960912400211495
A3: -10524994985190571386
A4: -7840244511848
A5: 546480
skew 3595166.53, size 2.227e-015, alpha -7.379, combined = 1.813e-012 rroots = 3[/CODE]

Secondary option, but not as good:

[CODE]R0: -2155578372482669365926084128205
R1: 1127260259541943
A0: -40328988024591975714087292182419509328
A1: -66820151154132769497071567631796
A2: 6728377480314191201183140
A3: 25120365631094055045
A4: -7734154948932
A5: 1651716
skew 2071236.75, size 2.021e-015, alpha -6.944, combined = 1.693e-012 rroots = 3[/CODE]

[CODE]
n: 76869571906355491906959871188888092471250708894766586952516503475089113869342318053568967454050390052064136129541897584223796694069753683292240241556441311641

# norm 2.556114e-015 alpha -8.042625 e 1.746e-012 rroots 3
skew: 8507980.48
c0: 697527486164130532279865895120392673275
c1: -120763715291162991817394881816805
c2: 160759075606110254468802459
c3: -54065299020536806483
c4: -5257027044422
c5: 129960
Y0: -3584173698956592806915920474584
Y1: 98597680939668659

# norm 2.416781e-015 alpha -6.370630 e 1.697e-012 rroots 5
skew: 3404723.60
c0: -34793535780275951024922226403477538000
c1: 18037274809042927554180583138536
c2: 52891988988073712409582830
c3: 2731440434913531781
c4: -4988767267408
c5: 159600
Y0: -3439889529759048116671987007159
Y1: 228778027137143873
[/CODE]

Two days on the C152 got me nothing better than 1.58e-12. I yield to wombatman's poly.

HP[SUB]6[/SUB](96) has a c168. I ran a full t55 on it.

If any of you feel like looking for a poly you should go ahead.

But I have a few numbers waiting for GNFS, so it might be a while before I get to it.

Here's a quick initial one:

[CODE]polynomial selection complete
R0: -351185084381712098737001356638663
R1: 6845968206438053
A0: 1576437232461335934077481224737190849790080
A1: -59688835068123814058351767034530824
A2: -11054644286537809640172862132
A3: -16485688320011887029
A4: 20579079041540
A5: 140556
skew 30830220.54, size 1.753e-016, alpha -7.577, combined = 3.959e-013 rroots =
3
elapsed time 00:17:13[/CODE]

And here's a much better one:

[CODE]R0: -148138110612077723916876612082378
R1: 8339809050965341
A0: -12770912230199211021871868495951883270795
A1: 13079726119391213221893906059256122
A2: 1496544624011384736520582339
A3: -1484564397432902982318
A4: -82524300618068
A5: 10524360
skew 5555477.07, size 2.099e-016, alpha -7.736, combined = 4.460e-013 rroots = 5[/CODE]

Aliquot Sequence 3366 is now ready for GNFS with a c154.

The last term is [url="http://factordb.com/sequences.php?se=1&aq=3366&action=range&fr=2120&to=2120"]here[/url].

The composite is:
[code]3308009973407192231807077520689238593649607835059580923229693691366589979099091729341044080076948136375169476605174148416558932726440741541408142888471481[/code]

A nice poly is requested.

Starting at 100,000 and running overnight.

A little low, but a good start.

[CODE]expecting poly E from 3.10e-012 to > 3.57e-012

polynomial selection complete
R0: -376204185518317275648739469673
R1: 1226119532674199
A0: -24163564970005681572050262530157268480
A1: 41684539050837063806347703372056
A2: 13347999928080102193859026
A3: -12359935788342500071
A4: 1096104627318
A5: 438984
skew 3049806.24, size 5.349e-015, alpha -7.266, combined = 3.033e-012 rroots = 3[/CODE]

Small improvement:

[CODE]polynomial selection complete
R0: -313426207408478690741807350442
R1: 4985154433895227
A0: 64274525925439519677314518508387370465
A1: 50955331804223207969796444100169
A2: -16775534299649772885367939
A3: -14160958980613019665
A4: 335875177338
A5: 1093680
skew 2722245.55, size 5.610e-015, alpha -7.448, combined = 3.081e-012 rroots = 3[/CODE]

C166 @ AS3408:i1399
 

Aliquot sequence 3408 is ready for GNFS and a poly.

The current term is [URL="http://factordb.com/sequences.php?se=1&aq=3408&action=range&fr=1399&to=1399"]here[/URL].

The C166 composite is:

[CODE]1123367150134848120647467256535952482658686476885324744360520386356344637194859571636314844955603685901980138488599119535795850301781912277982445337403841501308675373[/CODE]

A nice poly is requested.

Since no one else seems interested I'll have a go. I'll let you know when I've got something.

