mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Cunningham Tables (https://www.mersenneforum.org/forumdisplay.php?f=51)
-   -   Contributing to Cunningham Project (https://www.mersenneforum.org/showthread.php?t=24211)

VBCurtis 2019-06-30 18:47

It was a surprise to me to learn that the smaller tasks sieved by Greg on 16f are so well- sieved that the matrices can be solved on single machines. For instance, the C206 I am post-processing had both filtering and LA fit in a 32GB machine. That means that with a little patience, anyone with a 48GB+ machine can solve some of Greg's easier matrices for him, allowing him to use his cluster-time to solve the ones sized 60M+. I hope there are such jobs available to us, that they're not all too big already.

I plan to do just this with my 64GB box this fall. Perhaps Mr Womack will also run one matrix for NFS@home this year also, reducing Greg's backlog by 2 jobs. I still plan to expand my box to 128GB memory around the time 2330L finishes sieving, which would make patience rather than capacity the limiting factor for my own matrix-solving.

Again, C206 LA requires just 6 weeks on my machine; Mr Womack's best machine is likely twice as fast. I'm fine with 10 weeks per matrix personally, which is likely somewhere around 55M matrix size. There's no reason to think the cluster Greg uses to be limited to 100M matrices; at 120M to 140M, SNFS-330 should be covered?

I personally enjoy variety in projects a bit, but mainly I enjoy efforts to optimize tools for peak speed. I've learned quite a bit during the 2230L project, and I'd like to try to apply that knowledge to a C201-C205 before tackling 2330M. Perhaps that means Mr Silverman's advice to leave these C210-C217 GNFS jobs to NFS@home is what we must do; but I very much enjoy the team-sieve happening now, and I think it has attracted contributors that would not have sieved for NFS@home. With that in mind, I think CADO-team-sieve has merit for attracting CPU cycles and for advancing the development of CADO parameters for faster factorizations.

R.D. Silverman 2019-06-30 18:48

[QUOTE=swellman;520388]Mr. Silverman, thank you for the update. My records now reflect the new information. A few clarifications:

- Why skip ECM 2,2246M? It’s a C221 that seems beyond SNFS.
[/QUOTE]

It was (accidently) listed out of order in my file....

[QUOTE]
- Isn’t 2,1157+, with a SNFS difficulty of 322, also within reach of NFS@Home?
[/QUOTE]

Yep. I missed that 1157 is divisible by 13.

[QUOTE]
- 2,2162L and M both seem similar in difficulty to 2,2158L, i.e. SNFS 325. Which is scary difficult but are either a feasible candidate for NFS@Home?
[/QUOTE]

One must choose a cutoff. I chose 1080 bits. One can usually find a number that is
"just a little bit bigger" to do. The cutoff is a pure guess, based on some informal
remarks made by Greg about the LA.

[QUOTE]
- Lastly, 2,1139+ pops up as an octic(!) with difficulty 323. Another reachable candidate perhaps, or a complete nonstarter?
[/QUOTE]

I think out of reach; The LA is problematic, as is sieving with an octic.

R.D. Silverman 2019-06-30 18:54

[QUOTE=VBCurtis;520403]It was a surprise to me to learn that the smaller tasks sieved by Greg on 16f are so well- sieved that the matrices can be solved on single machines. For instance, the C206 I am post-processing had both filtering and LA fit in a 32GB machine. That means that with a little patience, anyone with a 48GB+ machine can solve some of Greg's easier matrices for him, allowing him to use his cluster-time to solve the ones sized 60M+. I hope there are such jobs available to us, that they're not all too big already.

[/QUOTE]

I get the impression that the current set of NFS@Home numbers waiting for LA require
> 64GB and a few dozen cores minimum.

I would be [b]very[/b] pleased to find that it is not true and that others might help Greg with the LA.

VBCurtis 2019-06-30 18:55

[QUOTE=R.D. Silverman;520404]One must choose a cutoff. I chose 1080 bits. One can usually find a number that is "just a little bit bigger" to do. The cutoff is a pure guess, based on some informal remarks made by Greg about the LA.[/QUOTE]

Thank you; this addresses my questions about SNFS-330 and what candidates are reasonable for Greg on SNFS.

