SkipTrialFactoring=1 doesn't work?

cmokruhl · 2002-10-15, 17:56

On my P4 machines, I want to skip the trial factoring before running the LL test since the LL test runs very fast. The undocumented feature "SkipTrialFactoring=1" seem to be just what the doctor ordered. However, I have that in the prime.ini files and it still does trial factoring. Am I doing something wrong?

Assuming this doesn't work, can I accomplish the same thing by changing the worktodo.ini files:

Test=13949693,65,0
to
Test=13949693,65,1

By the way, I'm using 22.10 on Linux.

dswanson · 2002-10-15, 19:04

The "SkipTrialFactoring=1" feature bypasses the trial factorization up to n bits. In your example, "Test=13949693,65,0", trial factorization has already been completed to 65 bits so no further factorization is needed. Thus skipping the trial factoring does nothing.

It sounds like what you're really trying to skip is the P-1 factorization. Yes, changing the 0 to a 1 as you show accomplishes this.

However, from an overall GIMPS efficiency viewpoint, it'd be better if you allowed the P-1 factorization to run. George has picked the P-1 limits such that, on average, the CPU cycles expended per factor found using P-1 is less than the CPU cycles expended per LL test completed.

In other words, if you did 100 first-time LL tests with P-1 turned on, on average you'd find about 5 factors and complete 95 LL tests. If you did the same 100 first-time LL tests with P-1 turned off, you'd find 0 factors and complete 100 LL tests. But the total time spent doing those 100 LL tests would be longer than the time spend doing 100 P-1 factorizations plus 95 LL tests. So you're less efficient without P-1.

2002-10-15, 17:56	#1
cmokruhl Sep 2002 2·3 Posts	SkipTrialFactoring=1 doesn't work? On my P4 machines, I want to skip the trial factoring before running the LL test since the LL test runs very fast. The undocumented feature "SkipTrialFactoring=1" seem to be just what the doctor ordered. However, I have that in the prime.ini files and it still does trial factoring. Am I doing something wrong? Assuming this doesn't work, can I accomplish the same thing by changing the worktodo.ini files: Test=13949693,65,0 to Test=13949693,65,1 By the way, I'm using 22.10 on Linux.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GPU72 doesn't show all available DC work	Mark Rose	GPU to 72	2	2014-01-24 00:33
mfaktc doesn't work for me	sonjohan	GPU Computing	11	2012-01-27 15:10
autoreconf (from autoconf 2.68) doesn't work...	Ralf Recker	GMP-ECM	1	2011-07-25 14:14
Prime95 as a service in XP doesn't always work	petrw1	PrimeNet	8	2008-01-27 01:59
Desktop doesn't work.	clowns789	Software	7	2003-10-19 21:33

2002-10-15, 19:04	#2
dswanson Aug 2002 2³·5² Posts	The "SkipTrialFactoring=1" feature bypasses the trial factorization up to n bits. In your example, "Test=13949693,65,0", trial factorization has already been completed to 65 bits so no further factorization is needed. Thus skipping the trial factoring does nothing. It sounds like what you're really trying to skip is the P-1 factorization. Yes, changing the 0 to a 1 as you show accomplishes this. However, from an overall GIMPS efficiency viewpoint, it'd be better if you allowed the P-1 factorization to run. George has picked the P-1 limits such that, on average, the CPU cycles expended per factor found using P-1 is less than the CPU cycles expended per LL test completed. In other words, if you did 100 first-time LL tests with P-1 turned on, on average you'd find about 5 factors and complete 95 LL tests. If you did the same 100 first-time LL tests with P-1 turned off, you'd find 0 factors and complete 100 LL tests. But the total time spent doing those 100 LL tests would be longer than the time spend doing 100 P-1 factorizations plus 95 LL tests. So you're less efficient without P-1.