User Xolotl has an "LL Success"

[NBtarheel_33]

Looking at the latest hourly Top 500 LL Producers Report, the user Xolotl is not only making a nice run up the leaderboard, but also has reported one LL "success" (i.e. a new Mersenne prime) in 1,877 LL attempts.

Has Santa arrived early this year, or is this some false-positive coal in our stocking?

[Dubslow]

Hasn't this user turned in several other false positives?

[ixfd64]

I wouldn't hold my breath: http://mersenneforum.org/showthread.php?t=21625

[Zr40]

Could it be M74551753? It's showing as C though, and someone else has already been assigned this exponent as a first-time check.

[Madpoo]

Originally Posted by NBtarheel_33

Looking at the latest hourly Top 500 LL Producers Report, the user Xolotl is not only making a nice run up the leaderboard, but also has reported one LL "success" (i.e. a new Mersenne prime) in 1,877 LL attempts.

Has Santa arrived early this year, or is this some false-positive coal in our stocking?

That user has submitted many false positives over the past 6+ months. I wouldn't make anything of it.

Up until now, all of his false positives have been (curiously) from a double-check. This is the only time a first-time check of his has resulted in a (presumably) false positive.

There seems to be a funny difference in what counts as a "success" in first-time and double-check top producer reports. In DC, a success seems to be a matching residue, so they happen all the time. All of xolotl's previous false positives while doing DC would have been lost in there. In the first-time check, apparently it's only "is prime" results that chalk up a "success" notch.

Anyway, I'm pretty sure this is another false alarm but I'm doing a check anyway. Unfortunately the count of "success" or not comes from what the user reported, so even when it turns out to be wrong, it'll still show up as a success in that report. Oh well.

[Madpoo]

Originally Posted by Zr40

Could it be M74551753? It's showing as C though, and someone else has already been assigned this exponent as a first-time check.

That's not the one that made his "success" count show up as "1".

In fact, besides the false positives from this user, he also has a number of "false not-quite-positives". Those are the ones where the partial residue is zero (like that one) but it said it was composite. Why or how that would happen, we're not sure.

[S485122]

Originally Posted by Madpoo

...
In fact, besides the false positives from this user, he also has a number of "false not-quite-positives". Those are the ones where the partial residue is zero (like that one) but it said it was composite. Why or how that would happen, we're not sure.

I suppose PrimeNet has data about that users hardware, program and its version ?

Jacob

[retina]

[conspiracy]

Oh yes, "probably a hardware problem". How can we be sure that it isn't all just a ruse to throw us off the scent and in fact a real discovery has been made? Hmm? HMM?

Oh, and the Moon landings were faked, and 911 was an inside job, also! We know!!



[/conspiracy]

[Madpoo]

Originally Posted by S485122

I suppose PrimeNet has data about that users hardware, program and its version ?

Jacob

George and I have actually emailed with him about this and there is some unusual hardware installed on these boxes, but nothing jumped out as being a real problem. I don't know if this is 100% true, but I think in all cases these boxes have a Xeon Phi (Knights Corner, not Landing) coprocessor card installed.

Maybe there's a driver issue causing some interference, or the system gets too hot inside which creates CPU problems.

Anyway, Prime95 is running on Xeon E5-26xx processors, same as I do and I haven't had any issues.

It would be annoying except for the fact that he also submits a lot of good double-checks, so... whatever.

Meanwhile, this did indeed turn out to be another false positive.

If someone wants to do a double-check of my result, this is the exponent:
M76849343

[ewmayer]

Originally Posted by Madpoo

George and I have actually emailed with him about this and there is some unusual hardware installed on these boxes, but nothing jumped out as being a real problem. I don't know if this is 100% true, but I think in all cases these boxes have a Xeon Phi (Knights Corner, not Landing) coprocessor card installed.

Maybe there's a driver issue causing some interference, or the system gets too hot inside which creates CPU problems.

Anyway, Prime95 is running on Xeon E5-26xx processors, same as I do and I haven't had any issues.

It would be annoying except for the fact that he also submits a lot of good double-checks

Thanks, Aaron - 2 questions:

1. Do I understand you correctly in that some/most of the KC-installed systems return good DCs?

2. Is there any commonality among the user's systems which have returned false positives (I.e. some specific and hopefully-small subset of the KC-installed systems), or is it a case of apparently sporadic glitches, in which a given system returns a bunch of good DCs, then a bad one, then reverts to good ones?

[Madpoo]

Originally Posted by ewmayer

Thanks, Aaron - 2 questions:

1. Do I understand you correctly in that some/most of the KC-installed systems return good DCs?

2. Is there any commonality among the user's systems which have returned false positives (I.e. some specific and hopefully-small subset of the KC-installed systems), or is it a case of apparently sporadic glitches, in which a given system returns a bunch of good DCs, then a bad one, then reverts to good ones?

Hmm... I don't remember for sure. I'll have to refresh my notes on that. It seemed like it might be the case that only the systems with Phi copros were having issues, but I don't know if *all* such systems had errors, or just a subset. My memory on that could be wrong though; we were only casually discussing it via email and that was several months back.