Ranges

Xyzzy · 2004-01-05, 21:08

Here is 100M to 101M... I have taken the results and made two files... The first file is the factors found and the second file is the nofactor data... The files use the same configuration as George's data files, except I added a third field to the factor file to show who found the factors...

For this 1M range I tested 58,043 exponents to 50 bits and I found 28,208 factors in 3h52m35s on a 2.67GHz P4...

http://www.mersenneforum.org/lmh/

I don't know if this work is productive or not, but it is kind of fun... I'd be willing to coordinate these searches if others feel it is worthwhile and want to join in... It could fall under the LMH forum... Adding a finder ID field makes it a bit fun too... I figure we could use a two letter code for each person... I'd also be willing to merge all the results into master factor/nofactor files...

This would be especially well suited for older computers... Even a 1M range would be easily done in a reasonable amount of time on a P90...

All the factors were verified with GP2's verification program...

http://www.mersenneforum.org/showthr...?threadid=1346

Code:

mv@opteron:~/lmh> time ./factorverify factor

real    0m0.114s
user    0m0.106s
sys     0m0.003s

On a test run I even put in a bogus factor to test his program to make sure it works right and it caught it...

Comments?

adpowers · 2004-01-05, 23:53

I'd be interested in doing a few of these to low bit depth. I have an Athlon machine to donate or a P4, which one would work better on these numbers?

You don't get primenet credit for this, do you?

Andrew

Xyzzy · 2004-01-06, 00:24

I imagine an Athlon would fly on this, but probably we should let the real slow computers do this...

You won't get PrimeNet credit for this either...

I'm not even sure this is a legit idea yet!

Xyzzy · 2004-01-06, 00:26

I've updated the above files to 100M-102M... The second batch took 3h50m03s and I found 27,967 factors out of 58,229 exponents tested... I'll stop at 105M so we can let the really slow computers take over...

bayanne · 2004-01-06, 07:56

I have P166 that I could use for this work. It is sitting idle at the moment, whilst I was getting together another batch of normal LMH to run.

Let me know if this would fit the bill!

hbock · 2004-01-06, 12:47

Well, I've also played a little within that range.
It seems that version 21.4.1 of Prime95 is working up to 251M (252M fails), at least with my Athlon 1.2G under Win2K.
Versions 22.12 and newer didn't work higher than 79M.
The syntax for the worktoi.ini is :

AdvancedFactor=start_exponent,end_exponent,b1,b2

as described in the help file under Advanced Usage - Undocumented Features.
B1 and B2 are the start and end levels for the factoring depth (bit). Note that factoring is done for level b1 to b2 inclusive, ie. b1/2 = 28-32 gives a printout :

Mxxxxxxxx no factor from 2^27 to 2^32 bit.

I chose b1=28 since this is the smallest size of a M100000000 factor (2*exp +1). Smaller values are no problem, same factors, same timing. 32 is the minimum for b2.
Here are some timings for the range 100-100.01M:

Code:

28-40 bit  140s
28-50 bit  140s
28-51 bit  205s
28-52 bit  274s
28-53 bit  474s
28-54 bit  859s

51-51 bit  161s
51-52 bit  295s

So 50 bit is the absolute minimum that makes sense. In my opinion 52 bit seems to be more reasonable because it costs less time than an additional factoring from 51 to 52 bit and it's just a factor of 2 for 2 bits more.
If more than one factor is found for an exponent they occur in different orders ( I don't know why) :

Code:

M100000421 has a factor: 5033453790751247
M100000421 has a factor: 8164834373809

M100000687 has a factor: 253801743607
M100000687 has a factor: 25458774900583

M100006927 has a factor: 540037405801
M100006927 has a factor: 3494642057089
M100006927 has a factor: 95406608359
M100006927 has a factor: 294620406943

So, as George mentioned, there is no garanty for correct function (no more supported, not fully tested in that high ranges). Of course we can check the factors found, but we can't see if a small factor is possibly not found (Ok, maybe statistics might help).
Nevertheless it's fun, but it could become a huge own project that needs some coordination. I think most important is the data administration.
Xyzzy, if you (or any other) are able to do that, why not. In the near future there is nothing more to do for slow machines, this is an alternative.
Let's think about the ranges (79-251M ?) and the appropriate factoring depth to avoid dublicating work (contact Will Edgington ?).

