mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Cunningham Tables (https://www.mersenneforum.org/forumdisplay.php?f=51)
-   -   Planning & Coordination for 2,2330L (https://www.mersenneforum.org/showthread.php?t=24292)

VBCurtis 2019-04-13 00:27

Planning & Coordination for 2,2330L
 
We have a GNFS candidate from the Cunningham project at 207 digits:
[code]334377437706404684733884220732190564550147039308686365251932207592010868660163280944362994773042289837865752943086095333357861608005044927356022538834108845220872215597190097852981295940487497385774178731621[/code]

A first guess for difficulty is 80-100 core-years for sieving, ~10 core-years for the matrix. This is thinly based on a GNFS-186 I'm currently doing with CADO, which requires ~4 core-years for sieving; my historical data for smaller CADO tasks indicates matrix time is often 1/6th to 1/5th of the sieving time; 207 is 4 doublings tougher than 186, so ~16x as much work.
I expect we'll oversieve this number enough to get matrix time down under 10 core-years since a volunteer will be solving it on a single machine (that's me, unless fivemack has other ideas with his fastest machine). We have a C205 and C206 sieving on 16f at NFS@home presently, with Tom and I each committed to a matrix to finish the jobs. This provides us a natural comparison with CADO on this C207, the better to see how CADO compares to msieve for resource needs and time.

swellman has scheduled a t65 on this number at yoyo; anyone eager to set an initial poly-score mark is invited to begin by reserving a coefficient-range on msieve-GPU or CADO in this thread.

If there is medium-term interest in 2,1165+, we can coordinate poly select and such while we sieve 2,2330L. If CADO post-processing proves unwieldy on resource needs or we find insufficient forum-user resources to sieve it, we can always send it to frmky and NFS@home for sieving.

While ECM and poly select is going, we need to collectively decide whether to team-sieve with CADO, or send it to NFS@home. I *strongly* prefer a team sieve, for a few reasons: To learn how well CADO scales for future GNFS-220+ tasks, because NFS@home already proved it can handle a number this size but we haven't, and because teamwork in this forum is fun.

To decide it's worth proceeding as a CADO-team-sieve, we ought to collect pledges from interested parties in units of core-years. Sieve tasks will be handed out as a 4-threaded job, using either 3-3.3GB of memory if I=15, or 12-14GB if I=16. (We can select any fixed number of threads, and that number can be different for the I=15 server versus I=16)
So, if you have a 16GB or greater machine, please estimate how many core-years you can pledge for I=16 and complete within calendar 2019 (figure we'll start by 1 June, so 6-8 months for sieving).
If you have 6-16GB ram, and wish to contribute, please pledge core-years for I=15. We'll test-sieve to find out how much less efficient I=15 is versus I=16; the CADO team factored RSA-704 (GNFS212) entirely on I=15, though they noted the suboptimal parameters meant it took them 500 core-years (eek!). This job is half as hard as GNFS-212, so a wild guess is that I=15 work will be half as efficient as I=16 work.
If we get pledges of I = 16 years + 0.5 * I=15 years = 80 or more, we can do this as a team sieve. Once we have a polynomial we can further refine this estimate via test-sieving.

VBCurtis 2019-04-13 00:30

I'm willing to put 20 cores toward this project, so I pledge 10 core-years I=16.
I'll use this thread to track pledges:
[code]vbcurtis I=16 10 core-years
swellman I=16 2 core-years
RichD I=16 1 core-year
lukerichards I=16 2 core-years
ET I=16 4 core-years
fivemack I=16 30 core-years
Total: 49 core-years[/code]

Edit: Both swellman and I pledged a number of cores for as long as it takes; I'm using 6 months as a calendar-time goal for the sieving. Fivemack's cores begin 1 July, so I scaled his down a bit to reflect a bit less time available.

swellman 2019-04-13 00:50

I will pledge 4 cores for the duration.

