50th Mersenne prime???

[Miszka]

Maybe prime?
http://www.mersenne.org/report_expon...6092673&full=1

[Mark Rose]

Originally Posted by Miszka

Maybe prime?
http://www.mersenne.org/report_expon...6092673&full=1

Xolotl has had quite a few bad results before. The last two hexits are probably not 00.

I've put it at the beginning of the queue on my fastest machine.

[ATH]

Yeah the last 2 digits are almost certainly 02. If it was 00 the server would have masked the entire residue.

If a computers ram get zero'ed but some hardware error, the iterations will be 0,-2,2,2,2,2,2,2.... and it will end in 00000....00002.

[Miszka]

Originally Posted by ATH

Yeah the last 2 digits are almost certainly 02. If it was 00 the server would have masked the entire residue.

If a computers ram get zero'ed but some hardware error, the iterations will be 0,-2,2,2,2,2,2,2.... and it will end in 00000....00002.

That is why I ended title of thread "???"

[CRGreathouse]

Anyone want to do a Bayesian calculation here?

[Madpoo]

Originally Posted by ATH

Yeah the last 2 digits are almost certainly 02. If it was 00 the server would have masked the entire residue.

If a computers ram get zero'ed but some hardware error, the iterations will be 0,-2,2,2,2,2,2,2.... and it will end in 00000....00002.

This user has had some issues lately with a subset of machines. I've mentioned this in another thread but it was probably buried in some other conversation.

The issue manifests itself as a large # of false positives (residue = 0 but a double-check shows that was not the case) or even stranger, it sometimes has a residue of zero but says it was composite.

Well, the *partial* residue is zero, but since it says it was composite we have to assume it wasn't entirely zero.

As far as we can tell it's running stock mprime, latest 28.9 build, and it's on a server running a Xeon with (I assume) ECC. I'm guessing it's some very specific case related to whatever hardware itself.

In the past several months I think those machines have turned in about 20-30 false positives or "zero residue" composites, many of which I've run verification runs on. It was to the point where I've stopped bothering with them and just let them get picked up by others. In the cases where it did a *double* check and reported a false positive, it'll get picked up again shortly for a triple-check. For the few first time tests it does with the same bizarre result, I think George has been marking them suspect so they get reassigned as a new first time check.

[ewmayer]

Originally Posted by Madpoo

This user has had some issues lately with a subset of machines.

Any luck contacting the user about the problematic machines?

Mark, what is ETA on your DC? (And what's the per-iter time, and running on how many cores?)

I'm still trying to get my head around what kind of data corruption could yield "zero res64, but marked composite." That implies that the low words of the full residue are zero but at least one higher-order word is not. I'd love to see the actual full-length residue for such a case, but without a reporducible hardware problem *and* access to the user's system to run some kind of debug-enhanced custom build, that will be impossible.

[Mark Rose]

Originally Posted by ewmayer

Mark, what is ETA on your DC? (And what's the per-iter time, and running on how many cores?)

4 cores, 5.32 ms/iter. 3d 16:21 to go as I post this. 21.44% done already.

[kladner]

Originally Posted by Mark Rose

4 cores, 5.32 ms/iter. 3d 16:21 to go as I post this. 21.44% done already.

w00t! Impressive. Is this a Xeon? 13.57 ms for a GTX480 at 848 MHz.

[Mark Rose]

Originally Posted by kladner

w00t! Impressive. Is this a Xeon?

Hardly. i7-4770k at 3.7 GHz with DDR3-2400. I should try overclocking it someday.

[kladner]

Originally Posted by Mark Rose

Hardly. i7-4770k at 3.7 GHz with DDR3-2400. I should try overclocking it someday.

I long to upgrade to Intel, (from FX8350). But even cheap X99 boards are expensive, by my standards. Wanting a 5820K doesn't help.