[QUOTE=Mini-Geek;445125]Prime Pages does run a double check of submissions. However, this is not the preferred way to weed out false positives (they don't want the Prime Pages servers testing junk, so they have some semi-strict rules about cutting you off if you submit too many composite numbers). Before submitting a prime, you should check it twice.

Most of the time this doesn't really matter, because a number generally only is reported as prime if it is prime. Sometimes, a computer glitch makes it report that it's prime when it's not. That's what happened this time. Running it a second time, especially on a different computer, significantly reduces the possibility of a false positive.[/QUOTE]

And how do they know about the double check if you only reporting a prime? Thats odd.

[QUOTE=paulunderwood;445123]1. It does not waste The Prime Pages resources.

2. It saves false congratulations.
[/QUOTE]
3. It saves from being on Caldwell's raised-eyebrow list. Trust me - you don't want to be on this list. [SPOILER]Speaking from experience, I've been there (but for a [I]wrong[/I] reason); so -- I know how it feels. In my case, the 'composite' was after all... a prime![/SPOILER]

[QUOTE=Batalov;445127]3. It saves from being on Caldwell's raised-eyebrow list. Trust me - you don't want to be on this list. [SPOILER]Speaking from experience, I've been there (but for a [I]wrong[/I] reason); so -- I know how it feels. In my case, the 'composite' was after all... a prime![/SPOILER][/QUOTE]

[url]http://primes.utm.edu/primes/page.php?id=122375&deleted=1[/url] has the Res64, if someone is willing to triple check for peace of mind... :smile:

On 28 Dec 2014 I posted this at your forum. It was a warning.

[url]http://srbase.my-firewall.org/sr5/forum_thread.php?id=91&postid=345#345[/url]

[QUOTE=rebirther;445126]And how do they know about the double check if you only reporting a prime? Thats odd.[/QUOTE]

They don't. They only know if you submit a composite number. Which will happen far too often if you submit "primes" from computers that may or may not be stable without first doing a double check (which Carlos's suggested quorum of two setting will accomplish).

I (try to) run all of the primes I find with LLR through pfgw before posting them anywhere. This provides some confidence that:

1. There were no software issues with LLR
2. There were no hardware issues on my systems

If you make a few small efforts to ensure that you post accurate results, you shouldn't have any problems in the future.

Iam testing the number now. I cant find any warnings or round off errors in the results file so this is odd. I have already tested all these warnings again for other results on my computer or SRBase to double check these files.

Reb-
The longer the test, the higher the chance of a weird/random hardware error. I've posted a couple of false primes to top5000, learned the same lesson you've learned today. The first time a new machine reports a prime, I re-run the test on a different machine (I'm not paranoid enough to use other software- I fear hardware errors rather than software mistakes).

If the prime is false, I re-run a few days worth of that machine's tests to check residues and find out if it was a one-time freak error, or an unreliable machine. Naturally, since your project has submissions from many contributors, you don't have this same ability to track and confirm submissions.

If you do continue to post primes to top-5000 without doublechecks, I suggest you wait to announce them here until top5000 confirms them. If I were in your shoes, I'd double-check all megadigit primes and submit the rest as-found.

Alas,
[CODE]PRP test of 29*2^8727880-1 using FMA3 FFT length 512K, Pass1=256, Pass2=2K, 4 threads
...
[Work thread Oct 15 15:32] Iteration: 8720000 / 8727884 [99.90%], ms/iter: 0.855, ETA: 00:00:06
[Work thread Oct 15 15:32] 29*2^8727880-1 is not prime. RES64: 9D673329E37E4DFD. We8: 8B9D3ABA,00000000[/CODE]That's PRP in base 3 / mprime...
will run base 5 and larger FFT size now... done.
[CODE][Work thread Oct 15 15:41] Starting 5-PRP test of 29*2^8727880-1 using AVX FFT length 576K, Pass1=384, Pass2=1536, 4 threads
[Work thread Oct 15 18:14] 29*2^8727880-1 is not prime. (5-PRP) RES64: 68E874083F04FC7F. We8: 8B9D3ABA,00000000[/CODE]

I am sorry that the number was not prime.

In response to some overly harsh words from a couple of people here, Reb is a BOINC CRUS administrator. I am a CRUS and NPLB administrator. Never have I nor any co-admin of either of the projects double checked anyone's prime before they submitted it to the top-5000 pages nor have I known anyone being chastised in such a manner for failing to double-check that a prime was really prime before submitting it there. There have been many primes submitted to top-5000 by people that I am not familiar with. 99+% of them are confirmed as prime. Occassionaly the bad one slips through. If I noticed that one person submitted several bad "primes" in short order, I would insist that he have his primes double-checked before submitting them in the future. Regardless, it is not up to the administrator of a project to double-check EVERY prime before it is submitted to the top-5000...that is UNLESS the administrator knows that is from a person who is not to submit bad results. Reb has indicated that he was not aware that the person who submitted the prime had bad results in the past.

Reb, as Curtis wisely suggested I would second the notion that you double-check only primes of more than 1-million digits and/or use a quorum of 2 on your servers before submitting it to top-5000 and/or posting it here. As for smaller primes, it would be a far bigger waste of your resources to double-check every smallish top-5000 prime that is truly prime 99+% of the time (and then yet again have the top-5000 site effectively triple check them) then to have the top-5000 site double-check the occassional 1-% of the composites that happen to slip through. It is to be expected. If prime submissions to top-5000 were 100% accurate there would be no reason for the site to double-check them itself.

[CODE]29*2^8727880-1 is not prime. LLR Res64: 0B55843E108EED8C Time : 33832.751 sec.[/CODE]

This was done on my i7-4930k and match the result from Caldwell, so its not prime. I will let run the user the test again.