[QUOTE]Apparently there wasn't supposed to be any interim residue discrepancy (besides the +2 shift between glucas/mlucas and mprime/prime95, which was accounted for). No point in finishing the test if the interim residues show that something already went wrong.[/QUOTE]

Was there any error reported in the results.txt?

Nehalem is there !
 

I have Glucas running on Nehalem 8 cores.
Verification result should be available Friday 1:45 pm USA CST.
Tony

[QUOTE]I have Glucas running on Nehalem 8 cores.
Verification result should be available Friday 1:45 pm USA CST.[/QUOTE]

:shock: A two day verification? Wow...

[QUOTE=Primeinator;177114]Was there any error reported in the results.txt?[/QUOTE]

Nope.

[QUOTE=mdettweiler;177108]Are you sure, though, that you weren't simply testing the wrong exponent from the start? If that's the case, then the interim residuals would be different no matter what, and you may as well finish the test rather than letting the already completed work go to waste. No harm in doing a plain old doublecheck on an exponent that needs it. :smile:[/QUOTE]
Kevin's interim residues (again, once the iteration starting counter difference of 2 was taken into consideration) *did* match those of the Glucas and Mlucas DCs for the first 10M iterations or so, but somewhere between 10M and 15M the run went off-track for reasons unknown. (The latest Glucas and Mlucas residues continue to match).

Here is the ensuing e-mail exchange between myself and George (hope George is OK with me posting it - no secrets in there, as far as I can tell):

[quote=ewmayer]On Wed, Jun 10, 2009 at 7:40 PM, <ewmayer@aol.com> wrote:
Yep, Looks like Kevin's run went astray ... he posted to the M-forum saying
he saw no roundoff error messages or anything else untoward, and IIRC his
machine is not overclocked.

George, you don't happen to know if Kevin saved any interim savefiles at any
point before the mismatch, do you? It would be interesting to have him rerun
from a last-good savefile to see if the error is reproducible.[/quote]

[quote=Prime95]There are no interim residues or save files earlier than 15M. There is another
possibility - a bug in prime95's multithreading. I've asked him to halt the
run and do a torture test using multiple threads on that specific FFT length.
If that fails, I'll ask another quad core user to do the same.[/quote]

Kevin, it might be interesting to run, say 100,000 iterations both in single core and quad core mode and see if there is any discrepancy. I wouldn't expect there to be, I imagine George has tested the multi-core mode pretty thoroughly, and yet... My impression is that keeping a quad-core cool can be tricky with certain mobos.

[quote=T.Rex;177117]I have Glucas running on Nehalem 8 cores.
Verification result should be available Friday 1:45 pm USA CST.
Tony[/quote]

No waking up early then! (I'm in Britain.)

See @ [url]http://www.mersenne.org/report_LL/?exp_lo=1&exp_hi=45000000&exp_date=&user_only=1&user_id=stigmv&exbad=1&exfactor=1&dispdate=1&B1=Get+LL+data[/url]

@ the bottom of the page...

[QUOTE]See @ [url]http://www.mersenne.org/report_LL/?e...B1=Get+LL+data[/url]

@ the bottom of the page... [/QUOTE]

The page will not load.

I ran the torture test for the 2560k FFT for about 16 hours (I believe using the multi-threaded code), and no problems. Temperature shouldn't be an issue, either. Running at stock settings, Rosewill RCX-Z775-EX 92mm 2 Ball CPU Cooler, two 120mm intake fans, one 120mm rear outtake, and one 150mm top outtake. I'm going to try running a few actual double-checks (21M) multithreaded, and maybe do an early double-check of a 2560k FFT exponent (one that I can confirm was done single-threaded and with no errors). If I can't recreate any anomalies, then I guess I'll just write it off as "one of those things" and go back to my usual single-threaded testing.

Glucas on Nehalem
 

[QUOTE]Saved Interim file at iteration 31000000. Res64: 41ACAB138FCFFBB4.[/QUOTE]Since Nehalem is not in GA, I will provide no information about its speed.
However, it is really fast !!!
And the scalability between 1 and 8 cores is very good: 7.51 .
I confirm that the verification ends at: ~8:45pm France time (~1.45pm USA CST) tomorrow (Friday 12th).
I did not start the run on Nehalem from scratch. I used a save file from Itanium and put it on Nehalem. Guillermo wrote Glucas so that the saved file is compatible with many architectures. I've checked that the next residues are the same both on Itanium2 and on Nehalem.
Tony