mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data

Reply
 
Thread Tools
Old 2017-07-11, 07:39   #188
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

2×3×5×7×43 Posts
Default

I wonder if there are any traces of "James N. Seed" anywhere else in the world?
Is he really that less ambitious than Don Blazys?
Batalov is offline   Reply With Quote
Old 2017-07-11, 16:05   #189
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

22·2,137 Posts
Default

My feeling is that what he doesn't really say is that you waste the clock cycles, he takes the money and the glory (in case new primes are found) From all explanations he gives, you connect to a central server which gives you work to do, you report your results, but you can't see much of what you are doing, except that you are doing something. The reticence of the people is quite understandable, I wonder if he has any "customers"... For the last version, he speaks about the fact that you can see the iteration times, in.... seconds per iteration (huh?) but doesn't mention if you can see the exponent, or what you do if you find a prime... or how to actually verify that the program reports "prime" if a prime is found (to you! not to the server) instead of reporting some crap residue (to you), and only notify the server of the fact that a prime was found. This is very tricky, and from guy's attitude, I somehow expect it.. Later on, he can easily blame a "computing error" in your computer, when he claims the prime for himself. He even has the courage to say that his "setup" does not allow the program to be run, haha... because only some single, old GPU.... WTF? how did he develop and test?

Did anybody verified any of the claims? (this is a tricky question because in fact, I only want to find out who of you guys has the five GPUs required )

Last fiddled with by LaurV on 2017-07-11 at 16:06
LaurV is offline   Reply With Quote
Old 2017-07-11, 21:18   #190
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
Rep├║blica de California

2×35×23 Posts
Default

Quote:
Originally Posted by LaurV View Post
Did anybody verified any of the claims? (this is a tricky question because in fact, I only want to find out who of you guys has the five GPUs required )
By way of a simple put-up-or-shut-up test, someone should ask him to produce the 10000-iteration LL residue for a 'small' gigadigit M-number, say M3321928109, with iter=0 being the initial LL seed of 4. (I can produce that easily by way of cross-checking, should he accept.)

If he produces the correct Res64, ask him what he ran his code on and how long that test interval took. Of course he could lie about the latter, but at least we'll have an upper bound on the actual runtime based on the time between the above challenge-posting and his posting of the result.
ewmayer is offline   Reply With Quote
Old 2017-07-11, 21:27   #191
CRGreathouse
 
CRGreathouse's Avatar
 
Aug 2006

133448 Posts
Default

Quote:
Originally Posted by ewmayer View Post
By way of a simple put-up-or-shut-up test, someone should ask him to produce the 10000-iteration LL residue for a 'small' gigadigit M-number, say M3321928109, with iter=0 being the initial LL seed of 4. (I can produce that easily by way of cross-checking, should he accept.)
I think I've exhausted my goodwill, but it would be great if someone could convince him to do something along those lines. It would at least show that he's able to do the calculations, and assuming good faith (with I'm still prepared to do, signs to the contrary notwithstanding) it would give performance bounds as well.
CRGreathouse is offline   Reply With Quote
Old 2017-07-11, 21:28   #192
CRGreathouse
 
CRGreathouse's Avatar
 
Aug 2006

16E416 Posts
Default

Quote:
Originally Posted by LaurV View Post
Did anybody verified any of the claims? (this is a tricky question because in fact, I only want to find out who of you guys has the five GPUs required )
The new version apparently only requires one GPU.
CRGreathouse is offline   Reply With Quote
Old 2017-07-11, 21:30   #193
CRGreathouse
 
CRGreathouse's Avatar
 
Aug 2006

22·5·293 Posts
Default

Quote:
Originally Posted by Batalov View Post
I wonder if there are any traces of "James N. Seed" anywhere else in the world?
Is he really that less ambitious than Don Blazys?
My guess at this point (I give it optimistically 60%) is that his program works correctly modulo possible bugs, but is much slower than existing software.
CRGreathouse is offline   Reply With Quote
Old 2017-07-12, 03:33   #194
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

28×3×5 Posts
Default fft length, iteration time relationship

Quote:
Originally Posted by Madpoo View Post
Even magically getting faster as the size of the exponent grows (which is another of his more remarkable claims).