R.D. Silverman 2019-06-30 19:07

[QUOTE=Batalov;520401]Here is what we have for the visible future. (ECM can be run for all 147 base-2 targets, regardless)
[CODE]224 2,2586L 259.4 0.861 /quartic/resvd
228 2,2150M 258.8 0.88 /5q
211 2,2210M 266.1 0.792 /5q
225 2,2230M 268.5 0.837 /5q
253 2,1115+ 268.5 0.942 /5q
223 2,1135+ 273.3 0.815 /5q
299 2,2098L 315.7 0.946 /resvd
282 2,2102L 316.3 0.891 /resvd
300 2,1052+ 316.6 0.947 /resvd
274 2,1144+ 317.9 0.875 /13
219 2,2126M 319.9 0.684
281 2,1063+ 319.9 0.878
270 2,1157+ 321.5 0.839 /13
271 2,1072+ 322.7 0.839
268 2,1253- 323.3 0.827 /7
309 2,2506M 323.3 0.954 /7
238 2,1076+ 323.9 0.734
296 2,2158L 324.8 0.911
236 2,2162M 325.4 0.725
258 2,2162L 325.4 0.792
318 2,1084+ 326.3 0.974
265 2,2534M 326.9 0.809 /7
314 2,2534L 326.9 0.958 /7
273 2,2174L 327.2 0.834
309 2,2174M 327.2 0.944 [/CODE][/QUOTE]

Let's finish the original tables before starting index > 1200......

swellman 2019-06-30 19:22

[QUOTE=R.D. Silverman;520404]
I think out of reach; The LA is problematic, as is sieving with an octic.[/QUOTE]

I believe Greg recently [url=https://www.mersenneforum.org/showpost.php?p=511744&postcount=259]updated msieve[/url] to better handle octics but I have no idea if those changes scale to composites of this difficulty.

Thanks again for the additional information - I’ve updated my records accordingly, including plans to enqueue 2,2246M for ECM by Yoyo@Home.

Got some good news: Yoyo is allowing me to enqueue the t65 work in blocks of 12000 curves rather than 9000. Hoping 6x submissions are a bit faster than 8x in the BOINC environment.

R.D. Silverman 2019-06-30 19:36

[QUOTE=swellman;520409]I believe Greg recently [url=https://www.mersenneforum.org/showpost.php?p=511744&postcount=259]updated msieve[/url] to better handle octics but I have no idea if those changes scale to composites of this difficulty.
[/QUOTE]



When the NFS norms are unbalanced it results in much larger matrices.
Norms are unbalanced when the degree is smaller or larger than optimal.

sweety439 2019-07-01 12:55

[QUOTE=swellman;520361]2,1165+ and 2,2210M are being worked to t65 thanks to Yoyo@Home. They should be fully ECMd in a few weeks. Any preferences on our next ECM target?

3,748+ c204 has been [url=https://www.mersenneforum.org/showthread.php?t=24548]suggested[/url]. Seems like a worthy candidate to me, but the effort so far has been focused on factoring all remaining 2+ (n<1200) and 2LM (n<2400). Do we limit our efforts to only that list? Scope creep has been the downfall of many projects...

I have asked Greg Childers where he is heading with the 16f queue so that we don’t step on each other. He may want our help in some sector. Perhaps he will chime in here.[/QUOTE]

Also 10, 323- c271 has been [URL="https://mersenneforum.org/showthread.php?t=24550"]suggested[/URL], not only 3,748+ c204

R.D. Silverman 2019-07-01 15:55

[QUOTE=sweety439;520455]Also 10, 323- c271 has been [URL="https://mersenneforum.org/showthread.php?t=24550"]suggested[/URL], not only 3,748+ c204[/QUOTE]

Suggested for what? NFS@Home will get to it, as I [b]already[/b] pointed out. It doesn't
need more ECM. It is beyond the capabilities of the mersenneforum. So I
therefore ask: What you are suggesting to do with it?

VBCurtis 2019-07-01 15:59

[QUOTE=sweety439;520455]Also 10, 323- c271 has been [URL="https://mersenneforum.org/showthread.php?t=24550"]suggested[/URL], not only 3,748+ c204[/QUOTE]

You suggested this one, and were told it's going to wait until easier ones are done first. Please don't keep asking.

sweety439 2019-07-02 12:42

[QUOTE=R.D. Silverman;520471]Suggested for what? NFS@Home will get to it, as I [b]already[/b] pointed out. It doesn't
need more ECM. It is beyond the capabilities of the mersenneforum. So I
therefore ask: What you are suggesting to do with it?[/QUOTE]

Currently, there is [B]NO[/B] "10,323- c271" in the page [URL="http://homes.cerias.purdue.edu/~ssw/cun/who"]http://homes.cerias.purdue.edu/~ssw/cun/who[/URL].


All times are UTC. The time now is 11:58.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.