How about P-1 for srb5?

Rincewind · 2011-03-19, 18:35

Hi,

how are the opinons about P-1 for this project?
I think it could be nice because it is posible to 'pre-test' n-ranges more directly.

Joe O · 2011-03-28, 00:33

Have you run any timings? Start with B1 ~ 1E8 and B2 ~ 200E8. Check to see that the time for Stage2 is the same as the time for Stage1. Adjust B2 until it is. Run 40 k n pairs. If no factors found run 40 more. If no factors found after 80 k n pairs, try 20 more. If still no factors after 100 pairs, increase B1 and B2 and try again. Once you have factors and can determine the number of runs to find a factor you can determine a probable time to find a factor. Compare this time to the time for an LLR run. Please report your finding. It is important that all the runs be done on the same machine.
Thank you.

NeoGen · 2011-03-29, 12:55

I would be willing to help if it's something simple, and it appears to be simple to a basic user like me, I can understand the step by step procedure. But in terms of exactly what/how it is to do and which application(s) to use I have no idea...

I don't even know how to set the SR5 sieve app to lower cpu priority (instead of normal), I just go and do it on the windows task manager after having fired it up.

em99010pepe · 2011-03-29, 18:44

Quote:

Originally Posted by NeoGen

I don't even know how to set the SR5 sieve app to lower cpu priority (instead of normal), I just go and do it on the windows task manager after having fired it up.

Create batch file with:

start /low /min sr5sieve.exe

Rincewind · 2011-03-29, 19:51

Quote:

Originally Posted by Joe O

Have you run any timings? Start with B1 ~ 1E8 and B2 ~ 200E8. Check to see that the time for Stage2 is the same as the time for Stage1. Adjust B2 until it is. Run 40 k n pairs. If no factors found run 40 more. If no factors found after 80 k n pairs, try 20 more. If still no factors after 100 pairs, increase B1 and B2 and try again. Once you have factors and can determine the number of runs to find a factor you can determine a probable time to find a factor. Compare this time to the time for an LLR run. Please report your finding. It is important that all the runs be done on the same machine.
Thank you.

OK, that's kind of what I thought. So I'll do some timings (I don't know when I'm able to do this, but it should be this week).
Just a short question: Is there a reason for this B-Values? And why should both stages take the same time?
I ask because if I use mprime/Prime95 for P-1 with 'pfactor' in the worktodo.txt it uses B2-Values which causes stage 1 to be roughly 5 times longer than stage 2.

@NeoGen: After I have the timing and it makes sense, I can send you a howto.

Joe O · 2011-03-31, 20:18

The B1 value was chosen to get factors of the size above those that are being returned by sieving. It will also get smaller factors. mprime/Prime95 chooses B2 to fit in the memory you provide it. It's also geared toward the mass user and toward Mersenne numbers. I'm hoping you have a large amount of memory and are willing to use it for P-1. The equal time recommendation comes from the GMP-ECM documentation and has served me well throughout the years. The program you use will also be a factor in the choice of B2 values. GMP-ECM is usually faster in stage2 than mprime/Prime95. In fact some people, myself included, run stage1 in Prime95 and stage2 in GMP-ECM. It's more work, so sometimes I just use one or the other.

Rincewind · 2011-04-01, 06:10

Thank you for the answers. I think I can start the tests tomorrow.
I'll use mprime for this tests.

Rincewind · 2011-05-02, 16:57

Sorry that my answer took so long.
I tested some pairs and realised that this makes no sense atm.
Usin Joe O's values has the result that one P-1-test need much more time than a prime test.
Reducing the B-values decreased the time, but I found no factor after several hours.
So I think sieving and prping is fine for now.

2011-03-19, 18:35	#1
Rincewind Oct 2006 103 Posts	How about P-1 for srb5? Hi, how are the opinons about P-1 for this project? I think it could be nice because it is posible to 'pre-test' n-ranges more directly.

2011-03-28, 00:33	#2
Joe O Aug 2002 20D₁₆ Posts	Have you run any timings? Start with B1 ~ 1E8 and B2 ~ 200E8. Check to see that the time for Stage2 is the same as the time for Stage1. Adjust B2 until it is. Run 40 k n pairs. If no factors found run 40 more. If no factors found after 80 k n pairs, try 20 more. If still no factors after 100 pairs, increase B1 and B2 and try again. Once you have factors and can determine the number of runs to find a factor you can determine a probable time to find a factor. Compare this time to the time for an LLR run. Please report your finding. It is important that all the runs be done on the same machine. Thank you.

2011-03-29, 12:55	#3
NeoGen Dec 2005 110110₂ Posts	I would be willing to help if it's something simple, and it appears to be simple to a basic user like me, I can understand the step by step procedure. But in terms of exactly what/how it is to do and which application(s) to use I have no idea... I don't even know how to set the SR5 sieve app to lower cpu priority (instead of normal), I just go and do it on the windows task manager after having fired it up.

2011-03-31, 20:18	#6
Joe O Aug 2002 1015₈ Posts	The B1 value was chosen to get factors of the size above those that are being returned by sieving. It will also get smaller factors. mprime/Prime95 chooses B2 to fit in the memory you provide it. It's also geared toward the mass user and toward Mersenne numbers. I'm hoping you have a large amount of memory and are willing to use it for P-1. The equal time recommendation comes from the GMP-ECM documentation and has served me well throughout the years. The program you use will also be a factor in the choice of B2 values. GMP-ECM is usually faster in stage2 than mprime/Prime95. In fact some people, myself included, run stage1 in Prime95 and stage2 in GMP-ECM. It's more work, so sometimes I just use one or the other.

2011-04-01, 06:10	#7
Rincewind Oct 2006 103 Posts	Thank you for the answers. I think I can start the tests tomorrow. I'll use mprime for this tests.

2011-05-02, 16:57	#8
Rincewind Oct 2006 103 Posts	Sorry that my answer took so long. I tested some pairs and realised that this makes no sense atm. Usin Joe O's values has the result that one P-1-test need much more time than a prime test. Reducing the B-values decreased the time, but I found no factor after several hours. So I think sieving and prping is fine for now.