Chris

C166 @ AS3408:i1399 polys
 

[CODE]
N: 1123367150134848120647467256535952482658686476885324744360520386356344637194859571636314844955603685901980138488599119535795850301781912277982445337403841501308675373

# norm 4.740739e-016 alpha -6.805832 e 6.493e-013 rroots 3
skew: 4440977.86
c0: -177048535144577104002100358070738707625
c1: 365723560219633695438695599995825
c2: 495640540820121872411293295
c3: -111509067703870473337
c4: 7960709240186
c5: 1321320
Y0: -61080521272901563946054800533482
Y1: 875338835798211529

# norm 4.462039e-016 alpha -7.435898 e 6.237e-013 rroots 3
skew: 9112731.34
c0: -23808740138726624289321423454604296645443
c1: -1466967236654254780631224580887119
c2: 967665755570684778379380281
c3: -130649495154318981865
c4: -382514849014
c5: 1321320
Y0: -61080522378333716799855808469480
Y1: 875338835798211529

# norm 4.491349e-016 alpha -7.834517 e 6.233e-013 rroots 5
skew: 10120852.14
c0: -2487662169925619011414083785466603069415
c1: 17900717407620781734682035008680428
c2: 1738977030042289153018461673
c3: -780255315421125543468
c4: -15384833651358
c5: 1312740
Y0: -61160160247186338969070336749222
Y1: 864634448508699529

# norm 4.496146e-016 alpha -7.259351 e 6.231e-013 rroots 5
skew: 8596150.68
c0: 15831684823536804655058985745100802791721
c1: 10704450769273425265783975355355078
c2: 192151769714510808550440602
c3: -425569331212241159887
c4: 5995995318156
c5: 1447380
Y0: -59977426957982371068013390847954
Y1: 5137117526847881
[/CODE]

My best result is: [code]
n: 1123367150134848120647467256535952482658686476885324744360520386356344637194859571636314844955603685901980138488599119535795850301781912277982445337403841501308675373

# norm 5.501656e-16 alpha -7.179471 e 7.119e-13 rroots 5
skew: 36195688.86
c0: 75577562337689364028481050776760905394833
c1: 3224013476052476246459675851031357
c2: -1644146666659802871764747813
c3: -9297976523287591189
c4: 1376786174160
c5: 1692
Y0: -231432459847718112775382371692070
Y1: 2775839702823571
[/code]
Chris

My best one is below sashamkrt's polys.

And my second best result is: [code]
# norm 5.386430e-16 alpha -7.124421 e 7.002e-13 rroots 5 skew: 39711911.18
c0: 313768005569780256947621801516554049147768
c1: 5242099641663080151981840862637332
c2: -1623782187706610058329399198
c3: -12691336302789308989
c4: 1371563435460
c5: 1692
Y0: -231432459849431763536721989131065
Y1: 2775839702823571
[/code]
I should have posted that earlier, but sleep got in the way. Sorry.

Chris

If RichD does some test-sieving, I am curious to hear what yield is for Chris' polys vs sashamkrt's. I still don't have a firm grasp of the effect of skew on the sieving phase, but Chris' skews are very high due to his tiny A1 choice.

Or, for the NFS experts: what effect will sieving with a poly skew of 30-40M for this C166 have? Is that effect outweighed by the Murphy score running 10% higher than Sasahmkrt's polys, the best of which has skew 4M?

Is it valid that if the yield of the high-skew poly is good enough to produce enough relations before poly performance drops off, we don't care what skew is?

The Murphy score assumes a sieving region of fixed area, and factor bases of fixed (small) size. I don't think there's a set answer to your question, the sieving rate could be higher or lower than the Murphy score would indicate. Lattice sieving doesn't care very much that the sieving region is very wide and thin, and we don't have many samples of (good) polynomials with very large skew to compare with.

I've got a c154 from aliquot sequence 611156:i7547 that needs a poly:[code]1233834109316954251065406210584382514482486785123839242711421920174697279622119210044011032460138316615713715099127735270405764948375711542820719611655437[/code]Something nice would be appreciated.

611156:i7547 c154 poly
 

[CODE]
611156:i7547

n: 1233834109316954251065406210584382514482486785123839242711421920174697279622119210044011032460138316615713715099127735270405764948375711542820719611655437
# norm 8.191801e-015 alpha -7.444448 e 3.477e-012 rroots 5
skew: 1837012.21
c0: 9524788629911251806899820614452358175
c1: 19825309048338991034711681062715
c2: -12548193435218537869063512
c3: -16897387341480048942
c4: 5214672777134
c5: 1375980
Y0: -245770001251625498938715399762
Y1: 43639219601327273

[/CODE]

[QUOTE=sashamkrt;371951][CODE]
611156:i7547

n: 1233834109316954251065406210584382514482486785123839242711421920174697279622119210044011032460138316615713715099127735270405764948375711542820719611655437
# norm 8.191801e-015 alpha -7.444448 e 3.477e-012 rroots 5
skew: 1837012.21
c0: 9524788629911251806899820614452358175
c1: 19825309048338991034711681062715
c2: -12548193435218537869063512
c3: -16897387341480048942
c4: 5214672777134
c5: 1375980
Y0: -245770001251625498938715399762
Y1: 43639219601327273

