processed dc and tc posts

[Uncwilly]

Originally Posted by ChZ

Requesting LL triple-check:

DoubleCheck=109512329,77,1

The latest result is reliable.

DC'ing your own LL test is frowned upon. And how can you be sure that the old one is not correct?

[kriesel]

Originally Posted by Uncwilly

DC'ing your own LL test is frowned upon. And how can you be sure that the old one is not correct?

Second one was probably with ECC ram.
https://www.mersenne.org/report_LL/?...p_hi=109512329
https://www.intel.com/content/www/us...fications.html
Both were prime95, but different versions; neither shows an error code.
It would have been better if the second run had been PRP/GEC/proof-gen, by a software version that supports it.
And by different user.
By different software.

Took:
DoubleCheck=109512329,77,1

@ChZ:
please upgrade prime95 instances to ~v30.8b17 where practical.
Do you have any interim res64s to share by PM from either previous LL run? (10M or 20M spacings would be fine)

[kriesel]

Originally Posted by ChZ

The latest result is reliable.

Confirmed first bad, second good.
https://www.mersenne.org/report_expo...=1&swversion=1
Reasoning provided by PM appears confirmed.

[Uncwilly]

Originally Posted by kriesel

It would have been better if the second run had been PRP/GEC/proof-gen, by a software version that supports it.

333043493
PRP=1,2,333043493,-1

[Uncwilly]

Just a note, there is a new TC candidate. Also, I have been checking every few days the list for TC and the 1LL+1PRP list. Both have been zero entries of late. Once we get those SDC's done, we will be in a glorious era.

[kriesel]

Originally Posted by Uncwilly

PRP=1,2,333043493,-1

Originally Posted by Uncwilly

333043493
PRP=1,2,333043493,-1

Originally Posted by kriesel

It would have been better if the second run had been PRP/GEC/proof-gen, by a software version that supports it.

But not in the case of a little research effort on LL reliability at 100Mdigits, of which this exponent is a part. I guess enough time has passed that Uncwilly has forgotten that he is also participating in that effort on another exponent, or forgot or chose not to check his script's output against that short list he's been notified of before.

Consider 333043493 taken, as LLTC which I had already under way on Gpuowl & Radeon VII while prime95 progress was blocked earlier. (I'll queue it in a prime95 worktodo shortly.)
See https://mersenneforum.org/showpost.p...1&postcount=29 or here or here or previous discussion (in this thread at the time, since relocated) for why LL in just a few specific documented cases.
Also see https://mersenneforum.org/showpost.p...&postcount=862 and related posts in that thread for the LLDC run issue, entirely conducted on ECC ram systems.
I'm still cleaning up after my own runs: https://mersenneforum.org/showpost.p...&postcount=803

[Uncwilly]

Originally Posted by kriesel

But not in the case of a little research effort on LL reliability at 100Mdigits, of which this exponent is a part. I guess enough time has passed that Uncwilly has forgotten that he is also participating in that effort on another exponent, or forgot or chose not to check his script's output against that short list he's been notified of before.

We know that there is a high error rate on the 100M LL tests. That is why DC's should be PRPs. An LL TC against one of your machine's DC is a better bet than an LL DC against any random FTC. My current machines have had a perfect record of LL tests, for DC's, TC's, or QC's. Therefore I picked up that exponent that you linked to.

I think that the point that there were a lot of errors in the initial 100M LL's has been well proven.

[ric]

This time for a QC: requesting (as always, LL only, please) for this one:

DoubleCheck=66478897,75,1

TIA

[sdbardwick]

Originally Posted by ric

This time for a QC: requesting (as always, LL only, please) for this one:

DoubleCheck=66478897,75,1

TIA

Reserved and in progress. ETC 50 hours.

[kriesel]

Originally Posted by Uncwilly

Therefore I picked up that exponent that you linked to.

I think that the point that there were a lot of errors in the initial 100M LL's has been well proven.

If you are referring to 333043493, you are poaching. I have that exponent's reassignment. Perhaps you meant 332216777.

If I remember correctly, it was you that criticized the thin statistical basis of attachment 3 of https://www.mersenneforum.org/showpo...6&postcount=14 on the basis of too-small sample size.
(Or perhaps it was about polls. Or both.)

Today, ~21 months later, for https://www.mersenne.org/report_LL/?...2192810&unv=-1
52 exponents verified & not factored, with 117 tests total, up to 4 per exponent;
and 15 listed as bad;
so 15 bad of 132 tests of known status; 11.36% bad, compared to the 13.7% found in June 2021 with smaller sample size (7 bad, 44 matching tests, 21 exponents), both cases of which exclude exponents with only suspect or conflicting residues requiring a TC or higher to resolve which residues are wrong.
We have slightly better statistics now, but still +-5.4% for 95% confidence interval per
https://calculator-online.net/margin...or-calculator/

[Uncwilly]

Originally Posted by kriesel

If you are referring to 333043493, you are poaching. I have that exponent's reassignment. Perhaps you meant 332216777.

The attachment will speak for itself.