mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Riesel Prime Search (https://www.mersenneforum.org/forumdisplay.php?f=59)
-   -   Low weight stats page. (https://www.mersenneforum.org/showthread.php?t=2934)

Kosmaj 2005-09-19 03:05

[B]Citrix[/B]
All k's you listed above are already on the [URL=http://www.15k.org/lowweight.htm]low weight stats page[/URL]. Creating another page will require updating two pages instead of one. But if you want to create such a page you are welcome, I can put it up on the server.

About reserving 53 values, I think it's not realistic. How many machines do you have? Some of these k's are already checked to 1M or more. So please have a look at the stats page and tell me what k's do you want to reserve so that you can have some results within a month or so.

Citrix 2005-09-19 03:27

If you read my post it says if no one else is interested. I can always reserve work as needed from main thread as available if other people are also interested. But if no one else is interested, you reserve them for me.

Kosmaj, do you have a script to generate such a page. If you could generate such a page for me, I can update it myself.

Ill take 48973 to start with and when I am done, take more.

Thanks,
Citrix

Citrix 2005-09-19 04:14

Instead of working on all of them, I would like to search all of them for primes under n=100K. then take the best 2-3 candidates higher. Since I don't have enough power to work on all of them myself.

Citrix

Citrix 2005-09-19 04:30

Please help test these to 100K

48973 81 --4 primes // checked by citrix
74201 92
124679 80
131069 92
192509 88
216367 92
256453 86
278077 95
284579 83
313979 99
327163 73
332159 95
334331 81
370421 99
473279 91
487811 86
612509 74
643843 91
648433 95
671413 92
685183 99
700057 92
700477 84
780427 71
783073 75
953429 95
961573 75
967597 65
981493 49
1006469 79
1034503 66

Kosmaj 2005-09-19 07:40

[B]Citrix[/B]
Please stop this nonsense!

[QUOTE]48973 81 --4 primes // checked by citrix[/QUOTE] The above 4 primes have been already found by Joss, go and read the [URL=D:\minovic\proth\k15\webSite\lowweight.htm]Low weight stats page[/URL].

[QUOTE]74201 92[/QUOTE] k=74201 is already checked by Marcin to n=335k, 3 primes found.
And so on and so on. Your call is meaningless.

Now, with respect to the following:
[QUOTE]If you read my post it says if no one else is interested. I can always reserve work as needed from main thread as available if other people are also interested. But if no one else is interested, you reserve them for me.[/QUOTE] I'd like to inform you that this is not the way we discuss matters here! [I]Please[/I], calm down. Thank you.

Citrix 2005-09-19 07:46

[QUOTE=Kosmaj][B]Citrix[/B]
Please stop this nonsense!

The above 4 primes have been already found by Joss, go and read the [URL=D:\minovic\proth\k15\webSite\lowweight.htm]Low weight stats page[/URL].

k=74201 is already checked by Marcin to n=335k, 3 primes found.
And so on and so on. Your call is meaningless.

Now, with respect to the following:
I'd like to inform you that this is not the way we discuss matters here! [I]Please[/I], calm down. Thank you.[/QUOTE]

Sorry for the inconvinience. :no: :redface:

I have checked the k to 100K and there are no new primes to report.


Citrix

gribozavr 2005-09-19 16:35

[QUOTE=VBCurtis]Gribozavr-- how many candidates are left? I can give you a decent estimate of ideal sieving, if you can provide me the weight of the k-value (number of candidates left from n=518k to n=1M is sufficient).[/QUOTE]

I have sieved to 1,5T and from n=518482 to n=1M there are 1610 candidates left.

[QUOTE=VBCurtis]If you have recent data on time per LLR or time per exponent sieved, that will help corroborate my estimation (I want to see if my estimations are accurate, so I'd use this data to double-check).[/QUOTE]

For n=518k LLR time is about 3500 sec.

