mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-05-02, 19:28   #210
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
Rep├║blica de California

2BAA16 Posts
Default

Quote:
Originally Posted by retina View Post
Nah, the author was just greedy. Wanting to use other people's time, effort and resources to get the prize money for himself. That's all.

It was a mistaken path though, because people aren't stupid, and finding large primes is hard.
Oh, no, we *are* stupid ... we're just not *that* stupid. :)

Would appreciate link(s) to any of the "in the end he did listen" forum threads mentioned in DukeBG's post above.
ewmayer is offline   Reply With Quote
Old 2019-05-02, 22:56   #211
Dylan14
 
Dylan14's Avatar
 
"Dylan"
Mar 2017

22·32·13 Posts
Default

Just for prosperity sake, I had tried to install the CEMPLLA software, version 1.5, way back in 2017, on a system with a GTX 1060. From my memory, it didn't like being run in a sandbox and running it normally didn't work. I had managed to install it under safe mode, but when I went back to normal mode, it had said that something had went wrong.


Not all was lost, however. I managed to record the registry changes that the installer did, using RegShot. I have attached them here in a 7zipped file. Basically, what I can gather, is that this program is essentially a glorified registry cleaner.
Attached Files
File Type: 7z cemplla-reg-changes.7z (290.5 KB, 17 views)
Dylan14 is offline   Reply With Quote
Old 2019-05-03, 05:56   #212
DukeBG
 
Mar 2018

3×43 Posts
Default

Quote:
Originally Posted by kriesel View Post
Where did you see the implementation of res64's in CEMPLLA?
Quote:
Originally Posted by ewmayer View Post
Would appreciate link(s) to any of the "in the end he did listen" forum threads mentioned in DukeBG's post above.
I wrote that based on his words (assuming he's ignorant, but not a liar) here: https://devtalk.nvidia.com/default/t...83292/#5183292
Quote:
And yes, CEMPLLA v1.5 displays the 64-bit residue for every iteration it completes.
Would be lovely to see a single screenshot, though, but alas.

----

Quote:
Originally Posted by kriesel View Post
It's also my recollection that the author made reference to displaying timing info as seconds/iteration. If that's multiple seconds, that's low performance.

If the author could not run it, and others could not run it, it may have been a substantial accomplishment in coding (undocumented performance questions aside), but I'm not sure we should call it developing.
Couldn't agree more

Last fiddled with by DukeBG on 2019-05-03 at 05:58
DukeBG is offline   Reply With Quote
Old 2019-05-03, 14:12   #213
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

384010 Posts
Default

For posterity's sake
https://devtalk.nvidia.com/default/t...83292/#5183292 June 2017
Quote:
CEMPLLA is very easy to use,
Not according to various GIMPSters who tried to get it to run. Or perhaps I'm recalling difficulties with CEMPLLA v1.0.

Quote:
and is currently the only software in the world capable of discovering the first one billion decimal digit prime number.
Mlucas counterpoint

Quote:
And it's one of only two in the world capable of discovering the first one hundred million decimal digit prime number.
As of 6/28/17, when the above claim was made, I think that list was
  1. Mlucas 14 (Sep 2014) or earlier https://www.mersenneforum.org/mayer/...drel.html#14.0
  2. prime95/mprime since v24.12 (June 2005) could do up to around 596 million exponent (179 million decimal digits) https://en.wikipedia.org/wiki/Prime95#Release_history
  3. CUDALucas (up to at least 64M fft p<1143276383, or 344,160,485 decimal digits, by May 2017 or earlier; Laurv made reference to 20480K ffts, 100Mdigit capable, at https://www.mersenneforum.org/showpo...postcount=2157, in March 2014)
  4. clLucas (up to at least 64M fft, by January 2016 v1.04)
(but perhaps not CEMPLLA though; see first point, about trying to get CEMPLLA to run. Also there are concerns about the claim by CEMPLLA's author that longer fft lengths could be much faster; that's sometimes associated in other applications and perhaps the NVIDIA library that CEMPLLA also uses, failing somehow in certain circumstances to execute properly)

Gpuowl is now capable of 100Mdigit primality tests also, since V3.3, July 2018, and theoretically gigadigit since V3.5 also July 2018 (although run times are too long on gigadigit test candidates, as they are for any existing software on available hardware) https://www.mersenneforum.org/showpo...83&postcount=7

Last fiddled with by kriesel on 2019-05-03 at 14:16
kriesel is offline   Reply With Quote
Old 2019-06-04, 00:29   #214
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

28·3·5 Posts
Default

Quote:
and is currently the only software in the world capable of discovering the first one billion decimal digit prime number.
Gpuowl appears to have just recently joined that exclusive club;
Mlucas https://www.mersenneforum.org/mayer/README.html#faq (LL on cpu; ~20 years / exponent try)
gpuowl https://www.mersenneforum.org/showpo...postcount=1230 (PRP3 on gpu, for the sufficiently patient, ~24 years / exponent try on RX480, so ~6 years /try on Radeon VII)
CEMPLLA (allegedly, although there are reports of being unable to run it, and a lack of sample runs or timings)
kriesel is offline   Reply With Quote
Old 2019-06-04, 01:41   #215
CRGreathouse
 
CRGreathouse's Avatar
 
Aug 2006

22·5·293 Posts
Default

Thank you for the report, very useful!

