mersenneforum.org fond of a factor? Turn yourself in to become inane
 Register FAQ Search Today's Posts Mark Forums Read

 2011-01-21, 03:14 #78 KingKurly     Sep 2010 Annapolis, MD, USA 33×7 Posts From P-1: M1983503 has a factor: 186794246935371583225132033 Does anyone have recommendations for other places to run P-1? I did a bunch of M1983xxx to celebrate my recent birthday, which may have been in 1983. I have plenty of cores dedicated to "normal" P-1 (which is mostly in the 53M range right now), but I'm interested in running P-1 in other ranges too.
 2011-01-21, 17:42 #79 lycorn     Sep 2002 Oeiras, Portugal 1,399 Posts From the point of view of project progress (defining progress as moving forward the leading edge of first-time LL tests), the type of P-1 you are doing (53M range) is the most appropriate, specially if you are using considerable amounts of memory. Therefore, from that POV, that is my recommendation. But of course, as repeated ad nauseam by many forumites (including myself), it´s your machine, so do as you like it the most and enjoy the participation. Every bit counts! Last fiddled with by lycorn on 2011-01-21 at 17:42
 2011-01-21, 19:43 #80 KingKurly     Sep 2010 Annapolis, MD, USA 33×7 Posts Understood; I still have several cores dedicated to 'mainline' P-1 (53M range), and I am using considerable amounts of memory on my main machine... about 6GB dedicated to mprime with two cores running that sort of P-1. Then recently I've had a third core running "small" P-1. I found a few more factors in M1983xxx and decided to start poking through M1982xxx and M1984xxx. Not sure where I'll head from there, but it seems like I'm finding a lot of exponents in the small ranges that have never had stage 2 of P-1. Looks like the stuff under 1M has mostly been done, but there were a bunch I saw between 1M and 2M. It looks like we are assigning ECM curves when the exponent has not had a "proper" P-1, am I missing something there? I do understand that finding factors to smaller Mersenne exponents will not help find the next Mersenne prime. Fortunately for the project, I also have several cores dedicated to "real P-1", LL-D, and LL. Heck, I even have a few smaller cores running TF, and even a bit of ECM. (And yes, I understand that ECM finds factors to small Mersenne exponents, which like my 'small P-1' pursuit, does not help find the next prime.) You are right, they are my machines, and I like what I'm doing. :) I just thought it'd be fun to have one core working on "exploratory" work, although perhaps that isn't the right word. Non-standard work? I don't know. I did notice a bunch of exponents below 1M that have P-1 bounds with B1=B2=exponent. For example M909107 has P-1 bounds of 909107. Was that value as arbitrarily selected as the B1=1M, B2=30M that I've been using? Sorry for the jumbled stream-of-consciousness style of this post.
 2011-01-21, 23:35 #81 lycorn     Sep 2002 Oeiras, Portugal 1,399 Posts A suggestion is to do some work on the 332M range. That´s where the 100M digit numbers start, and there are a few people working on that range, mainly doing TF, but also some P-1 (and even LL tests!...). As you said you would like to do some "exploratory" work, you might enjoy joining. A generous amount of memory is recommended for numbers this big. You may wish to have a look at this thread: http://www.mersenneforum.org/showthread.php?t=10693
2011-01-21, 23:59   #82
Uncwilly
6809 > 6502

"""""""""""""""""""
Aug 2003
101×103 Posts

100001100100102 Posts

Quote:
 Originally Posted by lycorn A suggestion is to do some work on the 332M range. That´s where the 100M digit numbers start, and there are a few people working on that range, mainly doing TF, but also some P-1 (and even LL tests!...).
There are currently 18 available for P-1 in the 332M range. And from 332400000 to 332999999 is all essentially at 67 bits. There are tons of factors out there. (From 332192831 (the lowest 100M digit) to 332399999 is at 70 or better).