Also reserving c5 < 1M for poly search using msieve. (I’m assuming we are not searching at all for a sextic.)

Max0526 2019-04-13 13:28

C207 CADO poly select 0-1M
 
Reserving c5 < 1M for poly search using CADO. (Later on I'll take 0-1M for a CADO sextic, just to see the difference. It would be great to have somebody test-sieve top-c5 vs top-c6).

VBCurtis 2019-04-13 15:49

I'll surely test-sieve a few top polys, at worst in 16f GGNFS but likely the best two will get test-sieved in CADO to see if relative speed changes.

Max0526 2019-04-13 17:12

C207 score
 
@swellman
When you get to it, could you please post anything above 1e-15? I am stuck at 7.68e-16, and we are trying to get to 1.6e-15. Quite probably my parameters are way off.

swellman 2019-04-13 18:27

[QUOTE=Max0526;513620]@swellman
When you get to it, could you please post anything above 1e-15? I am stuck at 7.68e-16, and we are trying to get to 1.6e-15. Quite probably my parameters are way off.[/QUOTE]

You got it. Might take a while but I’ll post anything above that level.

RichD 2019-04-13 20:12

I'll pledge 4-cores at half time. I guess that amounts to 1-core year. This will work good for me since sometimes I have a box waiting 12, 24, 36 hours for sieving to complete at NFS@home and I can point it at the CADO server for a quick set of WUs (in addition to my pledge).

I'll start msieve-GPU at c5=4M tomorrow when I get home.

lukerichards 2019-04-13 21:13

Minimum 2 core years from me.

I've never used the software mentioned, but I propose to get started on one of my home computers and probably later devote some Google Cloud credit to it

Max0526 2019-04-14 15:32

C207 polys
 
[QUOTE=Max0526;513620]@swellman
When you get to it, could you please post anything above 1e-15? I am stuck at 7.68e-16, and we are trying to get to 1.6e-15. Quite probably my parameters are way off.[/QUOTE]
I got two so far:[code]# norm 1.534672e-20 alpha -7.343566 e 1.087e-15 rroots 3
skew: 101089318.80
c0: 155816101167630939131941933802766181845255176400
c1: -24270336909414063210712346128763647493743
c2: 1941539569306237251183718679832592
c3: -41711829877443769945297129
c4: -265321259164440600
c5: 7695600
Y0: -24893355147235925339215203881525122029329
Y1: 25390484222442108467594[/code][code]# norm 1.531481e-20 alpha -7.344243 e 1.081e-15 rroots 3
skew: 201773747.84
c0: 1238725398513114053251175452995353070314472059520
c1: -96455898644705516040804890845088702814652
c2: 3872983036737718442851378564937504
c3: -41754620765014257030865129
c4: -132659853865740300
c5: 1923900
Y0: -24893355147234901594891355015711708639249
Y1: 12695242111221054233797[/code]

Max0526 2019-04-14 23:05

C207 polys -- first 1.2's
 
[code]Y0: -19544845026703225729565432245313306145368
Y1: 10120116742258979082991
c0: 5973139622227297137449550863671749439844048257200
c1: -280720202442802238774504181104212579693210
c2: -1238186817666897836947815109808853
c3: 1780719008342744983579849
c4: -174906653491611360
c5: 123805440
skew 196606977.59
# size 1.416e-20, alpha -8.218, combined = 1.205e-15 rroots = 3[/code][code]Y0: -9772422513351884893520748044014403870764
Y1: 10120116742258979082991
c0: 1497056890375551628851845387618418364676688307500
c1: -140293528523944498177067344182767951355725
c2: -1238477045246699046008488861071573
c3: 3636669026514306751748498
c4: -699759729575533440
c5: 990443520
skew 98182419.04
# size 1.427e-20, alpha -8.240, combined = 1.203e-15 rroots = 3[/code]


All times are UTC. The time now is 02:10.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.