mprime ROUND OFF ERROR: Triple-check advised?

[Bdot]

I have run an LL-test (Linux64,Prime95,v27.7,build 2, 8 single-threaded tasks) with ROUND OFF errors:

Trying 1000 iterations for exponent 46194101 using 2400K FFT.
If average roundoff error is above 0.24261, then a larger FFT will be used.
Final average roundoff error is 0.24054, using 2400K FFT for exponent 46194101.
...
[Mon Nov 12 06:26:43 2012]
Iteration: 22288153/46194101, ERROR: ROUND OFF (0.40625) > 0.40
Continuing from last save file.
...
[Sat Nov 17 22:39:51 2012]
Iteration: 33876613/46194101, ERROR: ROUND OFF (0.40625) > 0.40
Continuing from last save file.
[Sat Nov 17 22:50:06 2012]
Disregard last error.  Result is reproducible and thus not a hardware problem.
For added safety, redoing iteration using a slower, more reliable method.
Continuing from last save file.
...
[Sat Nov 24 02:32:34 2012]
UID: xxx, M46194101 is not prime. Res64: 292A79D0C1A98503. We8: D48D3B43,22678181,01000200, AID: 7D2290571113FFB8C76DD31A032CFE25

When I just checked if the first round-off-error without "Disregard ..." and the We8:...,01000200 have lead to a suspicious result, I found out, that I actually did a successful DC.

While this means that "Wayne Beardsley" had the same residue, can we be confident that the result is correct?

This machine has never had bad results before this, and neither since then - I'm wondering if that would be a case where another check would better be done with higher FFT size, or ... ?

[LaurV]

Originally Posted by Bdot

While this means that "Wayne Beardsley" had the same residue, can we be confident that the result is correct?

Totally! (Master Oogway)
These runs come from two different versions and "shifts" of P95. It is impossible for a hardware (or even a software/FFT) error to produce the same residue.

[Dubslow]

Originally Posted by LaurV

It is impossible for a hardware (or even a software/FFT) error to produce the same residue.

Well, almost impossible. The chances are incredibly slim.

[Prime95]

and a roundoff error barely over 0.4 (yours was 0.40625) is perfectly normal.

[Bdot]

Originally Posted by Prime95

and a roundoff error barely over 0.4 (yours was 0.40625) is perfectly normal.

OK, maybe not "normal" as in happening every day to me, but I believe you .

Thank you all for your reassuring answers.

[Prime95]

Originally Posted by Bdot

maybe not "normal" as in happening every day to me.

Yeah, normal as in not uncommon when testing an exponent near the upper limits of an FFT length.