mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Factoring (https://www.mersenneforum.org/forumdisplay.php?f=19)
-   -   Number of relative primes for P-1 (https://www.mersenneforum.org/showthread.php?t=23777)

petrw1 2018-11-06 00:38

Number of relative primes for P-1
 
Now that I am doing a lot of P-1 I'm intrigued by the number of relative primes

[CODE][Nov 5 17:57] Available memory is 3151MB.
[Nov 5 17:57] Using 3141MB of memory. Processing 144 relative primes (259 of 960 already processed).
[/CODE]

All my recent P-1 work has been for exponents in the 40-59M ranges; yet I see "relative primes" ranging from 144 to 960 with many values between.

At one time I thought it might be some magical function of the exponent range...until yesterday.

I have a PC working on P-1 in the 47.2M range and for 5 days it was doing 480 relative primes; which by the way is the most common value I've seem for P-1.
However, last night I changed the CPU setting to increase RAM overnight (it is now 16G daytime and 20G overnight). Since then the P-1 work on this same PC is now doing 960 rp's.

Now I am thinking the value is a function of the RAM allocated.
I am going to further assume it is a function of the maximum RAM; otherwise assignments starting during the day (when I have 16G) would get 480 rp's and 960 rp's when started overnight.

Am I anywhere close is my guesses?
Is there a layman's answer to this question?

axn 2018-11-06 02:01

[QUOTE=petrw1;499683]Am I anywhere close is my guesses?
Is there a layman's answer to this question?[/QUOTE]
Memory allocated can affect the number of "relative primes" used.

However, there is not significant difference in performance between 16G and 20G of allocation. In fact, if an assignment crosses the day/night boundary (in either direction), it will stop and restart with new memory settings, and will cause so much additional work that it will wipe out any potential gains. Please stick with one setting for the entire day.

Xyzzy 2018-11-06 02:15

In the 87M range, 12GB is enough to get the "E=12" Brent-Sujama thingie.

Related thread: [url]https://www.mersenneforum.org/showthread.php?t=23331[/url]

:mike:

petrw1 2018-11-06 02:49

[QUOTE=axn;499690]Memory allocated can affect the number of "relative primes" used.

However, there is not significant difference in performance between 16G and 20G of allocation. In fact, if an assignment crosses the day/night boundary (in either direction), it will stop and restart with new memory settings, and will cause so much additional work that it will wipe out any potential gains. Please stick with one setting for the entire day.[/QUOTE]

So noted....thanks.

petrw1 2018-11-07 20:20

P-1 with 20GB is slower than 16GB
 
After a couple days at 20GB of RAM instead of 16GB I am seeing:

- 960 relative primes INSTEAD of 480 relative primes
- same GhzDays credit
- BUT it is not taking a little longer. With a total of 480 RPs stage 2 was taking almost exactly 1 minute per RP. i.e. if it was doing 60 RPs it took 60 minutes. But with a total of 960 RPs it is taking MORE THAN 2 minutes per RP. i.e. if it is doing 60 RPs it is taking about 135 minutes and so the entire P-1 is taking longer overall.

Seems odd that I am slowing down with MORE RAM.

NOTE a typical assignment is:

[CODE]Pminus1=N/A,1,2,47299781,-1,1000000,20000000,73[/CODE]


Hold the presses...

After a more formal and official calculation it seems it made not enough difference to measure yet...


All times are UTC. The time now is 02:28.

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