Redoing factoring work done by unreliable machines

[GP2]

Originally Posted by chalsall

As I said before, I did that kind of retesting myself. And I never found a cheater.

Put PLEASE! Rerun that test. Spend that money.

We will all be *very* interested if you can find a cheater. I didn't.

I remember reading an old thread where a particular machine was discovered to have found far fewer factors than expected. People here organized a retesting spree, and found a bunch of missing factors.

The person was not suspected of actually cheating because they had a bunch of other machines which did find the expected number of factors. It was concluded that it must have been bad hardware.

Anyone remember what I'm referring to? I could probably google it if sufficiently motivated...

[chalsall]

Originally Posted by GP2

I remember reading an old thread where a particular machine was discovered to have found far fewer factors than expected. People here organized a retesting spree, and found a bunch of missing factors.

I don't remember that event.

Might you be able to provide additional specific details?

[tha]

First of all, I am not looking for cheaters, I only care about error rates. We know that 2-4% of all LL residues are wrong due to overheated CPU's, overtweaked memory speeds etc. etc.

I redid a lot of P-1 work in the range 12M to 13M some time ago and found quite a lot more factors that should have been found through trial factoring or previous P-1 runs than I expected beforehand. Even though of course most factors were found then because of higher B1 and B2 bounds.

I personally attach a little more value to a factor found that was earlier missed than a new factor.

The queries that I would like to see running against the database would sniff out user/machine combinations with the highest ratio of missed factors during trial factoring and/or P-1 runs that should have been found with the settings used, and were later found by other machines through whatever method.

If we have a top ten or twenty list of suspect machines and factoring work they have done that has not been verified, I would like to redo the P-1 work or trial factoring from that list. But if others what to take portions of that list, that fine, provided the machines used are proven reliable.

[petrw1]

Originally Posted by chalsall

I don't remember that event.

Might you be able to provide additional specific details?

I participated in this event ... I have data that suggests it was January 2015 or so.
I have more data that suggest I found quite a few missed factors. I was testing in the 72-77M Range.

This was for work done by sannerud:
http://www.mersenne.org/report_expon...exp_hi=&full=1

[chalsall]

Originally Posted by petrw1

This was for work done by sannerud:
http://www.mersenne.org/report_expon...exp_hi=&full=1

Please bring this data forward.

[petrw1]

Originally Posted by chalsall

Please bring this data forward.

Sorry mis-worded. Not suggesting I am hiding anything.

I just have an old spreadsheet from that time period where I was using my GPU to re-factor a bunch of exponents from sannerud.com.

I found 23 factors....not sure out of how many but I spend almost 3 days with my GTX-750 running 500 Ghz/day

I thought it was Madpoo again that provided the lists of exponents to re-test but I have yet to find the forum posts.

[science_man_88]

Originally Posted by petrw1

I just have an old spreadsheet from that time period where I was using my GPU to re-factor a bunch of exponents from sannerud.com.

that website and you bring up the user TJAOI thread as the post you mention the site at.

meanwhile just that site without just you brings up this list of threads:

Redoing factoring work done by unreliable machines
OFFICIAL "SERVER PROBLEMS" THREAD
Newer milestone thread
fondled a factor? Turn yourself in to become inanely famous!
User TJAOI
More missed factors
Unexpected Factor
strange factors distribution??

[VBCurtis]

Originally Posted by chalsall

That assumes the person you are retesting was cheating.

As I said before, I did that kind of retesting myself. And I never found a cheater.

Put PLEASE! Rerun that test. Spend that money.

We will all be *very* interested if you can find a cheater. I didn't.

No, this addresses the question in this thread- if someone ran a bunch of TF on a known-bad computer, where known missed factors exist (as discovered by TJAOI), then an error rate can be estimated and optimal-recheck-effort can be imputed. I said nothing about cheating, and your fixation on that is your own slant, not mine.

[tha]

Originally Posted by GP2

You mention TF:63-64 factors uploaded by TJAOI today (Nov 1). Below is the list of all such exponents I see currently in the "cleared" listing, all uploaded by TJAOI on Oct 31 at 20:53.

However, this list doesn't match your description. Most of these exponents are > 100M and presumably done by GPU rather than mprime, more recently than ten years ago, and the "culprits" are all sorts of people, not "1997rj7". Can you clarify which exponents are involved? Perhaps they were uploaded a couple of days ago rather than today, and have now disappeared off the "cleared" listing.

The three exponents in the 66M range were done by one and the same user. Other users have multiple hits too.

Once again, I only care about identifying unreliable machines that have added multiple results to the database that were not verified one way or the other.

[Gordon]

Originally Posted by VBCurtis

No, this addresses the question in this thread- if someone ran a bunch of TF on a known-bad computer, where known missed factors exist (as discovered by TJAOI), then an error rate can be estimated and optimal-recheck-effort can be imputed. I said nothing about cheating, and your fixation on that is your own slant, not mine.

Seems like chalsall is having a bad week this week, see the completely unnecessary and over the top response here...


..the evidence was provided and no apology was forthcoming

[GP2]

Originally Posted by chalsall

I don't remember that event.

Might you be able to provide additional specific details?

It was in the middle of the "User TJAOI" thread, starting around page 14, and continuing for multiple pages. A bunch of people rechecked exponents by user sannerud.com

PS,

In the list of exponents I posted above, some TF:63-64 factors in the 500M range were missed by some guy named "For Research", for instance 532139263 and 532268537 and others. Looks like we found a cheater!