P-1 factoring anyone?

[petrw1]

Originally Posted by James Heinrich

So, at roughly 4.15GHz-days per P-1, we need (4.15 * 180 * 1.065) = 796GHz-days/day of P-1 power to keep up with LL.

And if I put my entire (albeit meagre) team of P-1 capable PCs (3 Quads and 6 Duals and 1 P4=25 cores) I estimate I could complete only about 10 or 12 per day or 5 - 6%.

Mind you, all 4 cores of a Quad on P-1 could be quite taxing.

[petrw1]

Well with about 15,500 LL tests now P1'd and ready in the 53M range that's about a 3 month head start.

[Brain]

I have an idea how we could motivate users for doing more P-1 work: Create a PrimeNet page "Top1000 factors found in the last 365 days" - ordered by bitlength.

I'd love to know where my 104 bit find would land. Most (all?) factors will be found by P-1 work. Thus more P-1 workers....

Rang | User | Exponent | Factor | Bitlength | Digits | found on | found by (worktype) | B1 | B2

One problem left: custom "giant" P-1 will always land in the Top 10...

[James Heinrich]

Originally Posted by Brain

Create a PrimeNet page "Top1000 factors found in the last 365 days" - ordered by bitlength

Brilliant idea! Can we have it, please?

Would composite factors (as found and reported by Prime95 in results.txt) be allowed on the list, or just prime factors? My own top-10 list ranges from 108 to 158 bits, but the top 7 of those are composite (largest prime is 128.6 bits).

[petrw1]

Like this one recently:

rx7350               RMCpc11          48781163  F-PM1    Apr  5 2011  4:51PM   1.3   1.4223  8075101174426524285127515871546822802477167420687

8 075101 174426 524285 127515 871546 822802 477167 420687 = 10385 840363 685189 198073 x 777 510619 425817 031240 450119

[James Heinrich]

Is it possible to get a list of recent results that has an entire hour's worth of data, not just the 1000 most recent results (which might only be 5 minutes worth) ?

[Christenson]

Originally Posted by ET_

Give me a pointer to a (somewhat easy) P-1 implementation, and I'll try to recode it for a GPU...

(as long as George is not working on it already)

Luigi

That would pull in the high-end GPU in my future by a few months....
without it, I think I'm turning in 1 P-1 per day, maybe three per day next week when more cores get going on that type of work.

[petrw1]

Originally Posted by James Heinrich

Is it possible to get a list of recent results that has an entire hour's worth of data, not just the 1000 most recent results (which might only be 5 minutes worth) ?

I keep thinking (ever since 63 bits) that as soon as the project gets one bit deeper in TF it will slow down enough to cover an hour but then something new comes along (i.e. GPU)

[aketilander]

There seems to be an upper limit for the P-1 test in Windows64 Prime95, v 26.5, build 5
The largest P-1 test done using Prime95 seems to be for exp 595999993

If I got it right there are presently available 13 larger exps ready (sufficiently TFed) for P-1
(670616621, 750000073, 872000011, 920000057, 930000107, 940000007, 950000063, 960000011, 970000021, 990000029, 999960011, 999999043, 999999751 and a couple of others (6) already reserved mostly by persons doing LL (4), P-1 (1), unclear (1))

I have tried to run a P-1 for some of these, but I get the following error message:

Cannot initialize FFT code, errcode=1002

So there seems to be an upper limit for P-1 using Prime95.
Is that correct?
Does anyone know the exact limit?
Are there similar upper limits for TF, LL, ECM, the program as such?

[xilman]

Originally Posted by James Heinrich

Brilliant idea! Can we have it, please?

Would composite factors (as found and reported by Prime95 in results.txt) be allowed on the list, or just prime factors? My own top-10 list ranges from 108 to 158 bits, but the top 7 of those are composite (largest prime is 128.6 bits).

How about this modified suggestion?

The larger Mersenne exponents are likely to produce larger P-1 factors because of the necessary inclusion of that exponent in the factorization of P-1. To compensate, rate the top 1000 by the size of P-1 after division by all powers of the Mersenne exponent. Note that this proposal scales down composite P-1 factors also.

The same observation about the Mersenne exponent also applies to factors found by trial division...


Paul

[cheesehead]

Originally Posted by aketilander

There seems to be an upper limit for the P-1 test in Windows64 Prime95, v 26.5, build 5
The largest P-1 test done using Prime95 seems to be for exp 595999993
< snip >
Does anyone know the exact limit?

Module gwnum.h in the Windows source has the definition

#define MAX_PRIME_SSE2 596000000L /* SSE2 bit limit */

595999993 is probably the largest prime < 596000000

Are there similar upper limits for TF, LL, ECM, the program as such?

LL, ECM and P-1 all use FFTs, and 596000000 is the maximum exponent the FFTs can now handle AFAIK.

Perhaps the higher exponents were processed with a Mac version of the software rather than a Windows version.