[/CODE][/QUOTE]Thank you for that! A local CPU-only search got this score after 100+ hours:[code]# norm 6.673765e-015 alpha -7.751098 e 3.299e-012 rroots 5
skew: 34808196.94[/code]

Greetings,

I've got a c161 from 96^127-1. (ecm pre-tested to t55)

[CODE]29402026346876404520391138812753203396665114819180002640111042211241267771985958243562132692098504888206091141479100367473237624904438034507578300300706369793341[/CODE]


Any help would be much appreciated. :smile:

I'll have a go. It'll probably take about 2 days to produce a poly.

Chris

Thanks Chris,

I've started 6 CPU cores on the old pol51 binaries for something to compare with. They should be done in a couple of days.

My best score was 1.388e-12, It appeared 3 times in msieve.dat.p, all for the same poly.

My best 3 ignoring duplicates are:
[code]
# norm 1.667441e-15 alpha -7.002235 e 1.388e-12 rroots 5
skew: 67950163.58
c0: -92018656678781734318812584786978993173168
c1: 17606278543589316445642626684590972
c2: -184245044024739887319988260
c3: -9925624700183687723
c4: 23633660818
c5: 660
Y0: -33866214470571755539145261052495
Y1: 119692786004264443

# norm 1.664533e-15 alpha -7.463580 e 1.375e-12 rroots 5
skew: 13851961.80
c0: -3980366090430387084049780075594783704630
c1: 1150963403949097102401214675385591
c2: 240601330168742354186263763
c3: -35973749870761629187
c4: -1525113916932
c5: 21672
Y0: -16845956005729258547405590420453
Y1: 22299707395951193

# norm 1.620840e-15 alpha -6.981357 e 1.359e-12 rroots 5
skew: 76448553.94
c0: 98489208632605476746240133473937531314816
c1: 26848738563981520237901585677130936
c2: 16059473682133602087809922
c3: -10261934626721014715
c4: 1907497018
c5: 660
Y0: -33866215258591475056417014729393
Y1: 119692786004264443
[/code]
I'll be interested if pol51 gets anywhere near as good a score. In my experience msieve generates better polys on a CPU. And my GPU should have searched a much larger range than a CPU could in that time.

Chris

pol51 holds its own
 

Top 3 Candidates:

[CODE]skew: 1195653.98
# norm 5.84e+22
c5: 22302000
c4: -464777566858860
c3: -99436795485673648928
c2: 578297457968152336612278175
c1: 48812371410286624110411117747644
c0: -91385745414121004257569049157219000311
# alpha -7.75
Y1: 1082295975254504197
Y0: -4207335607185413648657384925588
# Murphy_E 1.18e-12
# M 20751775683164952457242854624160015284564131656598338624140314606734216907514433651924011151071370571710908193183434894211011777674434219738391337425965679674713

skew: 603563.67
# norm 2.79e+21
c5: 17088660
c4: 55736335067836
c3: -24337375383614088601
c2: -22783222883034211139137917
c1: 3762710284495542565468049664129
c0: -83444455839241142352217880538651963
# alpha -4.90
Y1: 584188293422681497
Y0: -4437453482190738444802154305568
# Murphy_E 1.12e-12
# M 14893812980472872748426252973158126516305595988165839050371884840345042253354616381401043147671151873314700192158970186409363237060108595396222768522630896644523


skew: 961948.16
# norm 1.75e+22
c5: 53091120
c4: -60160923641566
c3: -99495731552476470845
c2: 54975499562688884099764359
c1: -32588660991353723457107645266731
c0: 2287737267985726330614672965275562463
# alpha -6.28
Y1: 4036339538754394157
Y0: -3537294243672844580035463385550
# Murphy_E 1.10e-12
# M 25747597891659421512785722580208176486812627500621563442893503693548474750169867971141097140926018640467241525952556349336043133438320153062186410883050302658875
[/CODE]


I must say that I was expecting a larger gap between the pol51 top poly and the msieve GPU top poly. Something like an order of magnitude.

For a single sample area only, SpecialQ range of 100000, I'm getting a 14.2% increase in yield.

Cheers

I should probably make the script search for larger HLQs. It's not really tuned for C161s (it just searches HLQs from 1 to 100000). But I don't know how to choose a range for a given size of number.

Chris

I doubt anyone would be able to get a 10x difference in sieving performance by choosing a better polynomial; for RSA512, the largest difference in sieving performance between totally unoptimized polynomials and the output of Kleinjung's algorithm, that anyone has been able to find, is something like a factor of four.

For a C161 I'd start looking at coefficients above 1000000.

