Question on going deep and using cores

[Uncwilly]

Originally Posted by Uncwilly

If you are in a rush to get a particular exponent done (and are willing to possibly waste some effort), you can have multiple cores work on the exponent. Here is how I would structure a worktodo.txt file

I assumed that one wanted to do work on a single expo and only TF. If we were to add P-1 into the mix the structure would be different.

I had it set to try and find a factor (supposing that there may be one) as soon as possible. Doing the 1/62 and 1/63 chance at the same time, then starting the next most likely etc.

This was how I would do it, YMMV.

[Mr. P-1]

Originally Posted by Uncwilly

Bit depth and P-1 are oranges and limes.

Bit depth refers to how far an exponent has been trial factored.
P-1 has bounds, B1 & B2.

Specifically trial factoring and P-1 search different factor spaces. TF searches the space of all potential factors up to the specified bit length, while P-1 searches the space of all potential factors "smooth" to the specified bounds. ("Smooth" is explained here.)

These two spaces overlap; some factors could be found using either method. But we do P-1 because it is capable of finding some factors well beyond the reach of TF. So far, I have found two prime factors >100 bits - by no means a record for GIMPS - although most are just a few bits beyond the amount of TF done.

Normally you want to factor an exponent to a predetermined bit depth (based up the size of the exponent) or until a factor is found, then do P-1.
(George has changed this a little, factor to goal_bit-2, P-1, continue to factor to goal_bit.)

The rationale for this is that the expected benefit of doing P-1 exceeds that of doing the last two rounds of TF. Those rounds are still worth doing if P-1 fails.

[TheJudger]

Originally Posted by Mr. P-1

These two spaces overlap; some factors could be found using either method. But we do P-1 because it is capable of finding some factors well beyond the reach of TF. So far, I have found two prime factors >100 bits - by no means a record for GIMPS - although most are just a few bits beyond the amount of TF done.

Are "highscores" (biggest factors) for P-1 factors available for GIMPS?

[Mr. P-1]

Originally Posted by TheJudger

Are "highscores" (biggest factors) for P-1 factors available for GIMPS?

Not really. As far as I can see the factor database is no longer published in its entirety, nor is there any way to obtain a list of the largest factors.

Even if you had such a list, I don't think the method used to obtain them was recorded by V4. V5 may do so, but this information is not exposed on the website.

Any factor larger than the TF limit is likely to have been found by P-1 if the exponent is greater than 10,000. Below that level, factors may have also been obtained by ECM. Intractable Mersennes below about 1,000 bits have been factored using the Special Number Field Sieve method. That very specialized and computation-intensive method is beyond the scope of GIMPS

One way to gain confidence that a particular factor was obtained by P-1 is to determine the limits to which it is smooth. If those limits are within the normal bounds for a P-1 test on an exponent of that size, then it was most likely found this way, else not. This is not a definitive test, though. A factor which could have been found with P-1 may in fact have been found by some other method, while the P-1 method does occasionally throw up non-smooth factors.

Here is a list of the largest factors ever found using the P-1 method. None of these are factors of Mersenne numbers, and I don't believe that any were found by GIMPS.

See also this thread.

[Mr. P-1]

Originally Posted by Mr. P-1

See also this thread.

I would add, in relation to that thread, that factor length is only significant for prime factors. Every factor found has a huge almost certainly composite cofactor.

[Mr. P-1]

Originally Posted by Mr. P-1

Any factor larger than the TF limit is likely to have been found by P-1 if the exponent is greater than 10,000. Below that level, factors may have also been obtained by ECM.

In fact, the server is handing out exponents in the 2-3M range for ECM factoring.

[petrw1]

Originally Posted by petrw1

First point: I have NOT overclocked.

With only one core doing a LL at 47.7M it was about 0.058 per iteration.
With 3 cores doing LL at that level and 1 core doing P-1 I was averaging about 0.066 seconds while P-1 was in Phase 1 and about 0.068 when P-1 was in phase 2.

Related to that, my P-1 core is working in the 50M range using 1024M RAM. It is taking about 21 hours for Phase 1 and about 36 hours for Phase 2 for a total of 57 hours. The Test... Status menu keeps telling me they entire P-1 should finish in 50 hours.

If these numbers are out of line then I need to check into it.

According to the Benchmarks page I am at the right levels for a single core of a Q9550 NOT OC'd. Though I am not sure if I am losing too much when all 4 go.

FFT   1-core   All-4   Inc.
1536  0.030   0.038  22%  (Two cores with these)
2048  0.041   0.051  24%
P-1   0.076   0.121  60% (ouch)

P-1 is Phase 2 of a 50M exponent with 1200M RAM

[miguel_chek]

I am a new member and just built custom i940. NOT an expert in PCs or Math but know enough to be dangerous ( mostly to myself..lol) Anyway, I was struck curious at the P-1 completion times for stage 1 & 2 so I thought I would post my first run (started 0400 11Jan) using 1 worker, 1cpu, 2Gb mem. Stage 1= 11.37 hrs Stage 2= 16.56 hrs. (calculated)

Also I really want to adjust my workers to the optimum so I invite advice on my present setup:(all runs 24 hours)

Cpu1= P-1 L Cpu2= default(TF) Cpu3= TF LMH Cpu4= default(TF)

1st 45 hours stats TF694215819.1427170Overall1627267019.1427170

avg loads: C0-80%/80% C1-80%/20% C2-60%/40% C3-60%/40%

cheers

[petrw1]

Originally Posted by miguel_chek

IAnyway, I was struck curious at the P-1 completion times for stage 1 & 2 so I thought I would post my first run (started 0400 11Jan) using 1 worker, 1cpu, 2Gb mem. Stage 1= 11.37 hrs Stage 2= 16.56 hrs. (calculated)

WOW....
Was that P-1 time with all 4 cores busy?
I could match that time if the P-1 core was running alone but it is 50% faster than my Q9550 P-1 time running P-1 on 1 core and LL and the other 3.


I'm confused by the huge numbers you have for your 45 hours stats. Your numbers are MUCH!!! higher than the combined total thruput of the entire project over the last 12 years.

[miguel_chek]

here is my 24/7 settings:
Gimp: Core0: P-1, C1: TF, C2: TF(LMH), C3: TF

Rig: i940 2.9Ghz, ASUS P6T DLX, 3GB Corsair ddr3 1066(2 allocated to GIMPS),
2 'raptors-Raid 0, 5Gb DSL.

Account created2009-01-10 13:04 UTC

TypeRankofGHz daysCountTF586218127.4329910P-1107616254.12221Overall1576270031.5551911

[petrw1]

Originally Posted by miguel_chek

here is my 24/7 settings:
Gimp: Core0: P-1, C1: TF, C2: TF(LMH), C3: TF

I wonder if my P-1 would run faster if I did TF instead of LL on the other cores?