2011-01-22, 03:50   #83
KingKurly

Sep 2010
Annapolis, MD, USA

33×7 Posts

Quote:
 Originally Posted by Uncwilly There are currently 18 available for P-1 in the 332M range. And from 332400000 to 332999999 is all essentially at 67 bits. There are tons of factors out there. (From 332192831 (the lowest 100M digit) to 332399999 is at 70 or better).
I've started a P-1 of M332205149, it's using an 18M FFT and "up to 6144MB" of memory, B1 of 3365000, B2 of 95902500, and currently estimated to complete in early May or thereabouts.

Now I've got high (332M), medium (53M), any suggestions for low? I notice there are 60k+ factored exponents below 1M, how can I help advance that number? (I wonder if I will live to see the day that they all have at least one factor?) Or if sub-1M exponents are pretty well covered, maybe sub-10M? I've been finding more factors in the 1982xxx-1984xxx area by manually adding Pminus1 lines. Though doing Pminus1 makes me nervous a bit because I don't seem to get assignment IDs for them; I certainly don't want to be stepping on any toes. Doing Pfactor and/or using the manual assignment form seems to work okay, but only if Primenet believes it needs further P-1. Using Pfactor for a low exponent seems to use pretty wimpy B1/B2, so I have specified some beefier values in a Pminus1 line.

...What am I doing, it's a Friday night and I'm playing with Mersenne numbers. Hoo boy... :P

2011-01-22, 10:41   #84
ckdo

Dec 2007
Cleves, Germany

10000100012 Posts

Quote:
 Originally Posted by KingKurly Using Pfactor for a low exponent seems to use pretty wimpy B1/B2, so I have specified some beefier values in a Pminus1 line.
Code:
Pfactor=...,1,2,8323009,-1,64,10
gives me

Code:
[Worker #2 Jan 22 11:19] Optimal P-1 factoring of M8323009 using up to 3560MB of memory.
[Worker #2 Jan 22 11:19] Assuming no factors below 2^64 and 10 primality tests saved if a factor is found.
[Worker #2 Jan 22 11:19] Optimal bounds are B1=510000, B2=15682500
[Worker #2 Jan 22 11:19] Chance of finding a factor is an estimated 7.7%
[Worker #2 Jan 22 11:19] Using Pentium4 type-3 FFT length 448K, Pass1=448, Pass2=1K
That's a start.

 2011-01-22, 17:29 #85 ckdo     Dec 2007 Cleves, Germany 232 Posts M8401051 has a factor: 278879821834503938295782902841 k=2*2*5*11*131*113947*165541*30531703 B1=520000, B2=15990000, E=12
2011-01-22, 22:52   #86

"Richard B. Woods"
Aug 2002
Wisconsin USA

769210 Posts

Quote:
 Originally Posted by KingKurly It looks like we are assigning ECM curves when the exponent has not had a "proper" P-1, am I missing something there?
I presume that by "proper", you mean that the B1/B2 are sufficiently high to satisfy your (or someone's) sense or propriety.

Though some exponents that had only very low B1/B2 (such as 30) have had their P-1 history expunged from the database, others with what-I'd-consider-inadequate B1/B2 were allowed to remain. What's the lowest B1 you see in the database in the ranges you've examined?

Quote:
 I did notice a bunch of exponents below 1M that have P-1 bounds with B1=B2=exponent. For example M909107 has P-1 bounds of 909107. Was that value as arbitrarily selected as the B1=1M, B2=30M that I've been using?
If you mean, "did the prime95/mprime algorithm ever select B1=B2=exponent, such as B1=B2=909107 for M909107?" the answer is "no". Those were all specified manually by whoever ran the tests.

That would be true generally for any B1/B2 that are not multiples of 5000 and 1250, respectively.

Quote:
 Sorry for the jumbled stream-of-consciousness style of this post.
Why should yours be any different? :-)

Quote:
 Originally Posted by KingKurly I've been finding more factors in the 1982xxx-1984xxx area by manually adding Pminus1 lines. Though doing Pminus1 makes me nervous a bit because I don't seem to get assignment IDs for them; I certainly don't want to be stepping on any toes.
Here's how to reserve them and get assignment IDs:

1. Before calling PrimeNet, substitute, or add, a Factor= line, specifying bit levels above whatever those exponents already have (you might be able instead to use Test= or any other keyword that's not for P-1; I used only Factor=) for each exponent you want to P-1 on.

