Cunningham ECM efforts

[frmky]

2,2222M is done. The 166.4M matrix took 84 hours to solve using 8x A100 gpus.

[xilman]

Originally Posted by pinhodecarlos

Just to share with you all Yoyo ecm subproject has been chosen as one of the disciplines for the boinc penthalon. We shall see some nice boost for 14 days.

Yoyo became so concerned at the rate the tasks were being deleted that he mailed some of us asking for more work to be uploaded.

Nothing had happened with the CW queue for months; turned out that it had been disabled because the queue had drained and never been re-enabled at the end of last year when I uploaded more work.

There are now 20K jobs at B1=260M (i.e. t60) for each of 150-ish GCW numbers lying in the range 210 through 229 digits. I doubt that lot will be cleared out any time soon but will be paying attention in case I am pleasantly surprised.

Those parameters were carefully chosen to ensure that the cofactor will be either prime or an easy GNFS runt.

One factor, a p65, has already been found.

[xilman]

From yoyo's status page.

5 May 2023 2:20:21 UTC: 222 numbers and 49,021,203,178,191,588,779,805,498,404,773,459,890,685,805,986,392,540,314,833,570,519,033,561,936,348,293,019,006,003,065,518,083,046,626,000,225,179,047,533,500,450,802,266,461,291,958,188,220,613,345,012,477,374,857,379,363,181,514,099,488,463,232,269,399,297,607,712,564,312,724,508,293,640,879,079,424 curves are waiting in input queue.
CPU Years: 3,371.39 Credits: 5,906,676,334 GFlop: 2,551,684,176,288

10²²⁹ curves is just plain silly. I suspect that someone put a number-to-be factored into the number-of-curves column.

[pinhodecarlos]

Think you'll be surprised, you might need to add more. Look now at https://www.rechenkraft.net/yoyo/y_s...ecm.php#tabs-3

At the moment there are more than 200,000 Wu's ( for all ECM projects) in progresswhere each Wu covers for 5 (I think) ECM curves.

[Rubiksmath]

Originally Posted by xilman

From yoyo's status page.

5 May 2023 2:20:21 UTC: 222 numbers and 49,021,203,178.................... curves are waiting in input queue.
CPU Years: 3,371.39 Credits: 5,906,676,334 GFlop: 2,551,684,176,288

I suspect that someone put a number-to-be factored into the number-of-curves column.

Unless this mistake has been made more than once you should be able to extract the number they were attempting to factor by grabbing the first 200 digits of this number and searching a database of numbers of these forms needing to be factored.

[xilman]

Originally Posted by pinhodecarlos

Think you'll be surprised, you might need to add more. Look now at https://www.rechenkraft.net/yoyo/y_s...ecm.php#tabs-3

At the moment there are more than 200,000 Wu's ( for all ECM projects) in progresswhere each Wu covers for 5 (I think) ECM curves.

I hope I will be surprised.

At the moment I have 3M curves still to be done at B1=260M. Each curve takes 1-2 core-hours on typical systems, according to my ECMNET server logs scaled by a factor of 260/43 to allow for the higher B1. If my arithmetic is correct, 3M curves comes to around 8200 core-years in total.Im am only 1 of six functioning projects right now, so unless one or more of the others run dry, something like 50M core-years will have been run by the time my inpout queue runs dry.

As noted previously, I will be paying attention.

[pinhodecarlos]

Paul, my baseline reference are the 20 numbers I see from that link I've posted before, not the 150 if all were added to it. Think there's another link to follow up all the queue but can't seem to remember where it is. Looking forward to see for all ECM projects yesterday picture and within 13 days. I can see some.of the heavy hitters not showing expected output so they might be holding for some of the challenge days ahead nuances

[Andrew Usher]

Yes, the 'gang of 31' are now 24 (after the in-progress number, 1109+, finishes), while I'd assumed 31. These are, of course, the remaining numbers from the print editions (the last higher-base number from the third edition was done last year). Just one can marginally use an algebraic factor (1139+ is planned as the /17 octic, though even at that size it can't be that much better than the sextic), and their sizes range from 338 to 361. Directly adding those difficulties, using the estimates here and the theoretical formula, makes the total of the 24 still pretty close to that of M1277, but probably a bit less. Surely, though, this depends on memory allocated for sieving, as larger numbers must benefit more from extra memory. At the current limits both would take many years; the factory approach has promise, and has been demonstrated, but would require at least as big a change as raising the limits.

Calling the limit of the exponent list 'arbitrary' was not an judgement but a fact - it has no mathematical basis. If it was established in the 1960s (before complete factorisations to that level were at all conceivable) that hardly makes it less so.

Greg doesn't post here himself, so I have no basis to predict his decision, which will be needed promptly as just that one number in progress remains.

[charybdis]

Originally Posted by Andrew Usher

Just one can marginally use an algebraic factor (1139+ is planned as the /17 octic, though even at that size it can't be that much better than the sextic)...

I'm not aware of it being planned one way or the other. The sextic and octic will need to be carefully test-sieved when the time eventually comes to do it.

Greg doesn't post here himself...

He's made 21 posts in this thread, though you're right that he rarely gives any indication of his plans.

[xilman]

Originally Posted by xilman

From yoyo's status page.

...

10²²⁹ curves is just plain silly. I suspect that someone put a number-to-be factored into the number-of-curves column.

Looking like my tip-off to yoyo worked. Now 179 numbers are waiting in the input queue.

[xilman]

Two factors found so far, one of which has left a C162 runt which I will complete by GNFS.