GP2 · 2004-01-06, 14:33

This is unrelated to the 100M range, but I've been in contact with Will Edgington recently and there may be a different project available for anyone that's interested in using the "AdvancedFactor" functionality of Prime95 (which seems to be slightly broken in recent versions, at least for mprime).

Namely, covering gaps where bugs in old versions of trial-factoring programs may have left undiscovered factors smaller than the reported bit depth. Is there any interest in this?

Xyzzy · 2004-01-06, 15:49

Quote:

Originally posted by hbock
So 50 bit is the absolute minimum that makes sense.

Yes, that is the conclusion I came to after several timing runs...

From what I can tell, 21.3.1 is the best version to do this with...

I am willing to coordinate this project if you all want me to, or we can just treat it like we do the current LMH stuff...

Is the way I have configured the data files okay?

We could even begin running some ranges up beyond 50 bits if someone wanted to do that, but there are plenty of virgin ranges to work on...

Xyzzy · 2004-01-07, 12:39

Only two people with ultra slow boxes have contacted me, so I guess we can open this up to regular boxes now...

So far 79.3 to 115M are reserved...

Probably reserving work in 5M increments will be best...

Post here with your reservation and desired two letter ID code...

Since the data files are so big, upload them to ftp.mersenneforum.org and email/pm me when you have done so...

There is no need to zip them up beforehand...

Please keep a local copy just in case...

Xyzzy · 2004-01-07, 19:57

Everything up to 125M is now reserved...

Some results have been returned so the data files will be updated in a few hours...

Xyzzy · 2004-01-07, 21:09

http://www.mersenneforum.org/lmh/

There is now a copy of Prime95 in that folder to download as well as a description of how everything works...

2004-01-05, 21:08	#1
Xyzzy Aug 2002 10000111101101₂ Posts	Ranges > 79.3M... Here is 100M to 101M... I have taken the results and made two files... The first file is the factors found and the second file is the nofactor data... The files use the same configuration as George's data files, except I added a third field to the factor file to show who found the factors... For this 1M range I tested 58,043 exponents to 50 bits and I found 28,208 factors in 3h52m35s on a 2.67GHz P4... http://www.mersenneforum.org/lmh/ I don't know if this work is productive or not, but it is kind of fun... I'd be willing to coordinate these searches if others feel it is worthwhile and want to join in... It could fall under the LMH forum... Adding a finder ID field makes it a bit fun too... I figure we could use a two letter code for each person... I'd also be willing to merge all the results into master factor/nofactor files... This would be especially well suited for older computers... Even a 1M range would be easily done in a reasonable amount of time on a P90... All the factors were verified with GP2's verification program... http://www.mersenneforum.org/showthr...?threadid=1346 Code: mv@opteron:~/lmh> time ./factorverify factor real 0m0.114s user 0m0.106s sys 0m0.003s On a test run I even put in a bogus factor to test his program to make sure it works right and it caught it... Comments?

