Help doing some quadrup1e+ checks

[guido72]

Originally Posted by Brian-E

Unless you are able to contact Sergey Nosov and arrange things, I don't think you can ethically do that work (unless and until the assignment completes or expires).

Precisely. I was just offering some spare cycles to do some targeted work, if needed... and without contraindications of any sort...

[VictordeHolland]

My apologies to the people who 'lost' GPU hours because I uploaded my results earlier then them. If you care about the TF credits you can PM me and I'll run some TF that they can report.

For the record: I'm still running some TF and P-1 on the second exponent.

[bloodIce]

I do not live for TF credit, but as we talk, I lost some in the beginning of 2M range, due to your excursion to TF65 . As a compensation I demand that you complete the expos to TF66 (you know the first 196 expos in 2M range) . No, seriously I do not care, but if you are willing to extent 2M's a bit more, please do, I will finish all to TF68, but my GPU is slower than yours. So please help, since you started there. By the way I am running Pminus1 for the second candidate with B1=1300000, B2=130000000. What are your B1s, B2s?

[VictordeHolland]

Originally Posted by bloodIce

I do not live for TF credit, but as we talk, I lost some in the beginning of 2M range, due to your excursion to TF65 . As a compensation I demand that you complete the expos to TF66 (you know the first 196 expos in 2M range) . No, seriously I do not care, but if you are willing to extent 2M's a bit more, please do, I will finish all to TF68, but my GPU is slower than yours. So please help, since you started there. By the way I am running Pminus1 for the second candidate with B1=1300000, B2=130000000. What are your B1s, B2s?

I was running P-1 with B1=2,000,000 and B2=40,000,000 but I've stopped it now, so we don't duplicate even more work.

Do you have more work queued/assignments in the 2M range that I should be aware of? I'm currently doing 2.0M-2.2M to 65 and after that 2.7M-2.9M to 65bits.

[bloodIce]

No, I am running 2M from TF65 to TF68 (slowly). So you can submit your 2M TF65 work without any concern. What I was asking is to help extend 2.0-2.09999M to TF66 (or if you wish to TF68). My game in 2.0-2.1M range is to see if we complete everything to TF68, also ECM's first level (B1=50000) and some extended Pminus1 (B1=1e08, B2=5e09) how many will survive. However, I do not mind some help (just to coordinate the effort in order not to do double work). I played this game for 2 years now, I expect to finish in 5 years.

[TheMawn]

With enough mismatched residues, does it become a good idea to triple-check as well? Maybe with a safer FFT or something if that might have been what caused the excessive mismatches?

[Madpoo]

Originally Posted by bloodIce

I would not do this if I were you. I lost many hours of TF-work on 47540719. Straight after I asked my question I put the number for TF, but before I woke up, "my work" was reported already to the server. Now, exponent 47540719 is among the most wanted factored and I assume we will have factor soon. This will make your work pointless. Of course, if we do not manage to factor it, a matching LL-residue is needed, but it is already assigned to "Sergey Nosov", who might be ahead of you.

@Madpoo: I would suggest that you give us a list of such problematic exponents. I think that the most of them can be factored in reasonable time. 38419327 was easier to factor, than to verify by LL-test. How many suspected cases you have?

I've pretty much covered all of the wildly multiple LL tests (the ones where 3 runs failed to find any 2 that matched).

There's a good number out there that have had 2 mismatched runs and need a basic triple-check.

By the way, Sergey Nosov who has the assignment on 47,540,719 may have seen this thread and did a manual reservation, asking for that exponent specifically. It's far enough ahead of the current DC assignments that any other explanation seems unlikely.

That's why I was hoping if someone took it, they'd reply here to let other people know.

[Madpoo]

Originally Posted by TheMawn

With enough mismatched residues, does it become a good idea to triple-check as well? Maybe with a safer FFT or something if that might have been what caused the excessive mismatches?

It did strike me as odd when an exponent had so many different bad residues. Still, if there was some particular problem at any given FFT size for something, the test Prime95 does at the upper margins of some FFT help out, and even then the odds of 2 systems coming up with the same wrong residue are essentially nil.

There are definitely some oddities in the data... I mean, there are some winners like this one:
http://www.mersenne.org/M2397103

Of course all of the bad ones were from the same user, and there's a triple verification on it. That one is the record for most bad residues.

Other multiple bad ones with more than 4 distinct residues:
16073131 = 9 different residues
23057101 = 8 (this one should be triple-verified)
12136727 = 7
2513153 = 6
10916119 = 6
26026433 = 6
30078407 = 6 (another worthy of a triple-verification, even though that one user apparently tested it *17* times and got the same matching residue each time, besides the 5 bad ones)

2525329 = 5
5094653 = 5
7021433 = 5
26146763 = 5
41940097 = 5

If you only see one entry in the history, that means the same user provided the matching residue. They are distinct in the database with the residues and different shift-counts, but there's something happening that coalesces the history to that single one... I should look into that. But yeah, the code will NOT treat it as a valid DC if the shift-count is the same, like if you submitted the same manual result more than once.

For example: http://www.mersenne.org/M23057101

There are another 147 exponents that have 4 distinct residues, 2766 with 3 distinct residues, and a whopping 53,825 that have 2 distinct residues. Undoubtedly some of those are still in need of a matching test, but I wasn't looking for that specifically.

[petrw1]

....

[NBtarheel_33]

I'll do 2525329...

[Madpoo]

Originally Posted by NBtarheel_33

I'll do 2525329...

That's good and all, but I should have been more clear that all of those have been verified at this point. Just those 2 that I mentioned probably need an independent check since the matching residues both came from the same user.

There's another thread going on where I just started talking about doing triple checks where the first and second tests were done by the same user. Those 2 exponents I mentioned might not be listed there once I put up the list since it only includes exponents where *only* 2 results are recorded, just as a start.