[QUOTE=VBCurtis;472991]Likewise. GNFS-195 is definitely in 15e land on NFS@home. Serge's offer happens to be pretty close to the likely matrix-solving time.

However, we haven't done a forum-factoring collaboration project in quite a while. That's more fun, spirit of cooperation and all, but perhaps it's better used for numbers just above the reasonable 15e-queue range; say GNFS 197-200, or SNFS-290s.

I have an 8-core xeon with a sort-of-static IP at work that could host a forum-cooperative-CADO factorization. Such an effort is fully automated; helpers just point the cado-slave executable at the server IP/port, and work is issued automagically. Workunit length can be set fairly short, and cores can come and go as they please. Before doing such a thing, I would want some willing forumites to help me speed-test CADO vs lasieve, to make sure CADO isn't more than 10-12% slower. CADO thinks a G195 should be done with 3 large primes on one side, 32/33LP, and 15e; pretty close to what we'd have NFS@home do.

The server has 32GB, so it could (probably) handle the matrix at the end, though it would take 6-8 months.[/QUOTE]

I'm generally interested in what CADO can do, though my earlier attempts at fiddling with its poly select a year or so ago left a poor taste in my mouth, mostly because it's not very configurable in the manner of NFS: it always assumes that its user is doing the full factorization front to back, and is incapable of doing any single phase or subphase of NFS without setting the parameters for *all* phases, so e.g. you need to enter sieving parameters just to run poly select. And you can't do fine grained control of the polyselect with the included scripts, it can only be done by manually using the underlying binaries. (Credit where due: the underlying binaries are all extremely modular, so the limitations are "just" with the primary python scripts included with it.)

Having gotten all that off my chest, I would still generally be interested in using it (not least of which because we are in fact doing the whole factorization, so it won't complain about settings). It is of course by far the most actively developed publicly available NFS tool, and sometimes I think this forum might fall behind the times :smile: lol. So yeah, count me in if we go this way.

I don't have any particular preference to the team sieve in lieu of using NFS@Home though. It would certainly be faster with NFS@Home.

Anyone still have interest in this effort? Any poly search work started?

Or shall we shelve it for a few years? It is a very difficult job!

Poly select has been running for 10 days. I'll leave it run until early Jan on my office machine. I've searched 4M to ~5.3M, so I expect to reach 9M by the time classes resume in Jan. I'll run 3-4M at home, so consider 3-9M reserved for poly select.

I've had no major flares, scores around 8.3e-15.

[QUOTE=VBCurtis;474021]Poly select has been running for 10 days. I'll leave it run until early Jan on my office machine. I've searched 4M to ~5.3M, so I expect to reach 9M by the time classes resume in Jan. I'll run 3-4M at home, so consider 3-9M reserved for poly select.

I've had no major flares, scores around 8.3e-15.[/QUOTE]

Ok. I’ll run 1 up to 3M starting on Monday.

[QUOTE=VBCurtis;474021]I've had no major flares, scores around 8.3e-15.[/QUOTE]

Finally, a nice flare (well, two):
[code]# norm 4.812643e-19 alpha -7.374157 e 9.500e-15 rroots 5
skew: 105719902.83
c0: 30048502188167042982552251951340835389231313491
c1: 1153129598686829538208765662628311582607
c2: -29205521491058283837637168531513
c3: -1021865566517253157135583
c4: 3735057032949966
c5: 3760680
Y0: -37263245542673733241779787089068172892
Y1: 5730494632266754927

# norm 4.821277e-19 alpha -8.470200 e 9.695e-15 rroots 5
skew: 301546020.38
c0: -1185037957455716126073603887248566830856410560725
c1: 38199003018831596437241964330438131702970
c2: 5818942545524263202913191274913
c3: -2175672369500903012286010
c4: -652818679637388
c5: 3491640
Y0: -37820569343015955014796390113724725788
Y1: 581612093036528939

# norm 4.916080e-19 alpha -8.417676 e 9.714e-15 rroots 5
skew: 286215181.97
c0: -1519279501926262498874158108025086779859398489645
c1: 37578066834236592169180427319532471905810
c2: -64180208504197424958058226363687
c3: -2199546001529609458600570
c4: -466665733841388
c5: 3491640
Y0: -37820569343009753413003002073684535368
Y1: 581612093036528939[/code]
Before these two, I had 10+ days of GPU search without a poly better than 8.4e-15; this -npr run produced dozens above that, with about a dozen over 9e-15. No others are above 9.2e-15; so far, these three deserve test-sieving.
The search continues for 1e-14!

