mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Linux (https://www.mersenneforum.org/forumdisplay.php?f=39)
-   -   option for round off checking in linux64 version? (https://www.mersenneforum.org/showthread.php?t=11840)

graysky 2009-05-07 13:49

option for round off checking in linux64 version?
 
I have used p95 under Windows before for the purpose of stability checking. I remember seeing an option under the “Advanced” menu and called “Round off Checking.” As I recall, this needed to be set to enable or else errors wouldn't be reported. I'm running the mprime259_linux64 version of p95 now under a 64-bit Linux OS and can't for the life of me find that round off checking option anywhere.

I'm doing the blend test to test out my new memory but am wondering if errors will get reported as-is?

cheesehead 2009-05-07 14:33

[quote=graysky;172720]I have used p95 under Windows before for the purpose of stability checking. I remember seeing an option under the “Advanced” menu and called “Round off Checking.” As I recall, this needed to be set to enable or else errors wouldn't be reported. I'm running the mprime259_linux64 version of p95 now under a 64-bit Linux OS and can't for the life of me find that round off checking option anywhere.

I'm doing the blend test to test out my new memory but am wondering if errors will get reported as-is?[/quote]Don't worry. "Round Off Checking" just means that extra checking, beyond the standard amount, will be done, and that the roundoff error amount, no matter how nonthreatening, will be reported on your iteration lines. Small roundoff errors in the 0.00-0.09 range are routine and don't threaten the accuracy of our integer results.

If there is a roundoff error that is big enough (> 0.2 or so) to actually imperil the accuracy of the calculation, it'll get reported no matter what "Round Off Checking" is set to.

When you observed that "this needed to be set to enable or else errors wouldn't be reported", the missing clarification was that instead of having only threateningly-large errors being reported [I]as is the default[/I], "Round Off Checking" caused [I]all[/I] errors, no matter how small or inconsequential, to be reported.

As for whether the mprime259_linux64 version of p95 has that option, and where it is: I don't know.

Prime95 2009-05-07 14:43

Round-off checking refers to extra error checking that an ordinary user can insist on when processing work units.

You are doing a torture test rather than processing work units. The torture test will catch any errors that occur during computations. The results of each test are compared to known good results.

graysky 2009-05-07 14:43

Thanks for the quick replies, all. It sounds like my stress test will be valid w/ the default options then.


All times are UTC. The time now is 03:42.

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