20180410, 18:19  #1 
Aug 2006
3×1,993 Posts 
Timing for different B1 values?
Is there a chart somewhere  or could we make one  for how long it takes to do one curve at different B1 values? Obviously there are lots of secondary variables (B2, size of number, computer used, etc.) but I'd be fine with whatever we had.
I feel like this is something that would be referenced often, just to get a ballpark. 
20180410, 18:30  #2 
Oct 2006
Berlin, Germany
2^{2}·157 Posts 
I did it roughly 110 years back to predict runtime of ecm in yoyo@home project and to calculate credits. This LibreOffice document contains a tab per B1 and in each tab different composite sizes, their runtime and memory consumption.

20180410, 19:25  #3  
1976 Toyota Corona years forever!
"Wayne"
Nov 2006
Saskatchewan, Canada
3×1,619 Posts 
Quote:
http://www.mersenne.ca/credit.php Here's an example: http://www.mersenne.ca/credit.php?wo...tton=Calculate 

20180411, 03:19  #4 
"Curtis"
Feb 2005
Riverside, CA
11·461 Posts 
Stage 1 time in GMPECM is linear in B1 to within 2% from 1e5 to 1e9, on every composite I've cared to try.
If someone has evidence to the contrary, I'd like to hear it. 
20180411, 06:38  #5 
Romulan Interpreter
"name field"
Jun 2011
Thailand
23120_{8} Posts 

20180411, 10:34  #6 
Einyen
Dec 2003
Denmark
2·1,601 Posts 

20180412, 04:12  #7 
Romulan Interpreter
"name field"
Jun 2011
Thailand
2650_{16} Posts 
Any tutorial for silly people if I want to run those tests for my own CPUs?

20180510, 00:23  #8  
Mar 2006
746_{8} Posts 
Quote:
http://www.mersenneforum.org/showthread.php?t=22924 I think I'll create a new thread in this subforum so that people looking here for timing info about GMPECM can more easily find it. You can find that new thread here: http://www.mersenneforum.org/showthread.php?t=23326 So far, I've gathered data for numbers of size 100500 decimal digits, with B1 in the range [10e3,990e6]. I'm gathering data for larger B1 values, but that is (understandably) taking much more time. 

20180512, 05:57  #9 
Jan 2013
2^{3}·7 Posts 
How precise you want to be?
Here is a decent formula: TIME [min] = 1min * (B1/10^7) * (Digits/100)^1.5 * ( 2Ghz / f ) Accuracy is at 510% level, you adjust the last factor to your hardware for example by plugging actual core frequency in f. In other words, at B1=1e7, 100 digit number and 2Ghz core this formula will give you 1min for stage1+2. 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
mprime #7 timing summary request  aurashift  Software  5  20171126 21:28 
Timing for large candidate  carpetpool  Conjectures 'R Us  6  20161231 06:02 
Strange timing for GMPECM 6.2.3  jyb  GMPECM  5  20100210 14:01 
Very strange mem timing behavior, Asus M3A67EM  jwh  Information & Answers  1  20090130 18:04 
Timing Options  Kevin  Software  3  20020912 14:03 