Nice run! All three are in record territory for a C195, at least above what’s published in [url=http://www.mersenneforum.org/showpost.php?p=457286&postcount=59]the matrix of msieve records[/url].

I’m plodding along my 1-3e6 range. Will post any comparable flares as they come.

Look what popped up!

[code]
# norm 5.150792e-19 alpha -6.895293 e 1.005e-14 rroots 3
skew: 205953807.19
c0: 14586426520656790166603684695846179081093909195
c1: 320025907913206973904563524908763132680
c2: -11623937105903994879764826112549
c3: -33595581283342108832054
c4: -407200273594396
c5: 75480
Y0: -81426839057924295592219470403716846092
Y1: 875368165139429171


# norm 5.069658e-19 alpha -7.410003 e 1.002e-14 rroots 3
skew: 384943586.39
c0: 371117796634382522495324382646226924594015314224
c1: -2668627366646746165734198158961668690334
c2: -10592409142608145921441819531525
c3: -1223447635548018295958
c4: -414633192634996
c5: 75480
Y0: -81426839057941536028632294855860303891
Y1: 875368165139429171
[/code]

Found using another search strategy suggested by Wombatman last year, i.e. Stg_1_norm / 20, Stg_2_norm / 10. These results were found in the range of 1 to 80K. And that took 4 days to process! Like a firehose of polynomials...

Invocation line was as follows

[code]msieve -v -g 0 -np1 -nps "12,3000000 stage1_norm=1.78e28 stage2_norm=2.30e27" -s output[/code]

How high should we aim? 1.5e-14? Or a 2-handle?

I don't expect to find anything higher, but I do want to finish the search from 0-9M. Looks like we've got 5 polys to test-sieve before we ask NFS@home for 15e sieving.

I'm using 2.5e26 for stage 2 norm, and still getting more hits per day than I really want to root-opt. I've got stage 1 at 3-3.5e28.

My 750ti seems to have failed this week; I'm doing 6-9M on a Quadro 2000. I think that'll be at least another week.

C195 poly
 

Correction: at least 6 polys to test-sieve. CADO optimized swellman's poly:
[code]
Y0: -81426839057950426757923623376601315651
Y1: 875368165139429171
c0: 204683503018401689426140406687133627010286803680
c1: -2452099432849828691451653120112759691470
c2: -10812552584853359237467981971685
c3: 15699401891095722879082
c4: -418466278378996
c5: 75480
skew: 345149382.50858
# lognorm 60.66, E 53.05, alpha -7.62 (proj -1.82), 3 real roots
# MurphyE = 1.07002154e-14
[/code]
CADO also insists that VBCurtis's poly should look like this:
[code]
Y0: -37820569343019612054376949851072328184
Y1: 581612093036528939
c0: -1424454413957310647069963366688072494404174212685
c1: 37868451184504634441381013416680343540914
c2: 46695746836688002052020455253945
c3: -2157872836374911253589882
c4: -762591721102188
c5: 3491640
skew: 302152376.28392
# lognorm 62.74, E 54.26, alpha -8.48 (proj -2.34), 5 real roots
# MurphyE = 9.69471073e-15
[/code]

[QUOTE=VBCurtis;474704]I don't expect to find anything higher, but I do want to finish the search from 0-9M. Looks like we've got 5 polys to test-sieve before we ask NFS@home for 15e sieving.

I'm using 2.5e26 for stage 2 norm, and still getting more hits per day than I really want to root-opt. I've got stage 1 at 3-3.5e28.

My 750ti seems to have failed this week; I'm doing 6-9M on a Quadro 2000. I think that'll be at least another week.[/QUOTE]

Ok, I’m currently grinding up to 3M. Hoping msieve works a bit faster once I get out of the mud of small coefficients.

@Max0526 - thanks for the tweaks. Between us all, we should nail down a top notch poly.

Max, can you explain why CADO "optimized" my 9.71 poly into one with a lower score? Is there still a difference between msieve's murphy score and CADO's?