20221022, 00:39  #1 
Aug 2022
3 Posts 
What happens to unsuccessful P1 trials?
A P1 factoring trial was performed on a certain Mersenne number with B1 and B2, and no factor was found.
1. Can the calculation be repeated with another pair B3,B4 where B3>B2? 2. Will it be worth to repeat the calculation on a machine with more memory (e.g. in 5 years when average user has more memory)? 
20221022, 00:49  #2  
If I May
"Chris Halsall"
Sep 2002
Barbados
11,087 Posts 
Quote:
P.S. Very importantly, ignorant != stupid. P.P.S. Ignorant simply means you don't understand something fully. Stupid is what I am quite regularly. 

20221022, 01:50  #3  
Mar 2019
147_{16} Posts 
Quote:
As you might say, "this is a serious question". 

20221022, 01:53  #4  
Jun 2003
2·2,719 Posts 
Quote:
Depends. If the goal is to find a factor and save PRP/LL test, it would be too late. If the goal is to find a factor, period, then it might be worth it. There would be other competing options like ECM, P+1, additional TF, but typically P1 would still offer better bang for buck (as long as it it not "too close" to previous P1 bounds, nor "crazy high" bounds with diminishing returns). 

20221022, 01:55  #5 
If I May
"Chris Halsall"
Sep 2002
Barbados
11,087 Posts 
To make an observation. On the meaning of Life, the Universe, and Everything. And that Children can be taught.
Fair. Serious (realtime) answer. I hope this satisfies your requirements. Any further questions? Seriously. 
20221022, 02:21  #6  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
3^{4}·7·13 Posts 
Quote:
It's preferable (more efficient) to use optimal bounds on adequate memory the first time. Some users don't know to adjust the initial conservative low allowed stage 2 ram setting upward, or just can't allocate enough ram. If lacking ram, they should probably leave P1 to those who have enough ram. Last fiddled with by kriesel on 20221022 at 02:23 

20221022, 04:08  #7 
If I May
"Chris Halsall"
Sep 2002
Barbados
11,087 Posts 
Yes. Many have thought deeply about this kind of thing.
The resolution of the problem space seems to be a bit of a personal thing... I don't fully understand why... 9^) 
20221022, 08:18  #8  
"Jacob"
Sep 2006
Brussels, Belgium
1,907 Posts 
Quote:
Code:
PRP (and PRP DC for manual assignments, or most versions) PRP=[<AID>,N/A,<nul>]<k>,<b>,<n>,<c>[,<how_far_factored>,<tests_saved>[,<prp_base>,<residue_type>[,"commaseparatedlistofknownfactors"]]] PRP=N/A,1,2,82589933,1 (mersenne prime record) PRP=N/A,1,2,268435459,1,80,0,3,5,"3" (Wagstaff number) PRP=1,2,82589933,1,82,0 (to have PrimeNet issue an AID for it at the next checkin) NOTE: as of v30.x, it's recommended to include <how_far_factored,<tests_saved>, to prevent repeating unnecessary TF from 0 bits, and prevent repeating unnecessary P1 factoring not signing to save space 

20221022, 13:25  #9 
Romulan Interpreter
"name field"
Jun 2011
Thailand
10281_{10} Posts 
"Tests saved" used to refer to PRP/LL tests that would be "saved" if a factor is found. PRP/LL tests take a while to do. If P1 founds a factor, there is no need to run the costly PRP/LL tests. Before, when we were doing LL, each factor you found saved between 0 and about 2.5 LL tests, depending when the factor was found (before any LL, after the LL, but before DC, or after DC, as people still like to find record factors and they were trying to factor even after DC  the fraction part comes from errors, redoing tests, triple checks, whatever). Since we do PRP+Cert, each factor you find saves between 0 and ~1.3 PRP tests, more or less. So, if a PRP test takes a week to run on your hardware, and a P1 test with 5% chance to find a factor takes 7 hours, then you better do P1, this way you will run ~24 tests in 7 days, having a much better chance to clear one exponent (by finding a factor in average for every 20 tests you run) in the same time (plus, chance to find a record factor, but on the other hand, no chance to find a prime).
In the past, this "tests saved" was used by P95 to establish the B1/B2 limits according with your hardware. You could artificially increase it (edit the file by hand) to force P95 to do more P1 (we did this, it was a common practice). Since new versions of P95 have other habits in doing stage 2, with a much larger B2, we don't know how the things are done currently, and how the parameter is still used. Last fiddled with by LaurV on 20221022 at 13:42 
20221022, 13:38  #10  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
1CCB_{16} Posts 
A better place for https://mersenneforum.org/showpost.p...58&postcount=8 would have been the reference info discussion thread.
Test (first LL) and DoubleCheck (LLDC) use a field P1_done (boolean, true=1, false=0). (Serverissued example for P1 already done: DoubleCheck=(aid),74213963,75,1) (LL and DoubleCheck worktodo formats were mostly set before P1 factoring was added to prime95.) PRP uses a field tests_saved (~number of primality tests expected saved by finding a factor, and prime95 and the server support limited precision reals, such as 1.3 which the server currently emits, or 0. Servermanualissued example: PRP=<aid>,1,2,74214079,1,75,0). There was a discussion re Mlucas only supporting for PRP, 0, 1, or 2 in that field a while back, and before that a discussion for gpuowl implementation. Similarly to PRP, Pfactor issues with number_of_tests_saved_by_finding_a_factor. Work assignments for P1 retry, on exponents that only got a stage 1 P1 and have been LL or PRP tested once but need a DC, are all being issued P1 with tests_saved=1 also: https://mersenneforum.org/showthread.php?t=28038 "Tests_saved" is in units of primality tests performed on the same hardware. LL and PRP are in math terms, about the same effort required, for the same efficiency of code. Generally, for the worktodo formats, anything I post there is verified by multiple inthewild examples and/or the result of also consulting the program authors, PrimeNet API, or application source code. Note, the posted API is not current, but see its 5.3.5.1.2 for parameters p1 and saved. They're quite different. Section 7.2 shows use of p1 for LL or LLDC. There's a bit of supplementary information provided in a reference post. See also the assignment examples explanations in https://www.mersenneforum.org/mayer/README.html toward its end. Quote:
Last fiddled with by kriesel on 20221022 at 14:22 

20221022, 15:02  #11 
"GIMFS"
Sep 2002
Oeiras, Portugal
623_{16} Posts 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Unsuccessful LLtest.  Jan S  GpuOwl  4  20210728 01:50 
Free Trials of GPU Cloud Computing Resources  NBtarheel_33  GPU to 72  9  20130731 15:32 
Early Trials with OpenCL (Barrett's Modulus)  chrisjp  GPU Computing  39  20110429 00:55 
Guantanamo trials to be restarted  garo  Soap Box  39  20110322 23:07 
Successful TF worth more than unsuccessful TF?!  NBtarheel_33  PrimeNet  5  20100617 00:17 