Best type of work for lowest exponent work?

cappy95833 · 2019-11-20, 02:48

What is the best selection for "Type of work to get" to work on sub-million exponents?

I have "Whatever makes the most sense" currently selected, but i keep getting 90,000,000+ exponents.

If there isn't a good choice, is there a way to make my own filter of what type of work to select?

LaurV · 2019-11-20, 03:03

For "sub million" you can only do cofactor primality test or ECM. As the table you posted says, the only available work type is the ECM work, so we suggest you do that if you like finding record factors. Not that you will find a lot, but you may get lucky.

Other work types below 1M make no sense, all the LL were done long ago (multiple-multiple times), TF has zero* chances to find a new factor (due to ECM and P-1 amount done in the range) and extending P-1 limits, albeit has some minuscule chances of getting you a new factor, like around 0.1% or so, involves repeating a lot of work already done by others in the past (because P-1 checkpoints are not saved in the server, it would be too much a haste, so if you extend P-1 bounds, first of all, that is manual work, you can not get assignments from the server, you have to make the worktodo file by yourself, and second, 99.9% of the work you do, like raising to the former boundaries, was already done by other people, you only contribute the additional step, but for that, you have to do all the work from scratch).

If you have a good computer, we suggest you forget about "under million" part, and do normal "what makes sense" LL/PRP assignments in the 90M, or DC in the 50-60M. If you have a good GPU which you like to assign to GIMPS work, than you may either do TF on GPU72, or LL/PRP using cudalucas or gpuowl (read around about).

ATH · 2019-11-20, 06:22

Cofactor primality test is done since a long time under 1M except a few once in a while when a new factor is found.

The CF wavefront is at 9.9M and the CF DC at 7.9M.

lycorn · 2019-11-20, 13:45

In fact, as LaurV and ATH pointed out, the only work type suitable to exponents in the 0 - 1M range is ECM. I am currently doing it, and have found a fair amount of new factors (64 since mid march, to be more precise). If you want to give it a try, use the Primenet manual assignments form, choose ECM Factoring as the type of work and specify a range < 1 M.

storm5510 · 2019-11-20, 17:36

Quote:

Originally Posted by LaurV

For "sub million" you can only do cofactor primality test or ECM. As the table you posted says, the only available work type is the ECM work, so we suggest you do that if you like finding record factors. Not that you will find a lot, but you may get lucky.,,,

Just for reference, I have found 23 factors in 16,644 ECM runs. So, yes, the possibility is quite low. I found a factor using this method which was 39 digits in length. I thought that was pretty good.

The area needing the most attention is double-check, referred to here as "DC." These can take time to finish, depending on your hardware.

kriesel · 2019-11-21, 04:24

Perhaps you'd find something at https://www.mersenneforum.org/showth...922#post521922 useful background.

2019-11-20, 02:48	#1
cappy95833 Mar 2014 3 Posts	Best type of work for lowest exponent work? What is the best selection for "Type of work to get" to work on sub-million exponents? I have "Whatever makes the most sense" currently selected, but i keep getting 90,000,000+ exponents. If there isn't a good choice, is there a way to make my own filter of what type of work to select? Attached Thumbnails Last fiddled with by cappy95833 on 2019-11-20 at 02:55 Reason: adding question details

2019-11-20, 03:03	#2
LaurV Romulan Interpreter Jun 2011 Thailand 7²×197 Posts	For "sub million" you can only do cofactor primality test or ECM. As the table you posted says, the only available work type is the ECM work, so we suggest you do that if you like finding record factors. Not that you will find a lot, but you may get lucky. Other work types below 1M make no sense, all the LL were done long ago (multiple-multiple times), TF has zero* chances to find a new factor (due to ECM and P-1 amount done in the range) and extending P-1 limits, albeit has some minuscule chances of getting you a new factor, like around 0.1% or so, involves repeating a lot of work already done by others in the past (because P-1 checkpoints are not saved in the server, it would be too much a haste, so if you extend P-1 bounds, first of all, that is manual work, you can not get assignments from the server, you have to make the worktodo file by yourself, and second, 99.9% of the work you do, like raising to the former boundaries, was already done by other people, you only contribute the additional step, but for that, you have to do all the work from scratch). If you have a good computer, we suggest you forget about "under million" part, and do normal "what makes sense" LL/PRP assignments in the 90M, or DC in the 50-60M. If you have a good GPU which you like to assign to GIMPS work, than you may either do TF on GPU72, or LL/PRP using cudalucas or gpuowl (read around about). Last fiddled with by LaurV on 2019-11-20 at 03:25

2019-11-21, 04:24	#6
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 2×3²×7×43 Posts	Welcome Perhaps you'd find something at https://www.mersenneforum.org/showth...922#post521922 useful background.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Type of work to get?	Bispen	Information & Answers	3	2016-01-27 16:46
Type of work I'm doing	Unregistered	Information & Answers	17	2013-06-22 04:49
Best type of work for my cpu	Unregistered	Information & Answers	11	2013-05-17 05:22
Type of work to get?	ZFR	Information & Answers	7	2011-09-17 08:43
What type of work should I request?	Malta Soron	PrimeNet	5	2006-10-07 22:03

2019-11-20, 06:22	#3
ATH Einyen Dec 2003 Denmark 3158₁₀ Posts	Cofactor primality test is done since a long time under 1M except a few once in a while when a new factor is found. The CF wavefront is at 9.9M and the CF DC at 7.9M.

2019-11-20, 13:45	#4
lycorn "GIMFS" Sep 2002 Oeiras, Portugal 3×491 Posts	In fact, as LaurV and ATH pointed out, the only work type suitable to exponents in the 0 - 1M range is ECM. I am currently doing it, and have found a fair amount of new factors (64 since mid march, to be more precise). If you want to give it a try, use the Primenet manual assignments form, choose ECM Factoring as the type of work and specify a range < 1 M.