The (pseudonymous) author of CEMPLLA may have done some good, if he played a part in spurring these packages to new heights with his unfounded claim.
CRGreathouse is offline   Reply With Quote
Old 2019-06-04, 02:23   #216
GP2
 
GP2's Avatar
 
Sep 2003

2·1,289 Posts
Default

Quote:
Originally Posted by kriesel View Post
Gpuowl appears to have just recently joined that exclusive club;
Mlucas https://www.mersenneforum.org/mayer/README.html#faq (LL on cpu; ~20 years / exponent try)
gpuowl https://www.mersenneforum.org/showpo...postcount=1230 (PRP3 on gpu, for the sufficiently patient, ~24 years / exponent try on RX480, so ~6 years /try on Radeon VII)
I hesitate to post in a thread that should have died a long time ago. (At the very least, it ought to be retitled.)

But the exponent M1,073,741,827 figures in the "new Mersenne conjecture". First it should probably be TF'd a bit more, though.
GP2 is offline   Reply With Quote
Old 2019-06-04, 03:31   #217
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

3·1,087 Posts
Default

Quote:
Originally Posted by CRGreathouse View Post
Thank you for the report, very useful!

The (pseudonymous) author of CEMPLLA may have done some good, if he played a part in spurring these packages to new heights with his unfounded claim.
I think it's pretty clear that these additional FFT sizes were going to be added no matter what. I wouldn't credit CEMPLLA with anything beyond giving us all a good chuckle. And hey, that's not so bad really. Everyone can use a good laugh from time to time.
Madpoo is offline   Reply With Quote
Old 2019-06-04, 03:58   #218
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

F0016 Posts
Default

Quote:
Originally Posted by GP2 View Post
But the exponent M1,073,741,827 figures in the "new Mersenne conjecture". First it should probably be TF'd a bit more, though.
TF is no problem. We have the technology.

P-1 is a problem. I'm not aware of a cpu or gpu based application that will handle that exponent. Prime95/mprime implementation top out at 920M. CUDAPm1 exhibits gpu-model-specific limits that are lower. I haven't had a chance to characterize gpuowl P-1 yet.

LL or PRP3 is not a problem. (Mlucas, CUDALucas, Gpuowl, even cllucas have the range.) Run time can be long, but hardware advances are addressing that. It is already feasible within hardware lifetime. M999999937 took 6 months. This would take ~7.5 months.

To my knowledge there's nothing above ~604M double-checked yet.

Quote:
Originally Posted by Madpoo View Post
I wouldn't credit CEMPLLA with anything beyond giving us all a good chuckle. And hey, that's not so bad really. Everyone can use a good laugh from time to time.
The CEMPLLA saga and its author's lack of context was part of what motivated me in developing and collecting reference materials, both for users and programmers. https://www.mersenneforum.org/forumdisplay.php?f=154

Quote:
Originally Posted by CRGreathouse View Post
Thank you for the report, very useful!

The (pseudonymous) author of CEMPLLA may have done some good, if he played a part in spurring these packages to new heights with his unfounded claim.
You're welcome.

CUDALucas could join the gigadigit club with a little work. It's already capable of fft benchmarking and threads benchmarking up to 256M fft length. Its exponent is currently capped at 231 because of signed type. It didn't take Mihai long to remove the equivalent limitation from gpuowl.
Ernst: "I suspect Jerry Hallett's CUDA code running on a high-end GPU can beat my KNL times - if it doesn't currently permit such large exponents perhaps he can fiddle it to do so, just for the grins."
https://www.mersenneforum.org/showpo...&postcount=186

Last fiddled with by kriesel on 2019-06-04 at 04:10
kriesel is offline   Reply With Quote
Old 2019-06-04, 12:55   #219
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·2,281 Posts
Default

Quote:
Originally Posted by kriesel View Post
TF is no problem. We have the technology.

P-1 is a problem. I'm not aware of a cpu or gpu based application that will handle that exponent. Prime95/mprime implementation top out at 920M. CUDAPm1 exhibits gpu-model-specific limits that are lower. I haven't had a chance to characterize gpuowl P-1 yet.
I think AVX-512 can handle that exponent.
Prime95 is offline   Reply With Quote
Old 2019-06-04, 15:58   #220
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

74008 Posts
Default

Re P-1, 1073M
Quote:
Originally Posted by Prime95 View Post
I think AVX-512 can handle that exponent.
Planning features for your next prime95 version?

I'd like to be able to go higher than current limits in prime95 P-1 with any of the following:
i7-8750H (currently 16GB ram; estimated run time is ~1.5 months at 901M)
dual e5-2690 (64GB ram)
dual e5-2670 (128GB ram)
Some run time scaling info is available at https://www.mersenneforum.org/showthread.php?t=23900
kriesel 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 06:01.

Fri Jun 5 06:01:21 UTC 2020 up 72 days, 3:34, 0 users, load averages: 1.33, 1.31, 1.29

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.