mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   GPU to 72 (https://www.mersenneforum.org/forumdisplay.php?f=95)
-   -   Should I TF 70—>73 or 72—>73? (https://www.mersenneforum.org/showthread.php?t=17699)

Chuck 2013-01-25 02:20

Should I TF 70—>73 or 72—>73?
 
Which is better? I thought it might be better to work the "last mile" 72—>73 since that helps get the exponents out of GPU to 72 and back to the PrimeNet pool.

But is time better spent starting from 70 (with StopAfterFactor=2 to bail out as soon as possible after finding a factor) since the chance of finding a factor is higher at the lower TF level, although clearing fewer exponents out in the same amount of time?

Tell me what to do...

LaurV 2013-01-25 03:11

[QUOTE=Chuck;325730]Which is better? I thought it might be better to work the "last mile" 72—>73 since that helps get the exponents out of GPU to 72 and back to the PrimeNet pool.

But is time better spent starting from 70 (with StopAfterFactor=2 to bail out as soon as possible after finding a factor) since the chance of finding a factor is higher at the lower TF level, although clearing fewer exponents out in the same amount of time?

Tell me what to do...[/QUOTE]

Please DO NOT skip bitlevels.
First, is not optimum, you have more chances to find factors at lower bitlevels, for the unit of the time spent doing TF.
Second, is not "orthodox", skipped levels tend to be forgotten and never done.
Third, you do it for free, as you will not be able to report it, PrimeNet does not accept "non-contiguous" bit ranges. So, you waste the time, no credit for you.

If the exponent is TF to 70 on PrimeNet's DB, then please do 70 up.

chalsall 2013-01-25 03:41

[QUOTE=Chuck;325730]Which is better? I thought it might be better to work the "last mile" 72—>73 since that helps get the exponents out of GPU to 72 and back to the PrimeNet pool.[/QUOTE]

It is best for GIMPS to take whatever is lowest (by Candidate) to what is deemed to the optimal bit level. Then it can be released back to GIMPS for assignment for LLing.

[QUOTE=Chuck;325730]But is time better spent starting from 70 (with StopAfterFactor=2 to bail out as soon as possible after finding a factor) since the chance of finding a factor is higher at the lower TF level, although clearing fewer exponents out in the same amount of time?[/QUOTE]

Entirely up to you. Choosing "Lowest TF Level", and taking it up to the release level is a reasonable strategy. The current "What Makes Sense" is taking the lowest available candidate up to the release level.

[QUOTE=Chuck;325730]Tell me what to do...[/QUOTE]

Choose to do what you enjoy most.

At the end of the day all the current candidates need to go up to (at least) 73. Some get fixated on finding factors, and/or the "GHz Days Saved" metric so only go one level. That just means that someone else has to step in and do the rest of the work.

chalsall 2013-01-25 03:44

[QUOTE=LaurV;325734]Please DO NOT skip bitlevels.[/QUOTE]

Chuck understands this.

Just in case you don't remember, he's the reason the "Highest TF Level" is an option. He volunteered about a year ago to do the less-coveted "last level".

LaurV 2013-01-25 05:34

[QUOTE=chalsall;325741]He volunteered about a year ago to do the less-coveted "last level".[/QUOTE]

:tu:

Sorry! :bow:
Indeed I did not know/remember that.

Chuck 2013-01-25 13:17

OK. I see the logic of working on the lowest exponents. I'll switch back to "What makes sense" option.


All times are UTC. The time now is 04:38.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.