Thanks Jason for straightening out my muddled thinking there.:davieddy:

(FWIW, I probably heard the order of magnitude ref in regards to how much time the GPU version takes for an equivalent MurphyE which could be totally correct. Of course, boosting the MurphyE score takes a non-proportional increase in effort.)

Requesting GNFS polys for both of these xyyxf composites. SNFS yields for both are just plain terrible.

[code]
C168_130_71 = 293577856524534308556608110931494014404182621098756377812259533965962071178386204940945650625875365752664844816196696488552291293374296950182835664858833152967071700503

C168_134_94 = 451591044633621500700127843125932943919387601290262860485200418433795934760393784972054631775554879954085888690144804817796633480540639229986141076270921955279078568333
[/code]

Thanks for any help!

[QUOTE=sashamkrt;371951][CODE]
611156:i7547

n: 1233834109316954251065406210584382514482486785123839242711421920174697279622119210044011032460138316615713715099127735270405764948375711542820719611655437
# norm 8.191801e-015 alpha -7.444448 e 3.477e-012 rroots 5
skew: 1837012.21
c0: 9524788629911251806899820614452358175
c1: 19825309048338991034711681062715
c2: -12548193435218537869063512
c3: -16897387341480048942
c4: 5214672777134
c5: 1375980
Y0: -245770001251625498938715399762
Y1: 43639219601327273

[/CODE][/QUOTE]18 days later:[code]prp62 factor: 15397886342298096881993126507804932212098047954161942939890657
prp92 factor: 80130095903332114576197636633232711661580076072271741043519900823811963455142901912702330541[/code]Thanks again!

C168_130_71 polys
 

[QUOTE=swellman;373248]
[code]
C168_130_71 = 293577856524534308556608110931494014404182621098756377812259533965962071178386204940945650625875365752664844816196696488552291293374296950182835664858833152967071700503
[/code][/QUOTE]

[code]
# norm 2.543780e-016 alpha -7.960966 e 4.428e-013 rroots 5
skew: 28988859.01
c0: 1062215008480970792901257152198983339519285
c1: 183447334833241615015815837774853009
c2: -11275670137112669226852509629
c3: -972496338763237373105
c4: 16927684068408
c5: 196560
Y0: -272173367430666031300850149081496
Y1: 137410100549285747

# norm 2.582543e-016 alpha -7.601118 e 4.411e-013 rroots 5
skew: 17409001.95
c0: -7955852503352268262565791828072786886325
c1: 29671650887957557467125296674725865
c2: -5547713341035381175926696651
c3: -1090949680455146448497
c4: 15110874091608
c5: 196560
Y0: -272173367684683167636863076700178
Y1: 137410100549285747

[/code]

Thank you - much appreciated!

C168_134_94 polys
 

[QUOTE=swellman;373248]
C168_134_94
[/QUOTE]

The best I've found:
[code]
C168_134_94 = 451591044633621500700127843125932943919387601290262860485200418433795934760393784972054631775554879954085888690144804817796633480540639229986141076270921955279078568333

# norm 2.823478e-016 alpha -6.482150 e 4.730e-013 rroots 5
skew: 3361097.76
c0: -307614352606874801585694333843444177201
c1: 1401169162224492060422172914867970
c2: 894215456445484381015296096
c3: -135179339454555085966
c4: -73678892093863
c5: 10197684
Y0: -134663146019366686167859460191928
Y1: 458872479559461347

# norm 2.559875e-016 alpha -6.802294 e 4.533e-013 rroots 3
skew: 3201711.31
c0: 1263970061965078222765677851115619674831
c1: 202421940857837946691009854043805
c2: -883109385510979954289052600
c3: -154827687975231270885
c4: 22102810195129
c5: 10778040
Y0: -133180642537639277642324211796970
Y1: 109998887046030013

# norm 2.459939e-016 alpha -6.667674 e 4.313e-013 rroots 3
skew: 1975697.11
c0: -324920921271212032847546328915881597625
c1: 48151046352838040184498268911605
c2: 1051996428693474022859339243
c3: 365477064513368583179
c4: -322817089512834
c5: 11671920
Y0: -131075223761471907716419321629736
Y1: 219812386093713251

[/code]

Again, my thanks.

Anyone like to have a go at finding a poly for this C150 from Aliquot sequence 829332:3534?

[QUOTE]317861755950589524663668143654241775869683357813561272912580550015875979854007874694938191790792005111631751001246771609088835690096693017036893082583[/QUOTE]

Thanks very much in advance!

[offtopic]
please use code section instead of quote section.
it makes the post very difficult to read on my two-dollars terminals...
[/offtopic]

Whoops, sorry about that LaurV!

I'll run overnight starting at 1,000,000.

