P-1 factoring anyone?
 

I'll be watching over the next few months how many people sign up for P-1 factoring. We have tons of people doing LL and TF, we need more people signing up for P-1.

In the new scheme of things, if P-1 doesn't get done then when an exponent is assigned to an LL tester he must do the P-1, last 2 levels of TF, and then the LL test.

If not enough people sign up for P-1, I'll have to tweak "do what makes the most sense" to assign some P-1 factoring. We have plenty of time to work this out as the leading edge of exponents that need P-1 factoring and are holding up TF is above 48 million. It will take a while before the leading edge of LL gets that high.

Why?
[QUOTE]... last 2 levels of TF ...[/QUOTE]
Doesn't that happen anymore through your standard TF assignments?

The summary also shows almost 34,000 P-1 available from 44-45M??

What is the preferred hardware for P-1? I have my new Q9550, 64 Bit, 4GB RAM. So it would certainly be a fast factoring maching but then it would also be near the upper end in LL performance too.

i'll take it up, once i find out the cause of prime95 not assigning p-1 work to one of my workers

I would like to have the 64 bits version of Prime95 25.8.2 (or later) first. Otherwise I will be wasting 6 GB of RAM. I seem to insist on overworking you in my many communications on different channels ;-)

Jacob

[QUOTE=petrw1;152316]Why? Doesn't that happen anymore through your standard TF assignments?

The summary also shows almost 34,000 P-1 available from 44-45M??

What is the preferred hardware for P-1? I have my new Q9550, 64 Bit, 4GB RAM. So it would certainly be a fast factoring maching but then it would also be near the upper end in LL performance too.[/QUOTE]

An example: Take an exponent near 51 million. It needs to be factored to 2^69. Tf assignments will take it to 2^67. Then the server wants to assign it to P-1.

If we have plenty of P-1 volunteers, then the P-1 assignment happens in a timely manner, and the exponent is assigned for the final two bits of trial factoring.

If we don't have many P-1 volunteers, when the leading edge of LL testers reaches 51 million the exponent is assigned for LL testing. The LL tester must do the P-1, 2 levels of TF, and then the LL test.

I'm intentionally assigning P-1 exponents that are holding up TF. The 34,000 exponents in 44-45M were TFed by the v4 server which did not stop 2 bit levels shy of completing TF.

Because of memory bandwidth limitations, consensus seems to be a Q9550 should be doing 3 workers of LL/P-1 and 1 worker of TF. If you had one of your 3 LL/P-1 workers dedicated entirely to P-1 then you would be doing more than your fair share.

[QUOTE=Prime95;152358]Because of memory bandwidth limitations, consensus seems to be a Q9550 should be doing 3 workers of LL/P-1 and 1 worker of TF. If you had one of your 3 LL/P-1 workers dedicated entirely to P-1 then you would be doing more than your fair share.[/QUOTE]

As S485122 suggested do I not need the 64 Bit Version of Prime95 to go along with my 64 Bit Vista to get the P-1 thruput benefit?

Out of curiosity, are you trial factoring residue classes that are more likely to be P-1-smooth to lower bound than the rest?

Alex

[QUOTE=petrw1;152398]As S485122 suggested do I not need the 64 Bit Version of Prime95 to go along with my 64 Bit Vista to get the P-1 thruput benefit?[/QUOTE]No, but you will not be able to use a lot of memory, which can be a plus since you would still have about 2GB free for other usage...

Jacob

[QUOTE=petrw1;152398]As S485122 suggested do I not need the 64 Bit Version of Prime95 to go along with my 64 Bit Vista to get the P-1 thruput benefit?[/QUOTE]

There are diminishing returns for extra memory. I suspect anything over 500MB yields very small benefits - but I do not have any benchmarks to back that up.

[QUOTE=akruppa;152402]Out of curiosity, are you trial factoring residue classes that are more likely to be P-1-smooth to lower bound than the rest?
[/QUOTE]

I did not get around to that bit of fine-tuning.

[QUOTE=Prime95;152423]There are diminishing returns for extra memory. I suspect anything over 500MB yields very small benefits - but I do not have any benchmarks to back that up.[/QUOTE]

Okay...but isn't it documented that 64Bit OS along with 64Bit Prime95 are faster for factoring? Or is that just for TF and not for P-1?