mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   GPU Computing (https://www.mersenneforum.org/forumdisplay.php?f=92)
-   -   mfaktc: Mis-reported composite factor (https://www.mersenneforum.org/showthread.php?t=20947)

mattmill30 2016-02-02 15:13

mfaktc: Mis-reported composite factor
 
Further to [URL]http://www.mersenneforum.org/showthread.php?t=20943[/URL], I understand that during a T77-81, using the 95bit_mul32_gs kernel, corruption went undetected.

Apparently mfaktc switched both the exponent and bitlevel being analysed.

I've repeated the test without the same factor being reported.
Note: The original factor result was still in results.txt during the retest.

The main change I can think of between the two tests, is the random intervals at which the tests were stopped to improve graphics responsiveness for whilst the computer was in use.

I believe mfaktc was configured for both tests with:
SievePrimes=200000
NumStreams=1
GridSize=3
Stages=0
StopAfterFactor=0
SieveOnGPU=1
GPUSievePrimes=1075000
GPUSieveSize=128
GPUSieveProcessSize=32

Since the factor is 95bits in length, is there a memory bug in the 95bit_mul32 kernel?
Also, the factor is returned as part of the self test, so does the kernel have bug which caused it to initiate and report on a self-test?
([url]http://www.mersenneforum.org/showpost.php?p=267068&postcount=1123[/url])
Note: Though the TF run would have been continuation from a checkpoint, I'm not certain whether the result was reported before (self-test), during (M332347303) or on termination (CTRL+C).

Why would mfaktc choose 95bit_mul32_gs over barrett87_mul32_gs for factoring T77-81?

TheJudger 2016-02-05 19:41

[QUOTE=mattmill30;424983]
Why would mfaktc choose 95bit_mul32_gs over barrett87_mul32_gs for factoring T77-81?[/QUOTE]

Because you set "Stages=0". The barett92/88/87 family can only do a single bitlevel at once.


All times are UTC. The time now is 23:31.

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