Here's the best one from a range that spanned from 1e6 to about 6.3e6:

[CODE]polynomial selection complete
R0: -42871029888439449149155215493
R1: 110147580964787
A0: 211421791366353129151562921662744800
A1: -1558626214321169860712785635120
A2: -811722799717554488528036
A3: -3314064974445742552
A4: -1802001073407
A5: 2194920
skew 997647.75, size 1.297e-014, alpha -7.112, combined = 5.049e-012 rroots = 3
elapsed time 00:07:29[/CODE]

Thanks a lot! My 3-1/2 day CPU search came up with a score of 6.348000e-012. I'll test sieve both and see which is better.

As noted [URL="http://www.mersenneforum.org/showthread.php?p=379863#post379863"]here[/URL], I have a [URL="http://factordb.com/sequences.php?se=1&aq=7044&action=last20&fr=0&to=100"]c170[/URL] that is ready for GNFS. I will be running this starting in the fall since my local daytime temps limit me to ~1/2 my cores. A nice poly would be greatly appreciated for this number:[code]44372976609219605920970917073060893237104077933564785933283944210773143942177660972637693825980692133877619049897368796520149382235143537600699965770310835229937169064489[/code]Thanks to [URL="http://www.rechenkraft.net/yoyo//y_status_ecm.php"]yoyo[/URL] and [URL="http://www.mersenneforum.org/showthread.php?p=379563#post379563"]Batalov[/URL] for running ECM for me that is out of reach with my horsepower.

Better metrics than Murphy_E ?
 

Hello all.

I've just done a moderately big GPU polynomial select for a C165, and, having trial-sieved rather more polynomials than I usually do, reached the slightly unexpected position where the 17th-best polynomial by Murphy_E was overwhelmingly the best by actual sieving yield.

I get the impression that Murphy_E is systematically larger for small leading coefficients; has anyone done empirical work on that scaling? Would it be worth trial-sieving a thousand polynomials on my next big GNFS job (this would be an effort about the size of the GNFS job) and seeing if I can quantify the effect?

How did your list of polynomials sort by the 'size' score? This is an integral by Bernstein that is independent of translation and skew, a higher score is better.

C165_134_120 poly request
 

Can someone have a go with a GNFS poly for this number?

[code]
459438041044149873293447911775750189334188171266352133449167595646830565714968065622756071729107336241581305604971832140291528446199347566513765523766901738465034561
[/code]

Thanks in advance.

Are the new CADO poly select tools available? Does anyone have them running yet? The paper they put out suggests their new tools find E-scores about 5% better than msieve.

I don't have a linux environment with a GPU, and I believe CADO is linux-only; am I mistaken?

I'll run a little bit overnight and see if anything good comes up.

[CODE]expecting poly E from 6.54e-013 to > 7.52e-013
polynomial selection complete
R0: -72282121430503005558365347862554
R1: 30143549754266929
A0: 10866930325383798521006629494754432253955
A1: 4586200568910877479742643013046673
A2: 1246880392864919530859391791
A3: -43030078174711951269
A4: -15521930381334
A5: 232848
skew 12300099.06, size 3.978e-016, alpha -7.237, combined = 6.508e-013 rroots = 3[/CODE]

Best from an overnight run.

Although the skew is a bit high, this poly sieves 5%-10% better.

[CODE]# norm 5.312242e-16 alpha -7.130420 e 6.877599e-13 rroots 5
skew: 43719798.98
c0: 11733960296493447965970979138987511920000
c1: 44340805857506327391452974614472304
c2: -4104856773330359330123402924
c3: 7221946156103196292
c4: 1875581206735
c5: 10800
Y0: -133585893493588108010583707471589
Y1: 350799735559930687
[/CODE]

Many thanks to both of you.

Can someone have a go with a GNFS poly for this number?

[code]
46862651776313668832684618638310007043245135907468247470585960688008180534742005269578548831878148535158738789789506710579367525183636389872513135592162572724499935530721
[/code]

Thanks in advance.

[QUOTE=piguy227;388648]Can someone have a go with a GNFS poly for this number?

[code]
46862651776313668832684618638310007043245135907468247470585960688008180534742005269578548831878148535158738789789506710579367525183636389872513135592162572724499935530721
[/code]

Thanks in advance.[/QUOTE]Why?

That is, why should we find a polynomial for you for an apparently uninteresting number. Make it interesting for us and we may do something.

Could I get someone with reasonably heavy resources to have a go at the C184

(125!+1)/(359*1003874788568233) ?

I need to do a search for a C129 first, but I'll take a whack at this for a few days or so. :smile:

My CUDA setup is dead (laptop power plug failed). I bought Xyzzy's 750ti, but it's in an ubuntu box, and he mentioned when I bought it that linux drivers weren't mature yet so I haven't tried installing CUDA on it.

If someone has info that says otherwise, I'll be excited to resume poly searching for myself and this thread after I get CUDA set up on the new card. A 750ti will be quite an improvement over a 460M, too!

[QUOTE=fivemack;379917]Hello all.

