20221116, 18:27  #13  
Random Account
Aug 2009
Not U. + S.A.
7×19^{2} Posts 
Quote:


20221116, 18:31  #14  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
2·29·127 Posts 
Yes. Well, it optimizes the bounds selection given projected odds of finding a factor versus many bounds values and run time estimates for P1 and PRP and allowed ram usage, for maximum expected computing time savings.
Quote:
Last fiddled with by kriesel on 20221116 at 18:43 

20221116, 18:36  #15 
"James Heinrich"
May 2004
exNorthern Ontario
5·19·43 Posts 
Prime95 selects optimal bounds based on the available RAM vs probability of finding a factor vs expected time saved not doing PRP when factor found. Here we don't want "really high" bounds, we want optimal bounds to maximize number of exponents factored (and therefore DC/PRP saved) per time. More RAM will give slightly better chance of factor and slightly longer runtime, but as long as you have at least (as requested in the thread title) 24GB allocated to Prime95 it should give you decent bounds. These assignments here are exponents that have been LLtested once, but with bad P1, so running a decent P1 could potentially save one PRP test (and associated certification). We don't want to spend any more effort looking for factors than is necessary to balance the time savings of finding a few factors, to optimize overall GIMPS throughput.
Last fiddled with by James Heinrich on 20221116 at 18:37 
20221119, 21:48  #16 
"/X\(‘‘)/X\"
Jan 2013
3,089 Posts 
Will 24 GB be sufficient for all these assignments under 100M?

20221119, 21:55  #17 
"James Heinrich"
May 2004
exNorthern Ontario
5·19·43 Posts 
Sufficient, yes. If you have more RAM by all means make use of it, but 24GB should be fine to give a "decent" P1 effort, far better than the poor P1 that was done previously.

20221123, 05:15  #18 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
2·29·127 Posts 
Here's one we would not have gotten with ordinary bounds, and its factor is 102 bits.
https://www.mersenne.ca/exponent/85408633 
20221128, 16:47  #19 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
16306_{8} Posts 
Mark or anyone can contribute to good firsttest wavefront P1. I have mine configured to queue up some of that automatically if they get close to running out of retry before dc assignments. Good first P1 prevents the need for retry, and reduces unnecessary first prp.
Last fiddled with by kriesel on 20221128 at 16:48 
20221128, 17:12  #20  
"/X\(‘‘)/X\"
Jan 2013
3,089 Posts 
Quote:


20221128, 17:19  #21 
"/X\(‘‘)/X\"
Jan 2013
3,089 Posts 

20221128, 17:32  #22 
"James Heinrich"
May 2004
exNorthern Ontario
111111110101_{2} Posts 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Strategic couple and dribble checks (PRP's, P1's, and special Certs too)  Uncwilly  Marin's Mersennearies  18  20230130 00:52 
GMPEcm as Pm1 requires very little RAM  jocelynl1204  Data  11  20210302 16:32 
Strategic Double Clicking  Madpoo  Marin's Mersennearies  1841  20190716 03:30 
factorial puzzle (requires maths)  graeme  Puzzles  7  20030819 20:40 
Perfect shuffling puzzle (requires programming)  NickGlover  Puzzles  18  20030726 01:10 