Prime95 P-1/Pfactor using to find factors on CRUS bases - Page 2

pepi37 · 2017-01-20, 16:53

Quote:

Originally Posted by Batalov

Can you run 45 P-1 tests (which will remove 1 candidate -- based on your own 2.2% value) faster than 1 LLR test?

If fact I can: I can run 90 tests with B1=20000 and B2=700000 for time of one LLR test on 4*155^930xxx+1

One P1 test is done about 345 seconds in and one LLR test is done in 31300 second

VBCurtis · 2017-01-20, 17:22

If that's true, that you can run 90 tests at B1=20000 in the time of one LLR test, then a P-1 pass on your entire input file will save you overall about half the time spent on P-1. If you increase B1 to 30000 (with B2 perhaps 1e6) for your one P-1 pass, you should find a success rate better than you had for 20000, and might save even more time.

pepi37 · 2017-01-20, 17:35

Quote:

Originally Posted by VBCurtis

If that's true, that you can run 90 tests at B1=20000 in the time of one LLR test, then a P-1 pass on your entire input file will save you overall about half the time spent on P-1. If you increase B1 to 30000 (with B2 perhaps 1e6) for your one P-1 pass, you should find a success rate better than you had for 20000, and might save even more time.

I am now experimenting with B1 70000 and B2 700000. I have small number of processed candidates, but it looks like chance for removing is little over 2.2% ( now is about 2.6%)
And yes, I will run P-1 on whole input file by time when candidate exceed 2M digits. It doesnot really matter if I lost few hours or even day in that, I will also removed some candidates. And since P-1 is trusted method, and all factors found by P-1 are really factors, then I am happy with this method :)

P.S
I noticed not only in this forum but also in many other forums that are members of very distrustful of facts that can be easily proved or disproved in a few minutes :) (thinking out loud)
If I lie, I lie to myself not to you, since you will not use P-1 ,so you will not loose time at all :)

pepi37 · 2017-01-21, 15:59

Last sequence finished.
From 4946 candidates removed 160. so removal rate is 3.2 %.
Values was: B1=70000 B2=700000.
Time on P-1 for one candidate is 137 seconds. ( 2 cores)
This sequence has algebraic factorization , so maybe that is reason of increased removal rate.

henryzz · 2017-01-21, 17:42

I just did some timings for 4*155^930000+1
P-1 B1=70k, B2=700k: 351 seconds
PRP test: 12600 seconds
The P-1 test takes 2.8% of the time of the prp test. We need a factor rate of >2.8% after sieving.
You claim to have 3.2% which sounds good. However, if there are algebraic factors then the rate will be doubled(assuming two factors). Your rate is probably actually 1.6% per factor.

There are potentially applications for P-1 at CRUS though. Testing on 1597*6^n-1 has reached n=5M. At this level:
P-1 B1=70k, B2=700k: 700 seconds
PRP test: 105000 seconds
The P-1 test takes 0.67% of the time of the prp test. We need a factor rate of >0.67% after sieving. This should hopefully be possible even with deeper sieving.

Batalov · 2017-01-21, 18:17

Quote:

Originally Posted by henryzz

I just did some timings for 4*155^930000+1

Kozma Prutkov once wrote:
"Throwing pebbles into the water, look at the ripples they form on the surface, otherwise, such occupation becomes an idle pastime”

Even if you were just throwing pebbles, why not take a realistic example, and not an obviously composite number?

Additional question for bonus points: now that have you done this pebble, are you any closer to proving or disproving the claim that started this vanity thread: "Very light P-1 factoring would further significantly reduce the likelihood of an unfactored, large exponent term with algebraic factors." It certainly did in this case, right? You lightly factored this candidate, and the doors of perception were suddenly cleansed ... and you saw this number for what it was - a composite.

pepi37 · 2017-01-21, 19:24

Quote:

Originally Posted by henryzz

PRP test: 12600 seconds

Tell me what CPU has that timing ( how many cores)?

Quote:

Originally Posted by henryzz

Your rate is probably actually 1.6% per factor.

If I take number of candidates I test with P-1 and number of found factor, and that give me 3.2 % then it is 3.2% It cannot be 1.6 %
And also not every range I take has algebraic factors or not every range was tested with same B1 and B2 values.

pepi37 · 2017-01-21, 19:29

Quote:

Originally Posted by Batalov

Kozma Prutkov once wrote:
"Throwing pebbles into the water, look at the ripples they form on the surface, otherwise, such occupation becomes an idle pastime”

Kuzma Prutkov also say this:

If you want to be happy, be so

No harm was done with P-1 factoring, and if that make me happy , then I will be happy :)
Not every one here has computing power, like you have, dont forget that fact

henryzz · 2017-01-21, 19:54

Quote:

Originally Posted by pepi37

Tell me what CPU has that timing ( how many cores)?

If I take number of candidates I test with P-1 and number of found factor, and that give me 3.2 % then it is 3.2% It cannot be 1.6 %
And also not every range I take has algebraic factors or not every range was tested with same B1 and B2 values.

My 1.6% assertion was based upon every number tested having two algebraic factors.
This was on one core of a skylake 6700k at 4GHz.

pepi37 · 2017-01-21, 19:59

Quote:

Originally Posted by henryzz

This was on one core of a skylake 6700k at 4GHz.

It is time to throw out my "old "Intels....

henryzz · 2017-01-25, 16:20

Quote:

Originally Posted by pepi37

It is time to throw out my "old "Intels....

I recently upgraded from a Q6600. 5x faster per core.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Find factors for non base2 candidates	pepi37	GMP-ECM	2	2017-03-07 20:13
Fails to find very small factors.	Mr. P-1	FactorDB	6	2013-03-22 02:30
Best Way to find large factors	mahnouman	Information & Answers	19	2013-02-22 06:11
Generalizing algebraic factors on Riesel bases	gd_barnes	Conjectures 'R Us	31	2010-04-06 02:04
How to find factors I found with TF?	edorajh	PrimeNet	3	2004-10-01 19:16

2017-01-20, 17:22	#13
VBCurtis "Curtis" Feb 2005 Riverside, CA 2⁸·19 Posts	If that's true, that you can run 90 tests at B1=20000 in the time of one LLR test, then a P-1 pass on your entire input file will save you overall about half the time spent on P-1. If you increase B1 to 30000 (with B2 perhaps 1e6) for your one P-1 pass, you should find a success rate better than you had for 20000, and might save even more time.

2017-01-21, 15:59	#15
pepi37 Dec 2011 After milion nines:) 5A8₁₆ Posts	Last sequence finished. From 4946 candidates removed 160. so removal rate is 3.2 %. Values was: B1=70000 B2=700000. Time on P-1 for one candidate is 137 seconds. ( 2 cores) This sequence has algebraic factorization , so maybe that is reason of increased removal rate.

2017-01-21, 17:42	#16
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 5881₁₀ Posts	I just did some timings for 4155^930000+1 P-1 B1=70k, B2=700k: 351 seconds PRP test: 12600 seconds The P-1 test takes 2.8% of the time of the prp test. We need a factor rate of >2.8% after sieving. You claim to have 3.2% which sounds good. However, if there are algebraic factors then the rate will be doubled(assuming two factors). Your rate is probably actually 1.6% per factor. There are potentially applications for P-1 at CRUS though. Testing on 15976^n-1 has reached n=5M. At this level: P-1 B1=70k, B2=700k: 700 seconds PRP test: 105000 seconds The P-1 test takes 0.67% of the time of the prp test. We need a factor rate of >0.67% after sieving. This should hopefully be possible even with deeper sieving.