Register FAQ Search Today's Posts Mark Forums Read

2020-01-20, 02:32   #1871
VBCurtis

"Curtis"
Feb 2005
Riverside, CA

33·163 Posts

Quote:
 Originally Posted by jagotu Thanks for any hints and/or recommended reading on how to better contribute :)
We're happy to have you aboard to help us out! For a linux user lacking a CUDA gpu, CADO is really the only way to go. Msieve's GPU code is about 50x faster than CPU, but the CADO poly select (which only runs on CPU, and only linux as far as I know) is competitive with msieve-GPU work.

That said, for msieve-CPU work I'd set the coefficient range you want to search in the command line, and otherwise let it work. After a couple tries, you might notice that you'd like to restrict the output of one or more parts of the process.

So, I'd run: ./msieve -np "1000000,1100000" -s polyfile1
on the first core, changing the coeff's on each instance in order to cover, say, a 1M block in a reasonable time (like a day, or two). Each instance will look to worktodo.ini for the input number, so you can run every instance in the same folder if you like (as long as you make different save files with -s for each).

Once you do this, you'll see how may polys are generated by the default msieve poly select settings, and you can then consider running the steps separately. The 3 steps are -np1 (generate raw polynomials for future polishing), -nps (optimize raw polys for size, takes a fraction of a second per poly), -npr (optimize for root properties, produces final polys usable by a sieve, takes 3-5 min per poly at ~C190 size, more for bigger composites)

The usual GPU-enhanced way to run things separately is to run -np1 (the GPU part) and -nps (the size optimizing step, which quickly improves and scores each raw poly generated by the np1 step) together. I'll run a GPU for a day or so, and then sort the output file by score and send only the top 100 or so scores to the -npr step; the rest would be a waste of time to try root-optimizing on.

So, you might run: ./msieve -np1 -nps "1000000,1100000" -s rawpoly1

and after it finishes, cat all the output files from your various runs together, sort them by the last column (which is a measure of quality), truncate to the top 100 or 150 lines, and then:
./msieve -npr -s [cat'ed file name]
Note the cat'ed file must have suffix .ms, just as the rawpoly files have.

Whether this effort is worth your time depends on how many intermediate polys msieve generates with the default settings. If running -np1 -nps produces a file with 1000 lines in a day, you should do the -npr step separately and filter the -nps output.

Hope this helps! I'm not very good at recalling what I didn't know when I set out to learn this stuff, so I likely left out some bits important to understanding; I'll be happy to try to remedy those when you identify them.

 2020-01-20, 20:22 #1872 RichD     Sep 2008 Kansas 26·72 Posts C189 poly An early baseline. Code: N: 171065344393143145598685691004020457412949305674191856490470856856353467726774164074336525717740050483725614204311545373938206064857869944567319589822085897560103957504493890195049294967729 R0: -1398161583686828531470118604065371984 R1: 1187142349205771171 A0: 15102286091227267832135864754379187414025981435 A1: 497879116946641926689720423820807937837 A2: -27439975218771539667117465183681 A3: -575684274285767307977493 A4: 12193816703534774 A5: 32016600 skew 56550159.73, size 1.367e-18, alpha -8.083, combined = 2.035e-14 rroots = 5
 2020-01-21, 01:15 #1873 VBCurtis     "Curtis" Feb 2005 Riverside, CA 33·163 Posts CADO says: Code: n: 171065344393143145598685691004020457412949305674191856490470856856353467726774164074336525717740050483725614204311545373938206064857869944567319589822085897560103957504493890195049294967729 skew: 29327175.183 c0: 122813740464986032965496322532662517388492480 c1: -46371620654869241552317485387440475878 c2: -1751856410174864380538219925817 c3: 63446716152916521165305 c4: -2079998547748350 c5: 1134000 Y0: -3902553125077415603716331429749427663 Y1: 214995590231742585983729 # MurphyE (Bf=4.295e+09,Bg=4.295e+09,area=2.147e+15) = 2.489e-08 cownoise's opinion: Skew 49310323.62764 score 2.40292176e-14 That's over 10% better than the previous record, so I won't try a second CADO run. I ran admax 1M, P=3M, incr=210, nq=15625. If I did a second run, I'd try admin 1M admax 20M, incr 4620.
2020-01-21, 02:31   #1874
swellman

Jun 2012

23·359 Posts

Quote:
 Originally Posted by VBCurtis If I did a second run, I'd try admin 1M admax 20M, incr 4620.
I’m running CADO admin = 1M and admax = 5M, with incr = 4620 and nq = 15625.

2020-01-21, 04:22   #1875
RichD

Sep 2008
Kansas

1100010000002 Posts

Quote:
 Originally Posted by VBCurtis That's over 10% better than the previous record, so I won't try a second CADO run.
Looks like we have a viable poly. This is my best so far. Shutting down my search.
Code:
R0: -1393599239599595290006064741998926876
R1: 946625746109546621
A0: 136027168288871325742499604367778620119838525
A1: 73272651910647714482943504075726120902
A2: 8610143338350156781304768474363
A3: -281080067656185322633130
A4: -4984025700534996
A5: 32544120
skew 34904462.42, size 1.448e-18, alpha -7.208, combined = 2.109e-14 rroots = 5

