Pépin tests of Fermat numbers beyond F24 - Page 6

ewmayer · 2019-05-03, 19:51

Has anyone heard from David Stanfill, a.k.a. airsquirrels, of late? I had my 2 side-by-side tests of F30 running on a pair of servers he hosts, one the GIMPS KNL (doing the run @64M FFT, last check ~70% done) another a 32-core Intel Xeon server (doing the run @60M FFT). The Xeon started its run some months after the KNL but was running faster (~55 ms/iter versus 68 ms/iter) and had pulled ahead of the KNL early this year. But I lost ability to ssh into it (from the KNL is only option, the way David configured things) a couple months ago, sent David several e-mails to that effect, no reply. And as of several days ago I can no longer access the KNL, either. I'm starting to get a bit worried - his forum profile indicates last activity back in January.

paulunderwood · 2019-05-03, 20:04

Quote:

Originally Posted by ewmayer

Has anyone heard from David Stanfill, a.k.a. airsquirrels, of late? I had my 2 side-by-side tests of F30 running on a pair of servers he hosts, one the GIMPS KNL (doing the run @64M FFT, last check ~70% done) another a 32-core Intel Xeon server (doing the run @60M FFT). The Xeon started its run some months after the KNL but was running faster (~55 ms/iter versus 68 ms/iter) and had pulled ahead of the KNL early this year. But I lost ability to ssh into it (from the KNL is only option, the way David configured things) a couple months ago, sent David several e-mails to that effect, no reply. And as of several days ago I can no longer access the KNL, either. I'm starting to get a bit worried - his forum profile indicates last activity back in January.

Maybe you can find out where he is through this site:

http://press.airsquirrels.com/2019/0...research-labs/

I guess he is really busy!

ewmayer · 2019-05-03, 20:09

Quote:

Originally Posted by paulunderwood

Maybe you can find out where he is through this site:

http://press.airsquirrels.com/2019/0...research-labs/

I guess he is really busy!

Thanks - I knew well how busy he was, but he never failed to reply to multiple e-mails over such a long span previously. (Including one more "you are still among the living, I trust?" missive last night). And I hate to nag, really don't want to have to resort to trying to reach him via phone.

paulunderwood · 2019-05-03, 20:17

Quote:

Originally Posted by ewmayer

Thanks - I knew well how busy he was, but he never failed to reply to multiple e-mails over such a long span previously. (Including one more "you are still among the living, I trust?" missive last night). And I hate to nag, really don't want to have to resort to trying to reach him via phone.

I did a "whois squirrelsresearch.com" and the closest I could get is squirrelsresearch.com@domainsbyproxy.com You could also try this form:
http://www.squirrelsresearch.com/contact/

HTH

ewmayer · 2019-05-03, 20:55

Quote:

Originally Posted by paulunderwood

I did a "whois squirrelsresearch.com" and the closest I could get is squirrelsresearch.com@domainsbyproxy.com You could also try this form:
http://www.squirrelsresearch.com/contact/

HTH

Just send a missive via the contact form - thanks!

Tooryalai · 2019-05-03, 23:21

Here’s the proof that there are no more Fermat Primes:

The following website calculates the total number of integers that equal a given totient number:

http://www.numbertheory.org/php/carmichael.html

Plug in n = 2k, k ≥ 33

Plug in e = 0

Plug in f = 0

All the solutions for n = 2k (k ≥ 33), contain exactly 32 solutions.

All 32 solutions are even and are equal to 2 * the solutions for the previous n.

Because there are no odd numbers, and that all subsequent solutions will just be a power of 2 times any of the 32 previous solutions, then there can’t be any more prime numbers.

Prime numbers are odd, and there’s only even solutions for n = 2k (k ≥ 33).

Thererfore, there are no more fermat prime numbers.

LaurV · 2019-05-04, 05:10

Disclaimer: "2k" in your post should be read as 2^k.

Do you understand what that calculator does, and how?

ewmayer · 2019-05-05, 21:54

Still nary a peep from David ... I think I need to start looking for a new Xeon server to finish my F30 runs on ... what is the cheapest 32-core avx2-capable Xeon server in the several-years-old used-hardware market, I wonder. The other problem is that while I have recent local-download copies of the run-status file and every-10M-iter savefiles for the 64M-FFT run, my 60M-FFT run-status file is 5-6 months out of date, i.e. I have no proof that I did the last ~150M iters or so at that FFT length. Of course one can use the 10Miter savefiles to do a sped-up parallel verification (multiple machines, each reproducing a separate 10Miter sbinterval), but rather a spot of bother in any event.

GP2 · 2019-05-07, 00:45

Before Robert Gerbicz came up with Gerbicz error checking for PRP tests for Mersenne numbers, I think he invented something similar for Pépin tests.

So I wonder if it would make sense to pause the tests on F30 in order to incorporate error checking, to increase the confidence in the final result.

ewmayer · 2019-05-09, 03:25

Quote:

Originally Posted by GP2

Before Robert Gerbicz came up with Gerbicz error checking for PRP tests for Mersenne numbers, I think he invented something similar for Pépin tests.

