suggestions for new work types

ixfd64 · 2011-09-17, 05:58

I think it would be useful if GIMPS offered the following work types:

Trial factor to high limits (TF-Hi): Searches for factors one or two above the current threshold. Good for older GPUs and future 128-bit operating systems.

Trial factor to ultra-high limits (TF-U): Searches for factors up to six bits above the current threshold. Good for high-end GPUs.

Error checking (ERR): Assignments with non-zero error codes and/or residue mismatches. Defaults to regular double-checks if no such assignments are available.

Optimal assignments (OPT): Focuses on (but not limited to) stuff that GIMPS needs the most - in this case, P-1 factoring and double checks (according to George).

Factoring only (No-LL): assigns trial factoring and P-1. Good for "factor freaks" or folks with slower computers.

Upper-range LL tests (LL-Hi): gives LL tests between currents "record size" exponents and those that yield 100 million-digit numbers. Perfect for those who like to use multiple cores on one exponent.

What does everyone think?

James Heinrich · 2011-09-19, 19:44

+1

If we're throwing ideas out, how about two more ideas:

a) "second-chance P-1", for exponents that have had 1 LL (or at least no matching DC) but no P-1.

b) "do-it-properly P-1", for exponents that have technically had P-1 done, but practically it was useless (there's a bunch, but for example a whole swath of exponents at around 400M that have B1=B2=5000 -> 0.092% chance of factor).

cheesehead · 2011-09-19, 23:11

Quote:

Originally Posted by ixfd64

Optimal assignments (OPT): Focuses on (but not limited to) stuff that GIMPS needs the most - in this case, P-1 factoring and double checks (according to George).

How would that differ from the existing choice "whatever makes the most sense"?

ixfd64 · 2011-09-19, 23:53

I suspect that "whatever makes the most sense" focuses more on the speed of the user's computer rather than GIMPS.

NBtarheel_33 · 2011-09-20, 07:20

Quote:

Originally Posted by ixfd64

I suspect that "whatever makes the most sense" focuses more on the speed of the user's computer rather than GIMPS.

This is true, though it can (and has been) tweaked by George in the past to hand out a certain kind of assignment at a particular interval (e.g. I believe that 15% or 20% of all such assignments now are DCs; this was done to accelerate the DC wavefront).

2011-09-17, 05:58	#1
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2·5·239 Posts	suggestions for new work types I think it would be useful if GIMPS offered the following work types: Trial factor to high limits (TF-Hi): Searches for factors one or two above the current threshold. Good for older GPUs and future 128-bit operating systems. Trial factor to ultra-high limits (TF-U): Searches for factors up to six bits above the current threshold. Good for high-end GPUs. Error checking (ERR): Assignments with non-zero error codes and/or residue mismatches. Defaults to regular double-checks if no such assignments are available. Optimal assignments (OPT): Focuses on (but not limited to) stuff that GIMPS needs the most - in this case, P-1 factoring and double checks (according to George). Factoring only (No-LL): assigns trial factoring and P-1. Good for "factor freaks" or folks with slower computers. Upper-range LL tests (LL-Hi): gives LL tests between currents "record size" exponents and those that yield 100 million-digit numbers. Perfect for those who like to use multiple cores on one exponent. What does everyone think?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
What do the different types of work each mean?	jrafanelli	Information & Answers	20	2019-02-01 05:27
Which Work Types are Multithreaded	tului	Software	6	2015-11-28 21:59
New work types	Unregistered	Information & Answers	0	2011-07-25 10:19
Work Types	Unregistered	Information & Answers	3	2010-07-28 09:54
v5 work types	S00113	PrimeNet	14	2008-12-10 00:26

2011-09-19, 19:44	#2
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 3×5×227 Posts	+1 If we're throwing ideas out, how about two more ideas: a) "second-chance P-1", for exponents that have had 1 LL (or at least no matching DC) but no P-1. b) "do-it-properly P-1", for exponents that have technically had P-1 done, but practically it was useless (there's a bunch, but for example a whole swath of exponents at around 400M that have B1=B2=5000 -> 0.092% chance of factor).

2011-09-19, 23:53	#4
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2×5×239 Posts	I suspect that "whatever makes the most sense" focuses more on the speed of the user's computer rather than GIMPS.