[QUOTE=VBCurtis]Also, do you have a P3 or Athlon to do the sieving? Ideal sieving depth if you only possess a P4 is nowhere near as deep as if you can sieve an athlon and LLR with a P4, since the P4 is relatively so much better at LLR than sieving. If you have only P4, I don't mind running a sieve for a week or so on an Athlon to give you a boost (you wouldn't have to credit me for any primes found).[/QUOTE]

I'm doing sieving and LLR'ing on Celeron 1.2 GHz. But I also have an Athlon 64 3000+ 1.8Ghz here.

VBCurtis 2005-09-21 21:06

Sieving on a celeron is a waste of resources, if you have an Athlon.. The athlon is over 50% more efficient at the same speed (and yours is even faster than your celeron, so even faster, relatively!)
edit: you said celeron 1.2-- that's P3 based, I think, making it better even than an Athlon for sieving. Never mind!

1.5T should be plenty for sieving to n=1M, for any low-weight number. I'll find my formula notes at home tonight and see if I can find the p-value estimation.

The most basic equialence says to sieve until time per factor found is equal to LLR time on an exponent 70% of the way from min-n to max-n. For you, that's around n=850k. LLR time increases roughly with the square of n, so LLRing an 850k power should take you about 9000 seconds per power. If 10 or more factors are found in a day, you should continue sieving. If 8 or fewer, stop for sure. If 9 factors in a day, it's right at the "sieved enough" point, and you should do whichever you prefer, as the choice is not relevant from an efficiency standpoint.

-Curtis

VBCurtis 2005-09-21 21:30

Citrix:
[QUOTE=Citrix]Instead of working on all of them, I would like to search all of them for primes under n=100K. then take the best 2-3 candidates higher. Since I don't have enough power to work on all of them myself.

Citrix[/QUOTE]
Getting all the powers up to some level isn't a bad idea, but why 100k? These candidates have such low weight that finding 3 small primes instead of 1 doesn't really indicate "best"ness for n-values above 50k or so. For proof, look at the number of bases that have no prime to even 500k, so running to 100k will not help in picking a "good" candidate.
If you want to LLR just small numbers, then sieve each candidate to 100G or so (for N-values up to 1M, since that is the standard on the stats page and server), LLR to your preferred cutoff, and post results. Doing this to 100k takes more time to post and admin than to actually run, so isn't very helpful. Choose a cutoff point that has you processing no less than 2 weeks on each number, and I'm happy to help admin/update the page/etc for your work. You can even email me the sieved-but-not-LLRed files, which I'll eventually post on the server.

I suggest n=500k as a nice balance between variety and depth, to keep you interested and the administration manageable. 400k is tolerable, but 300k will happen REALLY fast on each number--I think I used to get to 250k in a weekend on a P3-500, so a modern machine could sieve to 100G and LLR to 300k in something like 2 days. I imagine even 400k would happen in less than a week, thus my suggestion of 500k as your target. This would still get you through all 53 numbers by yourself in something less than 6 months, less if someone decides to join your method of progress.

Note that even 1M takes around a month on a mid-level (2.4Ghz) P4-- hardly a huge commitment. I have a bunch of public-school P3-class machines running, which is why my updates seem so much slower than these estimates indicate.

-Curtis

VBCurtis 2005-09-21 21:51

reserving k=403993, to help Citrix with his search. I'll put a P3 from school on it, and report times to sieve to 100G and LLR to 500k.
-Curtis

Kosmaj 2005-09-21 22:07

[B]Citrix[/B], no problems, please take a few at the time and keep on going.

But I agree that checking one k to n=100k will not be very helpful in deciding is it a prolific one. For example for k=80857169 for n up to about 1.3M (?) Thomas found only two primes, for n=4 and n=1251076.

Finally, I don't want to sound like a parrot :smile: :rolleyes: but k=403993 mentioned by Curtis has been already checked by Joss to 405k. Do you intend to double check? :cool:


All times are UTC. The time now is 16:52.

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