 2020-01-21, 21:19 #1876 swellman     Jun 2012 B3816 Posts C189 poly My search should finish tomorrow evening. I’m bailing after that unless someone can make a good case for the search continuing.
 2020-01-23, 03:06 #1877 swellman     Jun 2012 23·359 Posts CADO result: Code: n: 171065344393143145598685691004020457412949305674191856490470856856353467726774164074336525717740050483725614204311545373938206064857869944567319589822085897560103957504493890195049294967729 skew: 29985050.394 c0: 2513826780601637928378359984878765061986468900 c1: -117746085170220907393886509080764090404 c2: -17584276331630705778529665859825 c3: 645100683623259210556021 c4: 13652526139832520 c5: -70686000 Y0: -2356084567431967037174939101706834305 Y1: 126078896451210514076387 # MurphyE (Bf=8.590e+09,Bg=4.295e+09,area=5.469e+16) = 1.184e-08 # f(x) = -70686000*x^5+13652526139832520*x^4+645100683623259210556021*x^3-17584276331630705778529665859825*x^2-117746085170220907393886509080764090404*x+2513826780601637928378359984878765061986468900 # g(x) = 126078896451210514076387*x-2356084567431967037174939101706834305 Cow noise says skew of 37158140.01991 gives an optimal e-score of 2.18166e-14. VBCurtis has the top poly, and it seems time for GNFS. I’m done with this poly search. Unless someone else wants to search further, I’ll run some test sieving on 15e this weekend.
 2020-01-23, 10:15 #1878 jagotu   Oct 2019 3·7 Posts I've run about two days of poly select, but my best poly had combined e ~1.8e-14. Still, thanks VBCurtis for your help with getting me started, will try to tune the args so I can hopefully find some nice polys in the future. Not searching anymore though, so we can probably go with VBCurtis' poly. Last fiddled with by jagotu on 2020-01-23 at 10:16
2020-01-25, 21:08   #1879
Max0526

"Max"
Jun 2016
Toronto

701 Posts
C189 poly -- record score correction

Quote:
 Originally Posted by swellman Code:  171065344393143145598685691004020457412949305674191856490470856856353467726774164074336525717740050483725614204311545373938206064857869944567319589822085897560103957504493890195049294967729 It has withstood ECM up to 90% of t60, making it ready for GNFS. A poly is needed. Any search help is appreciated. If using msieve-GPU, suggest searching in the range c5 > 25M. A sextic is unnecessary. As a point of reference, the record e-score for a C189 is 2.078e-14. All my machines are fully tasked now, but I will be able to contribute to this search in a few days.
Quote:
 Originally Posted by VBCurtis CADO says: Code: n: 171065344393143145598685691004020457412949305674191856490470856856353467726774164074336525717740050483725614204311545373938206064857869944567319589822085897560103957504493890195049294967729 skew: 29327175.183 c0: 122813740464986032965496322532662517388492480 c1: -46371620654869241552317485387440475878 c2: -1751856410174864380538219925817 c3: 63446716152916521165305 c4: -2079998547748350 c5: 1134000 Y0: -3902553125077415603716331429749427663 Y1: 214995590231742585983729 # MurphyE (Bf=4.295e+09,Bg=4.295e+09,area=2.147e+15) = 2.489e-08 cownoise's opinion: Skew 49310323.62764 score 2.40292176e-14 That's over 10% better than the previous record, so I won't try a second CADO run. I ran admax 1M, P=3M, incr=210, nq=15625. If I did a second run, I'd try admin 1M admax 20M, incr 4620.
Hi everybody!
Got lost in elliptic curves, hacked into some new Z6 families, see the bottom of the page: https://web.math.pmf.unizg.hr/~duje/tors/z6.html
I am truly sorry to interrupt this holiday: https://www.mersenneforum.org/showpo...postcount=1537
It was also reposted later in https://www.mersenneforum.org/showpo...&postcount=136, the second post after the chart.
To be helpful and productive, I'll try to pull up VBCurtis's score and (maybe) run another CADO search with the suggested parameters (vebis would do it way faster than me, is he around?)

2020-01-25, 21:55   #1880
swellman

Jun 2012

1011001110002 Posts

Quote:
 Originally Posted by Max0526 Hi everybody! Got lost in elliptic curves, hacked into some new Z6 families, see the bottom of the page: https://web.math.pmf.unizg.hr/~duje/tors/z6.html I am truly sorry to interrupt this holiday: https://www.mersenneforum.org/showpo...postcount=1537 It was also reposted later in https://www.mersenneforum.org/showpo...&postcount=136, the second post after the chart. To be helpful and productive, I'll try to pull up VBCurtis's score and (maybe) run another CADO search with the suggested parameters (vebis would do it way faster than me, is he around?)
Hi Max! Don’t bother with CADO as I extended VBCurtis’s work and found nothing better.

If you want to run his poly through your alchemy that might improve things a bit. I’m test sieving now and am closing on the best parameters - the throughput is good for a C189. But I will pause in case you take a crack at it.

Thanks.

 2020-05-08, 22:45 #1881 EdH     "Ed Hall" Dec 2009 Adirondack Mtns 5×677 Posts I'm playing with some alternate scripts for my machines that aren't working on the 198 team project. I might as well try to make them useful during this time. Is there a polyselect request that I might try some things with? I won't be trying to set too many values ATM, so something where I wouldn't duplicate effort and if I don't perform well, nothing is really detracted from the overall picture, would be preferred.

 Similar Threads Thread Thread Starter Forum Replies Last Post ixfd64 mersennewiki 169 2018-09-21 05:43 carpetpool Miscellaneous Math 14 2017-02-18 19:46 cheesehead Forum Feedback 6 2009-07-28 13:02 R.D. Silverman NFSNET Discussion 13 2005-09-16 20:07 TauCeti NFSNET Discussion 0 2003-12-11 22:12

All times are UTC. The time now is 02:16.

Thu Oct 29 02:16:45 UTC 2020 up 48 days, 23:27, 1 user, load averages: 1.36, 1.74, 1.80