mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   Prime95 defaults to proof=8 for PRP-CF tests even though I have a very large disk space allocated (https://www.mersenneforum.org/showthread.php?t=27255)

techn1ciaN 2021-10-21 13:50

Prime95 defaults to proof=8 for PRP-CF tests even though I have a very large disk space allocated
 
I have my Prime95 in a dumb directory that I don't use much and so gave it 100GB for proofs. Wavefront PRP selects proof=10, as expected. But — I recently started running PRP-CF instead out of interest and every test selects proof=8.


I understand that this is almost completely inconsequential because even a power-7 proof on a cofactor-size number would take trivial time to certify. But — I have to say I'm still curious why the program does not select a higher proof power for this specific type of test, when the space for one is certainly available.


I searched for "PRP cofactor proof power" but could only find unrelated posts.

slandrum 2021-10-21 14:13

It's probably related to the size of exponent you are now getting, there's really no reason to generate a giant proof file, and I wouldn't worry about it. Those certs take seconds to run (minutes on older slow hardware).

slandrum 2021-10-21 14:16

If you are really concerned and want to force a particular proof level, read undoc.txt and it will tell you what line to add to local.txt

techn1ciaN 2021-10-21 14:18

I figured there would be something in undoc. I'll take a look.

kriesel 2021-10-21 14:42

Please read and apply [URL]https://www.mersenneforum.org/showpost.php?p=521664&postcount=2[/URL]
Specifically, "Use an existing topic-specific thread when practical."
There are multiple existing prime95 threads, version dependent.
A moderator may choose to merge this newly created thread into an existing thread.
Please do not make extra work for the moderators.

Also, [B]r[/B]ead [B]t[/B]he [B]f[/B]ine [B]m[/B]anual files; think; apply; before posting.

techn1ciaN 2021-10-21 17:28

[QUOTE] There are multiple existing prime95 threads, version dependent. [/QUOTE]


So, you're saying it would have been appropriate to post this in the "Prime95 v30.4/30.5/30.6" (I'm running 30.6b4) thread? If so, apologies; I was under the impression that the version-specific threads were strictly for testing and debugging of new features in the relevant versions and it would be incorrect to clutter them up with procedural questions.


[QUOTE] Also, [B]r[/B]ead [B]t[/B]he [B]f[/B]ine [B]m[/B]anual files ... [/QUOTE]


Of course. At the end of the day I'm only secondarily concerned with the ability to run proof=10 PRP-CF; I was more curious if there was any reason why the program doesn't select proof=10 for PRP-CF when the disk space is available, beyond the obvious "the certs are quick anyway so it's a non-concern." But I suppose not.

Prime95 2021-10-21 22:37

[QUOTE=techn1ciaN;591224]I have my Prime95 in a dumb directory that I don't use much and so gave it 100GB for proofs. Wavefront PRP selects proof=10, as expected. But — I recently started running PRP-CF instead out of interest and every test selects proof=8.[/QUOTE]

Running proof=10 on small exponents will cost more generating the proof than you will save doing the cert.

techn1ciaN 2021-10-22 00:00

[QUOTE=Prime95;591269]Running proof=10 on small exponents will cost more generating the proof than you will save doing the cert.[/QUOTE]


Thank you for the insight!


Approximately where are the cutoff points at which the optimal power goes up? Would Prime95 set proof=11 or proof=12 for a sufficiently large exponent (say, 100M-digit) if the disk space was allocated?

Prime95 2021-10-22 03:13

Over 414.2M the proof power goes to 11.

techn1ciaN 2021-10-22 03:44

[QUOTE=Prime95;591307]Over 414.2M the proof power goes to 11.[/QUOTE]


OK, so it scales very slowly. Interesting.


I happen to do DC-via-PRP using GPUOwl, which of course doesn't get more nuanced than using one, pre-set proof level for every test. I didn't have the information from this thread when I was tuning my installation and set proof=10 without thinking about it. At the current DC wavefront, would it actually be more efficient to use proof=9? If yes, around what exponent should I go back to 10?


Thanks again for your insights thus far

Prime95 2021-10-22 05:32

1.7M - 6.7M = 7
6.7M - 26.6M = 8
26.6M - 106.5M = 9
106.5M - 414.2M = 10
414.2M+ = 11


All times are UTC. The time now is 05:41.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.