2. Then, contact PrimeNet. PrimeNet will assign you an ID (if no one else is working on it, other than ECM, which can tolerate multiple simultaneous assignments) for each of your Factor= lines.

3. For the exponents that did not get an assignment ID, that means someone else has reserved it, so don't process those exponents now.

4. Then, copy the assignment IDs into your P-1 worktodo lines and put them back into the worktodo (and remove the Factor= lines), for only the exponents for which you _did_ get an assignment ID.

5. If you don't report any of the P-1 to PrimeNet until all are completed, your use of the assignment IDs for the P-1 won't cause any trouble. (I don't recall whether there was any trouble when I contacted PrimeNet while I had unfinished P-1s with the copied assignment IDs still in the worktodo. Maybe there wasn't then, either.)

Last fiddled with by cheesehead on 2011-01-22 at 23:16

2011-01-23, 00:30   #87
KingKurly

Sep 2010
Annapolis, MD, USA

33×7 Posts

Quote:
 Originally Posted by ckdo Code: Pfactor=...,1,2,8323009,-1,64,10 gives me Code: [Worker #2 Jan 22 11:19] Optimal P-1 factoring of M8323009 using up to 3560MB of memory. [Worker #2 Jan 22 11:19] Assuming no factors below 2^64 and 10 primality tests saved if a factor is found. [Worker #2 Jan 22 11:19] Optimal bounds are B1=510000, B2=15682500 [Worker #2 Jan 22 11:19] Chance of finding a factor is an estimated 7.7% [Worker #2 Jan 22 11:19] Using Pentium4 type-3 FFT length 448K, Pass1=448, Pass2=1K That's a start.
Aha, so the trick is to trick the software/server into believing that we'll be saving more primality tests than we really are. Sounds good.

2011-01-23, 00:38   #88
KingKurly

Sep 2010
Annapolis, MD, USA

33×7 Posts

Quote:
 Originally Posted by cheesehead I presume that by "proper", you mean that the B1/B2 are sufficiently high to satisfy your (or someone's) sense or propriety. Though some exponents that had only very low B1/B2 (such as 30) have had their P-1 history expunged from the database, others with what-I'd-consider-inadequate B1/B2 were allowed to remain. What's the lowest B1 you see in the database in the ranges you've examined?
Yes, you are correct in your understanding of my use of the word 'proper'. I'd seen some with B1=B2=40k; I guess I can poke around the DB a little bit more and see if I find anything lower.

Quote:
 Originally Posted by cheesehead If you mean, "did the prime95/mprime algorithm ever select B1=B2=exponent, such as B1=B2=909107 for M909107?" the answer is "no". Those were all specified manually by whoever ran the tests. That would be true generally for any B1/B2 that are not multiples of 5000 and 1250, respectively.
Gotcha.

Quote:
 Originally Posted by cheesehead Here's how to reserve them and get assignment IDs:
Between your advice and ckdo's advice, I think I've got it figured out now; thanks! It looks like Pfactor lines will suffice after all, as long as I fudge the number of tests that would be saved. I was being "too truthful" and answering 0, which was auto-selecting pretty low B1/B2. When doing Pminus1 lines, I wasn't using any particular formula to pick B1/B2, just using a larger value for B1 than I saw in the DB, and then B2 somewhere around 20 or 30x B1.

 Similar Threads Thread Thread Starter Forum Replies Last Post ewmayer Programming 3 2016-09-30 07:15 jasong jasong 18 2013-11-15 22:54 Rafael Information & Answers 12 2012-01-02 19:38 rogue Lounge 10 2008-11-21 05:25 nngs Hardware 0 2005-05-20 01:31

All times are UTC. The time now is 15:09.

Sun Sep 20 15:09:57 UTC 2020 up 10 days, 12:20, 1 user, load averages: 1.59, 1.56, 1.47