mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   PrimeNet Assignment Rules (https://www.mersenneforum.org/showthread.php?t=23153)

S485122 2018-03-12 09:35

PrimeNet Assignment Rules
 
It seems to me that on the [url=https://www.mersenne.org/thresholds/]PrimeNet Assignment Rules[/url] page the condition [code]Setting for "days of work to queue" <= 5.[/code] is missing for Category 1 Double-Checking.
For Category 0 it is <=3 and for category 2 it is <=10. It would then be coherent with the category 1 conditions for the Lucas-Lehmer tests.

Jacob

Prime95 2018-03-13 00:42

The web page is correct.

Cat 1 originally may have required days-of-work <= 5 and after some time not enough cat 1's were getting assigned. Thus, the restriction was removed.

storm5510 2018-06-01 00:55

[QUOTE]Your account is set for each computer to perform at least [B]one matching double-check yearly[/B] to ensure the computer is operating properly. [B]Highly recommended.[/B][/QUOTE]Does it matter how this test is ran? I am 55% through this on a 1080 using [I]CUDALucas[/I]. Just under 14 hours left. :confused2:

ATH 2018-06-01 13:15

[QUOTE=storm5510;488797]Does it matter how this test is ran? I am 55% through this on a 1080 using [I]CUDALucas[/I]. Just under 14 hours left. :confused2:[/QUOTE]

If this is a double check meant for your CPU that you moved to the 1080 GPU, then it defeats the purpose. The point is to run 1 double check per year on every CPU that you use for LL/PRP tests, to test if the CPU is reliable.
It is not a bad idea to double check on GPU as well, but that is a manual process.

storm5510 2018-06-02 01:05

[QUOTE=ATH;488848]If this is a double check meant for your CPU that you moved to the 1080 GPU, then it defeats the purpose....[/QUOTE]


I came to this conclusion on my own, but wanted to verify it. Thank you! :smile:

kriesel 2018-06-02 16:06

[QUOTE=storm5510;488906]I came to this conclusion on my own, but wanted to verify it. Thank you! :smile:[/QUOTE]
It's a good thing to occasionally run double-checks on your gpu, in addition to your cpu.
Not only does it double check both hardware units, it double-checks the software, potentially in an fft length not run on your hardware before, which might show up new issues.
(I'm currently running CUDAPm1 on multiple gpu models on an assortment of fft lengths each, and finding interesting behaviors that differ model to model. CUDAPm1 was derived from CUDALucas, so some of it may apply somewhat to CUDALucas also.)

henryzz 2018-06-08 12:38

[QUOTE=kriesel;488978]It's a good thing to occasionally run double-checks on your gpu, in addition to your cpu.
Not only does it double check both hardware units, it double-checks the software, potentially in an fft length not run on your hardware before, which might show up new issues.
(I'm currently running CUDAPm1 on multiple gpu models on an assortment of fft lengths each, and finding interesting behaviors that differ model to model. CUDAPm1 was derived from CUDALucas, so some of it may apply somewhat to CUDALucas also.)[/QUOTE]

If you are running prp there is probably no point because errors will be detected. If it detecting errors you should worry.

kriesel 2018-06-08 14:49

[QUOTE=henryzz;489429]If you are running prp there is probably no point because errors will be detected. If it detecting errors you should worry.[/QUOTE]
That's a good point on prp with Gerbicz check built in, providing continuous error checking. It applies to gpuOwL (on AMD or IGP gpus) and parts of Prime95/mprime. It does not currently apply to NVIDIA gpus, for which no PRP Mersenne application is available. (CUDALucas and CUDAPm1 do not have PRP, Gerbicz check, or Jacobi check.)


All times are UTC. The time now is 08:56.

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