(archive)MISFIT

[swl551]

Originally Posted by flashjh

I am currently working an exponent that I broke into two because it's so large:

Factor=2012-11-21,332240563,77,78
Factor=2012-11-21,332240563,78,79

It's telling me they're a duplicate even though they're not. Can you make it also check the from and to for duplicate checking?

SURE...give me about an hour

[swl551]

VERSION 2.1.2
1.Changed duplicate exponent checking to include the bit level as part of the evaluation

Factor=2012-11-28,74680841,72,73
Factor=2012-11-28,74680841,73,74

would no longer be considered duplicate. (always be careful if you fiddling your bits!)


2.Tightened up some interval configurations to prevent certain events from occurring too frequently.


Get from http://www.mersenneforum.org/misfit/

[Stef42]

Originally Posted by swl551

I suspect they are. Who do I give the list to so they can be reversed out?

If exponents are completed and send to GIMPS, the final residue (in results.txt) will be compared to the first LL test.
If they match, then the LL's are verified. In your case, I guess all the residue's do not match.
A third check is then handed out to the first person asking for work.
So, the system has a safe-guard for these kind of malfunctions.

You should CUDALucas set to automatic, so that the program can choose the correct N= size.
The error rate should be below 0.25 at all times. For example, for exponent 28993147, N=1600K with err = 0.08.

Keep in mind that many GTX570 are overclocked, with memory set at 1900Mhz.
Some people here on the forum found out that this speed can cause mismatch residue's.
Lower the speed to 1800Mhz when running CUDALucas. I've done the same and all residue's so far have matched, about 10 in total so far.

[swl551]

Originally Posted by Stef42

...Lower the speed to 1800Mhz when running CUDALucas.

Yes overclocking was the cause of all my problems and how I ended up with a low FFTLength which allowed CuLu to actually stay running. I never realized the condition being created.

[kladner]

The irony in this is that 1900 MHz is actually nVidia specification for GTX 570 RAM. On my Gigabyte card it will work OK running mfaktc, but has to be down clocked for CuLu. Gigabyte lists two different brands of VRAM: Hynix and Samsung. I suspect that the Samsung chips are not up to spec. One other user here AFAIK has the same card as I do (GV-N570OC-13I (rev. 2.0)), and has had the same problem. I don't know if this problem occurs on other manufacturers' cards.

[Stef42]

Mine is a Asus GTX570 DCII. Since I saw the messages involving a lower memory clock I just went ahead and lowered it. I didn't want to waste 28 hours of computing time....

[swl551]

Let's get this thread back to MISFIT specific topics.

Specifically the MISFIT for CULU release..

If you not familiar with MISFIT check it out here


http://www.mersenneforum.org/misfit/

[kracker]

I was fiddling with some exponents when I got this error:

12/4/2012 11:37:59 AM Error!
Analyzing WorkToDo.txt rows
Value was either too large or too small for an Int32.

[flashjh]

Originally Posted by kracker

I was fiddling with some exponents when I got this error:

12/4/2012 11:37:59 AM Error!
Analyzing WorkToDo.txt rows
Value was either too large or too small for an Int32.

Which exponents?

[kracker]

3,321,xxx,xxx, wasn't doing them on gpu actively,was just tinkering around.

[swl551]

Originally Posted by kracker

3,321,xxx,xxx, wasn't doing them on gpu actively,was just tinkering around.

int32 has a max value of

2,147,483,647

If you are making up unreasonable numbers then MISFIT works perfectly. If you are making up realistic numbers then MISFIT needs to move to a larger data type.