I've just done a moderately big GPU polynomial select for a C165, and, having trial-sieved rather more polynomials than I usually do, reached the slightly unexpected position where the 17th-best polynomial by Murphy_E was overwhelmingly the best by actual sieving yield.

I get the impression that Murphy_E is systematically larger for small leading coefficients; has anyone done empirical work on that scaling? Would it be worth trial-sieving a thousand polynomials on my next big GNFS job (this would be an effort about the size of the GNFS job) and seeing if I can quantify the effect?[/QUOTE]

Did any further conclusions ever come of this? This Murphy-E effect would explain why searchers in this thread routinely found "good" polys with tiny leading coeffs compared to the coeffs usually searched by the big guns/more seasoned factorers.

[QUOTE=VBCurtis;407923]My CUDA setup is dead (laptop power plug failed). I bought Xyzzy's 750ti, but it's in an ubuntu box, and he mentioned when I bought it that linux drivers weren't mature yet so I haven't tried installing CUDA on it.

If someone has info that says otherwise, I'll be excited to resume poly searching for myself and this thread after I get CUDA set up on the new card. A 750ti will be quite an improvement over a 460M, too![/QUOTE]

I have a GTX 750Ti working on Debian with the legacy nvidia drivers and some debian packages. CUDA and OpenCL are working fine. You should be able to install the same packages on Ubuntu but I don't know about the legacy Driver part.

These are my currently installed packages with *nvidia*, *cuda* or *opencl* for comparison. There may be some -dev packages that you don't need.
[CODE]dpkg-query -l '*nvidia*' | grep 'ii'
ii glx-alternative-nvidia 0.5.1 amd64 allows the selection of NVIDIA as GLX provider
ii libegl1-nvidia:amd64 340.76-2 amd64 NVIDIA binary EGL libraries
ii libgl1-nvidia-glx:amd64 340.76-2 amd64 NVIDIA binary OpenGL libraries
ii libgles1-nvidia:amd64 340.76-2 amd64 NVIDIA binary OpenGL|ES 1.x libraries
ii libgles2-nvidia:amd64 340.76-2 amd64 NVIDIA binary OpenGL|ES 2.x libraries
ii libnvidia-compiler:amd64 340.76-2 amd64 NVIDIA runtime compiler library
ii libnvidia-eglcore:amd64 340.76-2 amd64 NVIDIA binary EGL core libraries
ii libnvidia-ml1:amd64 340.76-2 amd64 NVIDIA Management Library (NVML) runtime library
ii nvidia-alternative 340.76-2 amd64 allows the selection of NVIDIA as GLX provider
ii nvidia-cuda-dev 6.0.37-5 amd64 NVIDIA CUDA development files
ii nvidia-cuda-doc 6.0.37-5 all NVIDIA CUDA and OpenCL documentation
ii nvidia-cuda-gdb 6.0.37-5 amd64 NVIDIA CUDA Debugger (GDB)
ii nvidia-cuda-toolkit 6.0.37-5 amd64 NVIDIA CUDA development toolkit
ii nvidia-detect 340.76-2 amd64 NVIDIA GPU detection utility
ii nvidia-driver 340.76-2 amd64 NVIDIA metapackage
ii nvidia-driver-bin 340.76-2 amd64 NVIDIA driver support binaries
ii nvidia-installer-cleanup 20141201+1 amd64 cleanup after driver installation with the nvidia-installer
ii nvidia-kernel-common 20141201+1 amd64 NVIDIA binary kernel module support files
ii nvidia-kernel-dkms 340.76-2 amd64 NVIDIA binary kernel module DKMS source
ii nvidia-modprobe 340.46-1 amd64 utility to load NVIDIA kernel modules and create device nodes
ii nvidia-opencl-common 340.76-2 amd64 NVIDIA OpenCL driver
ii nvidia-opencl-icd:amd64 340.76-2 amd64 NVIDIA OpenCL installable client driver (ICD)
ii nvidia-profiler 6.0.37-5 amd64 NVIDIA Profiler for CUDA and OpenCL
ii nvidia-settings 340.46-2 amd64 tool for configuring the NVIDIA graphics driver
ii nvidia-smi 340.76-2 amd64 NVIDIA System Management Interface
ii nvidia-support 20141201+1 amd64 NVIDIA binary graphics driver support files
ii nvidia-vdpau-driver:amd64 340.76-2 amd64 Video Decode and Presentation API for Unix - NVIDIA driver
ii nvidia-visual-profiler 6.0.37-5 amd64 NVIDIA Visual Profiler for CUDA and OpenCL
ii nvidia-xconfig 340.46-1 amd64 X configuration tool for non-free NVIDIA drivers
ii xserver-xorg-video-nvidia 340.76-2 amd64 NVIDIA binary Xorg driver
ii ocl-icd-libopencl1:amd64 2.2.4-1 amd64 Generic OpenCL ICD Loader
ii ocl-icd-opencl-dev:amd64 2.2.4-1 amd64 OpenCL development files
ii opencl-headers 2.0~svn28973-2 all OpenCL (Open Computing Language) header files
ii libcuda1:amd64 340.76-2 amd64 NVIDIA CUDA Driver Library
ii libcudart6.0:amd64 6.0.37-5 amd64 NVIDIA CUDA Runtime Library[/CODE]

Thanks, Christian! I've been meaning to move the card into my main system; your working setup encourages me to give this a go. Between CUDA-ecm and poly select, I miss having a GPGPU.

