mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   Double check LL test faster than first run test (https://www.mersenneforum.org/showthread.php?t=11060)

lidocorc 2008-11-30 14:46

Double check LL test faster than first run test
 
Why is a double check LL test about twice as fast than a first run LL test?

As far as I understood LL testing the number of iterations has to be the same for both kinds of LL tests. Is it because a first run LL test does some intermediate error detecting, which a double check leaves out? What else could it be to make a double check faster then a first run check?

axn 2008-11-30 15:57

[QUOTE=lidocorc;151365]Why is a double check LL test about twice as fast than a first run LL test?[/QUOTE]

For the same exponent, double check & frist time LL take the same amount of time (on the same kind of CPU, natch). They're basically the same computation sequence (the random shift count not withstanding).

However, at any given point in time, the exponents given out for first time testing will be much bigger than the ones handed out for doublecheck. Does that answer your question?

cheesehead 2008-12-01 01:14

[quote=lidocorc;151365]Why is a double check LL test about twice as fast than a first run LL test?[/quote]In addition to what axn explained:

[I]If[/I] what you're comparing is the elapsed time taken by a DC now to the elapsed time taken for the first LL test back when it was first tested, consider that the average computer used by GIMPS participants has gotten faster over time. So the (perhaps) 2.66 GHz system that does the DC may do so in half the time needed by the (perhaps) 1200 MHz system that did the first test.

lycorn 2008-12-03 15:12

[QUOTE=axn;151375]For the same exponent, double check & frist time LL take the same amount of time (on the same kind of CPU, natch). They're basically the same computation sequence (the random shift count not withstanding).

However, at any given point in time, the exponents given out for first time testing will be much bigger than the ones handed out for doublecheck. Does that answer your question?[/QUOTE]

Actually, for the current ranges (46M for 1st time LL and 23M for DCs) the time taken for a DC is roughly 1/4 of the time taken for a 1st time LL (1/2 of the iterations, each one done in 1/2 of the time)


All times are UTC. The time now is 10:28.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.