False primes

[Kosmaj]

Something appears to be wrong with 127*2^841793-1, it was found to be composite by the Top-5000 verification program. Can you check your output files? I hope it's not yet another false prime caused by a software or hardware error... The link is here

[paulunderwood]

Here's my LLR 3.7 test for the number:

127*2^841793-1 is not prime. LLR Res64: D556CA9818F86A8A Time : 1133.336 sec.

[lsoule]

I get the same result as Paul:
127*2^841793-1 is not prime. LLR Res64: D556CA9818F86A8A Time : 1359.384 sec.

[VBCurtis]

I noticed the composite myself, and am double-checking it presently, though the previous two posts make that effort only to see if my error is reproducible. I have checked my results file, and I did not typo the purported prime.

I do overclock my machine, but I have not had LLR fail in many months (I backed off the settings that caused the crash a few months ago, and re-ran a few days' work just in case). I assume this is an overclocking-related error, and I'm unsure how much of my own work to trust.

I assume if the error is not reproducible, it was heat/overclocking related. I'll slow the machine down a few notches, and re-run the work done since summer arrived on the thermometer. For those who care, the CPU is a P4-1800 running at 2450 or so, with CPU die temp around 56C. I've run as high as 2520 with LLR errors one per week or so, so I figured backing off 60Mhz from sort-of-stable was enough. I'll slow it to 2400 now. Prime95 torture test runs 4+ days without error in current config.

Ugh.
So much for a lucky week.
-Curtis

[paulunderwood]

For those who care, the CPU is a P4-1800 running at 2450 or so, with CPU die temp around 56C.

That temperature seems a bit too high -- I assume it's a BIOS reading. LLR will make it soar, maybe past what is considered safe.

[VBCurtis]

My residue matches the other two above.

The temperature is from motherboard monitor, which uses the motherboard temp sensors to provide live temp updates. The CPU with LLR not running idles around 43C.

I found two instances in the past 10 days in the results file where LLR had a suminputs != sumoutputs error. Perhaps I have dust accumulation in the computer or something.. I'll repeat all work starting from a day before the first error. If anyone has a linux utility to compare the results files, please instruct or link me to how to do this (I have a linux box available, but am a rookie at any file-manipulation commands-- even "grep" is just a vague acquaintance).

I'll report any residue discrepancies I find.
-Curtis

[paulunderwood]

Use "grep", "cut" and "diff" (and "man" )

grep prime lresults.txt | cut -d: -f2 > res1
grep prime lresults2.txt | cut -d: -f2 > res2
diff res1 res2

You can adjust things slightly with "head" and "tail"

[Kosmaj]

Moving posts related to 127*2^841793-1 here from the "Post Primes" thread.

[Jean Penné]

Hi,

I also verifyed this number and found the good residue :
127*2^841793-1 is not prime. LLR Res64: D556CA9818F86A8A Time : 1687.575 sec.

(LLR 3.6.2, P4 2.53 GHz, Win2000).

However, I think I would have to do an update for the future release, to avoid
the possibility of false zero results when an FFT error occurs, as George is doing
in Prime95...

Good luck for all participants of this project!

Jean

[Flatlander]

Is it possible that we are also getting false negatives?

The same likelihood? Less? More?

[Jean Penné]

Originally Posted by Flatlander

Is it possible that we are also getting false negatives?

The same likelihood? Less? More?

Indeed, it is possible, and more probable than false positives, because one
false bit on the complete residue is sufficient to miss the prime!

On the other hand, if you made a double check, and found the same 64 bits
residue, it becomes very unlikely...

That is why double checks are recommended!

Jean