P-1 Memory...

Xyzzy · 2003-11-27, 07:12

Some weird numbers... Any thoughts? (These are for a ~4M exponent!)

garo · 2003-11-27, 19:33

Umm... what's weird about this?

Xyzzy · 2003-11-27, 20:33

The slight increase in the chance you have of finding a factor doesn't look like it is worth the trouble of assigning extra memory... Of course I knew that giving it more memory would not make it any faster, but up until now I never would have considered using such a low amount of memory on a P-1 test...

In other words, work here in Marin's Mersenne-aries needn't be limited to people who have gobs of memory...

GP2 · 2003-11-27, 22:26

But that's for a small (~4M) exponent.

What do you get for, say, 11M exponents?

garo · 2003-11-27, 23:04

Well, if you look at the bounds calculation code, you will find that Prime95 is interested in maximizing the throughput of the project and not necessarily the individual machine. The bounds are calculated such that

(Time for 2 LL tests * chance of finding factor) - Time for P-1 test

is maximized

Note that GIMPS actually uses a multiple other than2 but I don't remember it exactly.

So, in effect it may be more beneficial for the project but your machine may end up finding factors at a slower rate.

nfortino · 2003-11-27, 23:38

This is the expected data trend from the command line:
Pfactor=exponent,how_far_factored,has_been_LL_tested_once

As garo has said, the program optimizes the bounds so that the processing time is most beneficial to the project. When you allocate more memory, it allows the second stage to run with larger bounds, but also allows the smaller bounds to run faster. This inturn allows deeper bounds with the same amount of time, explaining the slight increase in the B2 bound. However, since you goal is to find a factor, you should use a different command line.

From readme.txt: “The P-1 choice lets you factor Mersenne numbers using the P-1 method of factoring. There is presently no web site which tells you how much P-1 factoring has already been done on exponents. You can also edit the worktodo.ini file directly. For example:
Pminus1=751001,1000000,0,0,0
The first value is the exponent. The second value is bound #1. The third value is bound #2. The fourth value is 0 for 2^N-1 factoring, 1 for 2^N+1 factoring. The fifth value is no longer used.”

For some reason, this is not mentioned in undoc. I have no idea how to calculate the proper bounds from a set memory allocation.

Regards,
Nick

crash893 · 2004-01-06, 22:02

the time seems to be pretty stable

but by my ( very questiobable) perdictions

you would need something like 200 gigs of ram to get near 100% chance of finding a factor every time.

koekie · 2004-03-12, 13:34

Quote:

Originally Posted by crash893

you would need something like 200 gigs of ram to get near 100% chance of finding a factor every time.

Maybe we can ask the US goverment if we can borrow some of this http://www.techworld.com/news/index....ws&NewsID=1176 ram

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
"Hybrid Memory Cube" offers 1 Tb/s memory bandwith at just 1.4 mW/Gb/s	ixfd64	Hardware	4	2011-12-14 21:24
Memory available to P-1	lycorn	Software	23	2010-05-09 22:15
How much memory do you have?	fivemack	Factoring	17	2009-08-23 16:00
memory	victorvicentim	Information & Answers	5	2009-08-19 12:34
How much memory?	Unregistered	Software	29	2004-09-04 19:35

2003-11-27, 07:12	#1
Xyzzy "Mike" Aug 2002 2²·5·397 Posts	P-1 Memory... Some weird numbers... Any thoughts? (These are for a ~4M exponent!) Attached Thumbnails

2003-11-27, 19:33	#2
garo Aug 2002 Termonfeckin, IE 3×919 Posts	Umm... what's weird about this?

2003-11-27, 20:33	#3
Xyzzy "Mike" Aug 2002 2²×5×397 Posts	The slight increase in the chance you have of finding a factor doesn't look like it is worth the trouble of assigning extra memory... Of course I knew that giving it more memory would not make it any faster, but up until now I never would have considered using such a low amount of memory on a P-1 test... In other words, work here in Marin's Mersenne-aries needn't be limited to people who have gobs of memory...

2003-11-27, 22:26	#4
GP2 Sep 2003 2583₁₀ Posts	But that's for a small (~4M) exponent. What do you get for, say, 11M exponents?

2003-11-27, 23:04	#5
garo Aug 2002 Termonfeckin, IE 3·919 Posts	Well, if you look at the bounds calculation code, you will find that Prime95 is interested in maximizing the throughput of the project and not necessarily the individual machine. The bounds are calculated such that (Time for 2 LL tests * chance of finding factor) - Time for P-1 test is maximized Note that GIMPS actually uses a multiple other than2 but I don't remember it exactly. So, in effect it may be more beneficial for the project but your machine may end up finding factors at a slower rate.

2003-11-27, 23:38	#6
nfortino Nov 2003 3·5·11 Posts	This is the expected data trend from the command line: Pfactor=exponent,how_far_factored,has_been_LL_tested_once As garo has said, the program optimizes the bounds so that the processing time is most beneficial to the project. When you allocate more memory, it allows the second stage to run with larger bounds, but also allows the smaller bounds to run faster. This inturn allows deeper bounds with the same amount of time, explaining the slight increase in the B2 bound. However, since you goal is to find a factor, you should use a different command line. From readme.txt: “The P-1 choice lets you factor Mersenne numbers using the P-1 method of factoring. There is presently no web site which tells you how much P-1 factoring has already been done on exponents. You can also edit the worktodo.ini file directly. For example: Pminus1=751001,1000000,0,0,0 The first value is the exponent. The second value is bound #1. The third value is bound #2. The fourth value is 0 for 2^N-1 factoring, 1 for 2^N+1 factoring. The fifth value is no longer used.” For some reason, this is not mentioned in undoc. I have no idea how to calculate the proper bounds from a set memory allocation. Regards, Nick

2004-01-06, 22:02	#7
crash893 Sep 2002 2³·37 Posts	the time seems to be pretty stable but by my ( very questiobable) perdictions you would need something like 200 gigs of ram to get near 100% chance of finding a factor every time.