2004-01-06, 12:47	#6
hbock Feb 2003 163 Posts	Well, I've also played a little within that range. It seems that version 21.4.1 of Prime95 is working up to 251M (252M fails), at least with my Athlon 1.2G under Win2K. Versions 22.12 and newer didn't work higher than 79M. The syntax for the worktoi.ini is : AdvancedFactor=start_exponent,end_exponent,b1,b2 as described in the help file under Advanced Usage - Undocumented Features. B1 and B2 are the start and end levels for the factoring depth (bit). Note that factoring is done for level b1 to b2 inclusive, ie. b1/2 = 28-32 gives a printout : Mxxxxxxxx no factor from 2^27 to 2^32 bit. I chose b1=28 since this is the smallest size of a M100000000 factor (2exp +1). Smaller values are no problem, same factors, same timing. 32 is the minimum for b2. Here are some timings for the range 100-100.01M: Code: 28-40 bit 140s 28-50 bit 140s 28-51 bit 205s 28-52 bit 274s 28-53 bit 474s 28-54 bit 859s 51-51 bit 161s 51-52 bit 295s So 50 bit is the absolute minimum that makes sense. In my opinion 52 bit seems to be more reasonable because it costs less time than an additional factoring from 51 to 52 bit and it's just a factor of 2 for 2 bits more. If more than one factor is found for an exponent they occur in different orders ( I don't know why) : Code: M100000421 has a factor: 5033453790751247 M100000421 has a factor: 8164834373809 M100000687 has a factor: 253801743607 M100000687 has a factor: 25458774900583 M100006927 has a factor: 540037405801 M100006927 has a factor: 3494642057089 M100006927 has a factor: 95406608359 M100006927 has a factor: 294620406943 So, as George mentioned, there is no garanty for correct function (no more supported, not fully tested in that high ranges). Of course we can check the factors found, but we can't see if a small factor is possibly not found (Ok, maybe statistics might help). Nevertheless it's fun, but it could become a huge own project that needs some coordination. I think most important is the data administration. Xyzzy, if you (or any other) are able to do that, why not. In the near future there is nothing more to do for slow machines, this is an alternative. Let's think about the ranges (79-251M ?) and the appropriate factoring depth to avoid dublicating work (contact Will Edgington ?). Last fiddled with by hbock on 2004-01-06 at 18:17*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
New ranges at OBD	ET_	Operation Billion Digits	4	2010-07-13 11:34
Old ranges list	Xyzzy	LMH > 100M	54	2005-04-27 02:00
All Available Ranges	hbock	Lone Mersenne Hunters	0	2003-10-27 22:54
Available Ranges below 60 bit	hbock	Lone Mersenne Hunters	13	2003-10-24 23:40
Ranges for P4s	geoff	Lone Mersenne Hunters	1	2003-09-05 18:13

2004-01-05, 23:53	#2
adpowers Sep 2002 2⁴·5 Posts	I'd be interested in doing a few of these to low bit depth. I have an Athlon machine to donate or a P4, which one would work better on these numbers? You don't get primenet credit for this, do you? Andrew

2004-01-06, 00:24	#3
Xyzzy Aug 2002 3²·5·193 Posts	I imagine an Athlon would fly on this, but probably we should let the real slow computers do this... You won't get PrimeNet credit for this either... I'm not even sure this is a legit idea yet!

2004-01-06, 00:26	#4
Xyzzy Aug 2002 8685₁₀ Posts	I've updated the above files to 100M-102M... The second batch took 3h50m03s and I found 27,967 factors out of 58,229 exponents tested... I'll stop at 105M so we can let the really slow computers take over...

2004-01-06, 07:56	#5
bayanne "Tony Gott" Aug 2002 Yell, Shetland, UK 3×5³ Posts	I have P166 that I could use for this work. It is sitting idle at the moment, whilst I was getting together another batch of normal LMH to run. Let me know if this would fit the bill!

2004-01-06, 14:33	#7
GP2 Sep 2003 2·5·7·37 Posts	This is unrelated to the 100M range, but I've been in contact with Will Edgington recently and there may be a different project available for anyone that's interested in using the "AdvancedFactor" functionality of Prime95 (which seems to be slightly broken in recent versions, at least for mprime). Namely, covering gaps where bugs in old versions of trial-factoring programs may have left undiscovered factors smaller than the reported bit depth. Is there any interest in this?

2004-01-07, 12:39	#9
Xyzzy Aug 2002 3²·5·193 Posts	Only two people with ultra slow boxes have contacted me, so I guess we can open this up to regular boxes now... So far 79.3 to 115M are reserved... Probably reserving work in 5M increments will be best... Post here with your reservation and desired two letter ID code... Since the data files are so big, upload them to ftp.mersenneforum.org and email/pm me when you have done so... There is no need to zip them up beforehand... Please keep a local copy just in case...

2004-01-07, 19:57	#10
Xyzzy Aug 2002 3²·5·193 Posts	Everything up to 125M is now reserved... Some results have been returned so the data files will be updated in a few hours...

2004-01-07, 21:09	#11
Xyzzy Aug 2002 3²×5×193 Posts	http://www.mersenneforum.org/lmh/ There is now a copy of Prime95 in that folder to download as well as a description of how everything works...