Just as a heads-up, I started at 100M and am currently passing through 102M or so.

Here's the best two I've found so far. Neither is really close to the expected e-value, but they may be worth trial-sieving:
[CODE]expecting poly E from 4.57e-014 to > 5.26e-014
R0: -137547131193043341793334930193270566
R1: 70224319209054721
A0: -48715040541374867514878224348681604586285
A1: -275708572214962467144881300531857095
A2: -397914309979816733185360998057
A3: -8084283143447552273377
A4: 607205817813118
A5: 101468208
skew 8049123.93, size 4.248e-018, alpha -6.525, combined = 4.117e-014 rroots = 1

R0: -137052228970709324982229261366831248
R1: 38784020496691787
A0: -44875760234631994289855676839471716527683225
A1: 26415950906429837100672340418747601895
A2: 1882671010943471032317074656115
A3: -187234878234320539666327
A4: -4219616160069306
A5: 103313520
skew 21539398.15, size 3.540e-018, alpha -7.329, combined = 3.696e-014 rroots = 5[/CODE]

I'm going to play around in the 500M range for a bit, and if nothing good shows up there, I'll drop down to a lower coefficient like 1M.

Here's the best in the 500-510M range:[CODE]polynomial selection complete
R0: -99597953807168680476889254921237040
R1: 155679631953907381
A0: -92612284320573352971113764845050011599172263
A1: -14544423142209256027253616209189318319
A2: 3861979188784232836840703659203
A3: 261460166847969497979769
A4: -19772402213056250
A5: 509724600
skew 12734708.51, size 3.757e-018, alpha -8.313, combined = 3.829e-014 rroots = 3[/CODE]

I'll try down at around 100M (and maybe 10M).

Here's a good one:[CODE]R0: -278041983541936879812937287512480720
R1: 83839300817041243
A0: -3390086640132698923491414956173048512425481
A1: 3738332722384494113415434782471399053
A2: -1160199637312567120662841930769
A3: -40839181952932892416147
A4: 1087473801891780
A5: 3006324
skew 32140665.92, size 4.576e-018, alpha -7.218, combined = 4.328e-014 rroots = 3[/CODE]

Still under the expected, but best score so far.

[CODE]polynomial selection complete
R0: -256712836882622622457839933429900183
R1: 61840577050423613
A0: 602093201366745363780865731715111641933619000
A1: 92059674952581867600427558954630776650
A2: -1195424019900556240035714699635
A3: -58210719595944209591027
A4: 330526880428891
A5: 4480740
skew 64846617.97, size 4.451e-018, alpha -7.480, combined = 4.305e-014 rroots = 5[/CODE]

Anyone like to have a go at finding a poly for this C154 from Aliquot sequence 829332:3589?

[CODE]4537787384062167229294057374468826821223279315636792016447006047185936215587115121338899234285565806951562459365575674996087301727915180317496191381025877[/CODE]

This would be most appreciated! Thanks,

Rich

[QUOTE=richs;411875]Anyone like to have a go at finding a poly for this C154 from Aliquot sequence 829332:3589?[/QUOTE]

I should have something later tonight.

I went to 5M and the best is:
[CODE]N: 4537787384062167229294057374468826821223279315636792016447006047185936215587115121338899234285565806951562459365575674996087301727915180317496191381025877
# expecting poly E from 3.05e-12 to > 3.50e-12
R0: -399979656975273365701998369366
R1: 1067710737833939
A0: 519125690666296999062877195771789597
A1: 2025984829776720254430274320937
A2: -1154519980571878704863718
A3: -3142397915330686792
A4: 883463054330
A5: 443256
# skew 1379910.48, size 6.239e-15, alpha -5.942, combined = 3.375e-12 rroots = 5[/CODE]

Thanks, Rich!

Rich

I have a C166 that will be my first personal 15e factorization, and my first foray into the CADO tools for CPU poly select. Could someone give it some GPU time, so I might compare the effectiveness of msieve vs CADO for poly select?
Please list the GPU-hours you spent, so I can compare effort as well as results to my own CADO run.
[CODE]
N: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
[/CODE]
The number is the remaining cofactor of 13*2^924-1.

Stage 1: 22,3h GPU (GeForce GTX 650 Ti) + ~3,2h for one CPU thread.
Sizeopt and rootsieve: ~4h for one CPU thread.

