mersenneforum.org Reserved for MF - Sequence 4788
 Register FAQ Search Today's Posts Mark Forums Read

2021-09-21, 19:02   #3114
Batalov

"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

2·4,787 Posts

Quote:
 Originally Posted by bur Interesting, is it valid for most sizes of numbers? That'd be quite helpful.
Yeah, there are a few threads over years to the same effect. It is.

For smaller numbers/projects, it is very difficult to hit the exact cusp, so the usual rule of thumb for automatic scripts is "if you got the matrix, don't try to sieve extra - it will be a wash" (given that filtering itself is an overhead and together with sieving some more - you will not get any time savings). but for large projects, there can be a huge difference.

One could check logs that are kept in NFS @ Home collection, while binning for similar size, and very rarely you would see that the runner get a very large matrix and still decides to go along with it, and that's of course a fine personal preference. It depends on resources available to the runner; sometimes it is - optimizing for their own. human time, and sometimes it is the external resources (at the expense of human time for doing filtering twice or more).

Many logs (very educationally!) keep several recordings of filtering.

2021-09-21, 19:10   #3115
ryanp

Jun 2012
Boulder, CO

331 Posts

Quote:
 Originally Posted by charybdis What Q value have you sieved up to, and what is the overall duplication rate like? I'd be interested to know how well my estimates from test-sieving have held up.
That's unfortunately a bit tough to answer, due to the distributed nature of my sieving setup.

In any case, I now have a filtering run going with target_density=130 and 2.024B uniques. If this succeeds, I will hopefully be able to hand it off for frmky for LA.

 2021-09-23, 16:09 #3116 ryanp     Jun 2012 Boulder, CO 331 Posts Update: with 2.024B uniques and target_density=130, msieve produces a 133.3M x 133.3M matrix: Code: Wed Sep 22 18:39:19 2021 commencing 2-way merge Wed Sep 22 18:43:22 2021 reduce to 304431703 relation sets and 296088199 unique ideals Wed Sep 22 18:43:22 2021 commencing full merge Wed Sep 22 20:02:23 2021 memory use: 32549.0 MB Wed Sep 22 20:02:54 2021 found 134281671 cycles, need 133304399 Wed Sep 22 20:03:51 2021 weight of 133304399 cycles is about 17329637475 (130.00/cycle) Wed Sep 22 20:03:51 2021 distribution of cycle lengths: Wed Sep 22 20:03:51 2021 1 relations: 4545972 Wed Sep 22 20:03:51 2021 2 relations: 7597027 Wed Sep 22 20:03:51 2021 3 relations: 9794025 Wed Sep 22 20:03:51 2021 4 relations: 10584633 Wed Sep 22 20:03:51 2021 5 relations: 11171953 Wed Sep 22 20:03:51 2021 6 relations: 11129931 Wed Sep 22 20:03:51 2021 7 relations: 10851523 Wed Sep 22 20:03:51 2021 8 relations: 10319041 Wed Sep 22 20:03:51 2021 9 relations: 9582186 Wed Sep 22 20:03:51 2021 10+ relations: 47728108 Wed Sep 22 20:03:51 2021 heaviest cycle: 25 relations Wed Sep 22 20:04:20 2021 commencing cycle optimization I think we're at the point of diminishing returns, and frmky has graciously offered to help with the LA at this point.
 2021-10-01, 21:08 #3117 ryanp     Jun 2012 Boulder, CO 331 Posts frmky reports that this is now in LA on a multi-GPU system, with ~170 hours to go, plus or minus some time depending on his cluster's queueing.
 2021-10-02, 16:42 #3118 bur     Aug 2020 79*6581e-4;3*2539e-3 19316 Posts Nice, let's hope the 7 will vanish....
 2021-10-02, 17:08 #3119 charybdis     Apr 2020 1F916 Posts It can't - the 2^2 guarantees sigma(n) is divisible by sigma(2^2) = 7, so the next term which is sigma(n)-n will also be divisible by 7. Similarly the 7 means that sigma(n) will be divisible by 8, so the next term will also have exactly two factors of 2. To get rid of the 2^2*7 we would need a term where the 7 is raised to an even power and the remaining prime factors contribute at most two factors of 2 to sigma(n).
2021-10-04, 07:44   #3120
frmky

Jul 2003
So Cal

2·7·157 Posts

Quote:
 Originally Posted by ryanp frmky reports that this is now in LA on a multi-GPU system, with ~170 hours to go, plus or minus some time depending on his cluster's queueing.
Curiously,
Code:
linear algebra completed 103057 of 127518005 dimensions (0.1%, ETA 75h33m)
That's on 160 (!) A64FX nodes. Unfortunately I can't use that many for 3 days...

2021-10-04, 12:22   #3121
ryanp

Jun 2012
Boulder, CO

331 Posts

Quote:
 Originally Posted by frmky Curiously, Code: linear algebra completed 103057 of 127518005 dimensions (0.1%, ETA 75h33m) That's on 160 (!) A64FX nodes. Unfortunately I can't use that many for 3 days...
Still, pretty fun to see them chew through a matrix of this size so quickly!

How's the other run progressing?

2021-10-04, 18:44   #3122
frmky

Jul 2003
So Cal

2·7·157 Posts

Quote:
 Originally Posted by ryanp How's the other run progressing?
Live off the cluster:
Code:
linear algebra completed 60978666 of 127518005 dimensions (47.8%, ETA 90h39m)

 2021-10-04, 22:45 #3123 VBCurtis     "Curtis" Feb 2005 Riverside, CA 10011100100102 Posts Ryan- Now that sieving was shut down, do you have even a rough guess for Q-range that was sieved? Or the number of raw relations? It's nice to have confirmation that 2G uniques was enough for this job; that means we can go 35/35 and still use msieve for a future C22x GNFS job. Also, did you do any A=32 sieving, or all I=16 (same as A=31)? Last fiddled with by VBCurtis on 2021-10-04 at 22:45
2021-10-05, 12:40   #3124
ryanp

Jun 2012
Boulder, CO

331 Posts

Quote:
 Originally Posted by VBCurtis Ryan- Now that sieving was shut down, do you have even a rough guess for Q-range that was sieved? Or the number of raw relations?
Unfortunately I don't have a log of this now, but I can keep one for future jobs.

My methodology is essentially: fire off a large number of CADO jobs with different q ranges in parallel, then periodically gather the gzipped files, concatenate and remdups4 the merged file, and try an msieve run. Then wait a few hours and repeat. Unfortunately, I don't have the logs of the remdups4 output now.

 Similar Threads Thread Thread Starter Forum Replies Last Post kar_bon Aliquot Sequences 136 2021-10-21 16:17 RichD Aliquot Sequences 476 2021-10-04 20:47 RichD Aliquot Sequences 524 2021-09-06 21:00 prism019 GPU to 72 6 2020-09-21 22:11 petrw1 Lone Mersenne Hunters 82 2010-01-11 01:57

All times are UTC. The time now is 06:16.

Wed Oct 27 06:16:09 UTC 2021 up 96 days, 45 mins, 0 users, load averages: 1.10, 0.98, 0.97