I mean, we all know that the larger the FFT, the slower LLR will be per iteration. And the software out there walks that fine line of finding the smallest possible FFT size without getting into the roundoff errors.
If things are working as the knowledgeable here, Knuth, etc. predict re long multiprecision multiplication or squaring, that's so. Best known is proportional to n ln n ln ln n for large n.
However.
I've doggedly benchmarked wide ranges of fft lengths in fftbench and threadbench on multiple CUDALucas versions and multiple CUDA levels by bitness (win32, x64) by multiple GPU models etc, to multidimensionally explore the space more fully than I imagine an ordinary user is likely to have done. (V2.05.1 and two 2.06 betas to date) In doing that, I've seen counter-examples to the expectation of increasing iteration time with increasing fftlength in CUDALucas benchmarking, indicating trouble. That trouble indicator corresponds to 1024 square threads not working on a given card model, CUDALucas producing bad interim residues & too quickly, side effects of incomplete/inefficient fft files or threads files etc. Some of this trouble is CUDA library level dependent. Flashjh was notified back in April, May, June as I made finds, in considerable detail. CUDALucas has been in wide use for years and yet there they are, despite many people's dedicated efforts.

I've been involved with computer programs as an author, end user, tester, etc for nearly 50 years, and as a result, no longer trust them even after I've tested them. In my opinion number theory programs are even more demanding than real time controls/automation.

If CEMPLLA has not been exhaustively tested by third parties with lots of determination and patience and imagination, and access to manual controls and documentation, it may have some similar issues as yet undetected. If it's from a common code base it is likely to share the issues of the source of that common code base as well as to have some of its own.

When you were examining it, did you get any indication of what CUDA level he linked CEMPLLA to?
Try any string matching programs that would indicate if chunks of well known code were reused?

Last fiddled with by kriesel on 2017-07-12 at 03:43
kriesel is offline   Reply With Quote
Old 2017-07-12, 04:47   #195
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

28×3×5 Posts
Default James Seed

Quote:
Originally Posted by Batalov View Post
I wonder if there are any traces of "James N. Seed" anywhere else in the world?
Is he really that less ambitious than Don Blazys?
NVIDIA Announcement, LinkedIn, and here. The LinkedIn is interesting. https://www.linkedin.com/in/jamesseed
kriesel is offline   Reply With Quote
Old 2017-07-12, 22:50   #196
airsquirrels
 
airsquirrels's Avatar
 
"David"
Jul 2015
Ohio

10058 Posts
Default

I tried really hard to even get this running on two different machines with no luck. Both crash the installer on launch if the nvidia driver is loaded, windows 10 and server 2016.
airsquirrels is offline   Reply With Quote
Old 2017-07-12, 23:03   #197
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
Rep├║blica de California

2·35·23 Posts
Default

Thanks for generously allowing your time to be wasted in this attempt, David. So, to sum up:

o Even assuming one managed to install it, there is no way to tell if the code does what it purports to, much less to do any self-tests for correctness and speed;

o Author has 0 track record w.r.to computational number theory software, yet even trivial (at least they should be for him, since he presumably is building and testing his stuff on *something*) demonstrations along the lines of "does it work?" and "is it fast enough to be of interest?" are "beneath him."

o No one around here has even managed to install it sans crashes.

I do, however, congratulate him on a master class in "trolling Teh Internetz." Having mastered the "if we can only get enough people to buy in, we will see all sorts of magical kewl stuff happen" techno-vaporware business promotion model, may we expect an Initial Coin Offering sometime soon?
ewmayer is offline   Reply With Quote
Old 2017-07-12, 23:16   #198
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·3·1,499 Posts
Default

Quote:
Originally Posted by ewmayer View Post
I do, however, congratulate him on a master class in "trolling Teh Internetz." Having mastered the "if we can only get enough people to buy in, we will see all sorts of magical kewl stuff happen"
Isn't that how Trump became Commander in Chief?
chalsall is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Java applet alternative a1call Programming 19 2019-11-08 22:31
Alternative to LL paulunderwood Miscellaneous Math 36 2019-08-26 08:09
Fulsorials a1call Miscellaneous Math 41 2019-07-21 14:19
Is CEMPLLA 1.5 "the only software in the world capable of discovering" something? Not really. CRGreathouse Number Theory Discussion Group 51 2018-12-16 21:55
free alternative to EasyFit? ixfd64 Software 1 2008-04-26 21:28

All times are UTC. The time now is 07:21.

Fri Jun 5 07:21:33 UTC 2020 up 72 days, 4:54, 0 users, load averages: 1.25, 1.16, 1.09

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.