Best poly:
[CODE]
n: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
# norm 4.663644e-16 alpha -7.071486 e 6.464067e-13 rroots 5
skew: 29930803.76
c0: -179409669872419787875791142567559019315525
c1: 28507155776738138662132445463560139
c2: 142812890347046919926946189
c3: -97138933140798385355
c4: -790917891528
c5: 25200
Y0: -201367272691407156150843550143488
Y1: 805187391468667987
[/CODE]Note that I modified msieve: HIGH_COEFF_MULTIPLIER 420. The deadline per coefficient and the randomization are disabled.

[QUOTE=VBCurtis;417843]I have a C166 that will be my first personal 15e factorization, and my first foray into the CADO tools for CPU poly select. Could someone give it some GPU time, so I might compare the effectiveness of msieve vs CADO for poly select?
Please list the GPU-hours you spent, so I can compare effort as well as results to my own CADO run.
[CODE]
N: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
[/CODE]
The number is the remaining cofactor of 13*2^924-1.[/QUOTE]

[QUOTE=Gimarel;418008]Stage 1: 22,3h GPU (GeForce GTX 650 Ti) + ~3,2h for one CPU thread.
Sizeopt and rootsieve: ~4h for one CPU thread.

Best poly:
[CODE]
n: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
# norm 4.663644e-16 alpha -7.071486 e 6.464067e-13 rroots 5
skew: 29930803.76
c0: -179409669872419787875791142567559019315525
c1: 28507155776738138662132445463560139
c2: 142812890347046919926946189
c3: -97138933140798385355
c4: -790917891528
c5: 25200
Y0: -201367272691407156150843550143488
Y1: 805187391468667987
[/CODE]Note that I modified msieve: HIGH_COEFF_MULTIPLIER 420. The deadline per coefficient and the randomization are disabled.[/QUOTE]

For a number of this size, it's probably worth posting the top 3-5 polys by either CADO or msieve for a better comparison. (Unless by "best" you mean you trial sieved rather than relying upon the Murphy score.)

[QUOTE=Dubslow;418010]For a number of this size, it's probably worth posting the top 3-5 polys by either CADO or msieve for a better comparison. (Unless by "best" you mean you trial sieved rather than relying upon the Murphy score.)[/QUOTE]

I did trialsieve. The poly has the best score and sieved best.

[QUOTE=Gimarel;418011]I did trialsieve. The poly has the best score and sieved best.[/QUOTE]

Sweet :smile: Just checking.

More thorough trialsieving showed that the poly with the second best score sieves a little bit better.

[CODE]
n: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
# norm 4.504602e-16 alpha -7.446747 e 6.383632e-13 rroots 3
skew: 36418581.30
c0: -6206292264781510140235453638003213286408
c1: 53762734012950656320733897775170588
c2: -966037993270550888998755014
c3: -105304721455866115403
c4: -333326337528
c5: 25200
Y0: -201367269767225015489302863783315
Y1: 805187391468667987
[/CODE]And a poly with a lower score sieves almost as good as the highscore.

[CODE]
n: 8343435760616643489063187244256443424394562560772513554600760230830772046038334603336773710814168017089799741524226536908218074130479225552429408073640301885602291889
# norm 4.072475e-16 alpha -7.147561 e 5.982085e-13 rroots 3
skew: 36320379.21
c0: -10900050903348427457786768699738222025160
c1: 53136490784077126637321358609712604
c2: -1063694405957759563314685958
c3: -105692114573027200331
c4: -294449793528
c5: 25200
Y0: -201367269518789276975994168402387
Y1: 805187391468667987
[/CODE]I have 17 more polys with a score higher than 5.98e-13 that sieve worse.

I don't see a way to get a second-best poly out of CADO; it appears pretty dogmatic that the highest-scoring poly is the choice, period. So, I'm splitting my runs into ~12hr chunks in hopes of getting a few polys to compare; So far after 36 hr x 2 instances (3 core-days) I have just two polys competitive, with the other 4 runs producing nothing meaningful.

The E-score generated by CADO is not remotely comparable to msieve; I have multiple polys with score better than 2e-11 by CADO reckoning that test-sieve worse than an msieve 6e-13 poly I have from the current GNFS165 job.

I'm giving CADO 5 core-days total to search, and will post the best poly it produces for anyone interested to compare to Gimarel's (or another GPU searcher?) msieve poly.

Gimarel- thank you for the work, and for test-sieving, and for the detailed tracking of time-on-silicon.

The msieve poly with score 6.38e-13 sieves about 15% faster than my best CADO poly. I'll give CADO another ~40 core-hours, but 15% is pretty hard to overcome.

I was pleased with a 6.67e-13 poly for my current G165 project- these two (6.38 and 6.46) are nice finds for this G166!

CADO produced nothing better. I *just* got GPU-msieve configured on my linux box, so I'm running a day of my own poly select under msieve. That also means I return to contributing polynomials to this thread, rather than requesting them!