So I wonder if it would make sense to pause the tests on F30 in order to incorporate error checking, to increase the confidence in the final result.

Good suggestion, but in my case the whole point of doing side-by-side runs at FFT lengths 60M and 64M, with the hardware/FFT-length matched up to give similar throughput, was to be able to do regular cross-checking of interim residues. Before I lost access to the 32-core Avx2-Xeon system that run was perhaps 2-3% ahead of the KNL run, i.e. a few tens of millions of iterations.

I've had a generous offer from a fellow forumite to use a 32-core Skylake Xeon system he owns (or has access to) to finish F30 - depending on what his detailed tuning-run timings looks like, we may be able to use 16 cores for each of 2 runs at 60M and 64M, respectively, and get throughput not much less than I was getting before. The only issue is the missing run-status file documentation of the 60M run, where the last copy I ftp'ed over to my PC is iter ~530M, roughly 200Miter behind the last check of that run before the Avx2 Xeon went AWOL. I may simply start both 60M and 64M runs from the 740Miter savefile produced by the 64M run (last I copied to my PC before the KNL went AWOL), and hope to reach someone at Squirrels Inc who can help me recover the missing files between now and end of the year. Failing that I can re-do the missing 200Miter starting with the 530Miter savefile at a later date in order to fill in the gap.

Xyzzy · 2019-05-09, 15:01

For future runs, started from scratch, would Gerbicz error checking eliminate the need for parallel runs?

2019-05-03, 23:21	#61
Tooryalai May 2019 2 Posts	NO MORE FERMAT PRIMES Here’s the proof that there are no more Fermat Primes: The following website calculates the total number of integers that equal a given totient number: http://www.numbertheory.org/php/carmichael.html Plug in n = 2k, k ≥ 33 Plug in e = 0 Plug in f = 0 All the solutions for n = 2k (k ≥ 33), contain exactly 32 solutions. All 32 solutions are even and are equal to 2 * the solutions for the previous n. Because there are no odd numbers, and that all subsequent solutions will just be a power of 2 times any of the 32 previous solutions, then there can’t be any more prime numbers. Prime numbers are odd, and there’s only even solutions for n = 2k (k ≥ 33). Thererfore, there are no more fermat prime numbers.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
P-1/P+1 on Fermat numbers	ATH	Operazione Doppi Mersennes	2	2015-01-25 06:27
What are the Primality Tests ( not factoring! ) for Fermat Numbers?	Erasmus	Math	46	2014-08-08 20:05
LLT numbers, linkd with Mersenne and Fermat numbers	T.Rex	Math	4	2005-05-07 08:25
Two Primality tests for Fermat numbers	T.Rex	Math	2	2004-09-11 07:26
Fermat Numbers	devarajkandadai	Math	8	2004-07-27 12:27

2019-05-03, 19:51	#56
ewmayer ∂²ω=0 Sep 2002 República de California 2DDC₁₆ Posts	Has anyone heard from David Stanfill, a.k.a. airsquirrels, of late? I had my 2 side-by-side tests of F30 running on a pair of servers he hosts, one the GIMPS KNL (doing the run @64M FFT, last check ~70% done) another a 32-core Intel Xeon server (doing the run @60M FFT). The Xeon started its run some months after the KNL but was running faster (~55 ms/iter versus 68 ms/iter) and had pulled ahead of the KNL early this year. But I lost ability to ssh into it (from the KNL is only option, the way David configured things) a couple months ago, sent David several e-mails to that effect, no reply. And as of several days ago I can no longer access the KNL, either. I'm starting to get a bit worried - his forum profile indicates last activity back in January.

2019-05-04, 05:10	#62
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 11·911 Posts	Disclaimer: "2k" in your post should be read as 2^k. Do you understand what that calculator does, and how?

2019-05-05, 21:54	#63
ewmayer ∂²ω=0 Sep 2002 República de California 2²×5×587 Posts	Still nary a peep from David ... I think I need to start looking for a new Xeon server to finish my F30 runs on ... what is the cheapest 32-core avx2-capable Xeon server in the several-years-old used-hardware market, I wonder. The other problem is that while I have recent local-download copies of the run-status file and every-10M-iter savefiles for the 64M-FFT run, my 60M-FFT run-status file is 5-6 months out of date, i.e. I have no proof that I did the last ~150M iters or so at that FFT length. Of course one can use the 10Miter savefiles to do a sped-up parallel verification (multiple machines, each reproducing a separate 10Miter sbinterval), but rather a spot of bother in any event.

2019-05-07, 00:45	#64
GP2 Sep 2003 13·199 Posts	Before Robert Gerbicz came up with Gerbicz error checking for PRP tests for Mersenne numbers, I think he invented something similar for Pépin tests. So I wonder if it would make sense to pause the tests on F30 in order to incorporate error checking, to increase the confidence in the final result.

2019-05-09, 15:01	#66
Xyzzy Aug 2002 3²×23×41 Posts	For future runs, started from scratch, would Gerbicz error checking eliminate the need for parallel runs?