mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-02-22, 17:18   #12
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,311 Posts
Default

V2.06 May 5 2017 beta of CUDALucas, to be exact; it has a bit more error detection in it than the April 18 2017 beta, which has more than 2.05. It's pretty solid; see https://www.mersenneforum.org/showpo...82&postcount=4
https://www.mersenneforum.org/showpo...83&postcount=8
https://www.mersenneforum.org/showpo...24&postcount=3
https://www.mersenneforum.org/showpo...19&postcount=4
kriesel is offline   Reply With Quote
Old 2019-02-22, 17:25   #13
GP2
 
GP2's Avatar
 
Sep 2003

29·89 Posts
Default

The characteristic error of CUDALucas is sometimes zeroing out the residue, so then it stays zeroed until the end.

But there are three machines that I mentioned with a very high error rate, returning non-zero residues. Maybe those are running mprime and there's a hardware problem.

You should look into that, or at least run PRP with those machines, for far superior error checking.
GP2 is offline   Reply With Quote
Old 2019-02-22, 17:59   #14
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

10010000011102 Posts
Default CUDALucas repositories maintenance needed

Quote:
Originally Posted by ET_ View Post
Is there a 2.06 beta for Linux (or at least source code to compile)?
The sourceforge content could use an update. Promote 2.06 May 5 2017 version (r102) from beta. Deemphasize 2.05.1. Post there updated readme.txt. (https://www.mersenneforum.org/showpo...84&postcount=6) As I recall, the source code is common for Windows/linux, with conditional compile directives included. https://sourceforge.net/projects/cudalucas/files/ seems somewhat out of date.
The mirror at https://download.mersenne.ca/CUDALucas lacks the v2.06 may 5 2017 build of Windows multiple executables package, but has a CUDA9.1 supporting version. (Wasn't there a CUDA10 version of CUDALucas build done also? Maybe I'm thinking of an mfaktc build.)
As I recall, linux builds are distribution or version dependent, so executables posted built from any given linux environment are limited where they're useful in linux. MAYBE static linking would make it less so. For some recent experience in getting it to compile in linux, see the CUDALucas thread, such as
https://www.mersenneforum.org/showpo...postcount=2670

Please consider posting there some successful compiled linux executables, specifying what environment and CUDA level and compute capability range they are for.

Last fiddled with by kriesel on 2019-02-22 at 18:10
kriesel is offline   Reply With Quote
Old 2019-02-22, 18:14   #15
CRGreathouse
 
CRGreathouse's Avatar
 
Aug 2006

22×1,483 Posts
Default

Quote:
Originally Posted by GP2 View Post
The characteristic error of CUDALucas is sometimes zeroing out the residue, so then it stays zeroed until the end.

But there are three machines that I mentioned with a very high error rate, returning non-zero residues. Maybe those are running mprime and there's a hardware problem.

You should look into that, or at least run PRP with those machines, for far superior error checking.
We could use more buggy machines running PRP, I think we're still trying to improve its reliability further.
CRGreathouse is offline   Reply With Quote
Old 2019-02-22, 20:09   #16
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

10010000011102 Posts
Default

Quote:
Originally Posted by GP2 View Post
The characteristic error of CUDALucas is sometimes zeroing out the residue, so then it stays zeroed until the end.
A characteristic error for CUDALucas and some other applications is a premature zero value. There are more, for CUDALucas, corresponding to +-2. Bad residues observed versus application:
%badresidues=(
'cllucas', '0x0000000000000002, 0xffffffff80000000',
'cudalucas', '0x0000000000000000, 0x0000000000000002, 0xfffffffffffffffd',
'cudapm1', '0x0000000000000000, 0x0000000000000001, 0xfff7fffbfffdfffe, 0xfff7fffbfffdffff, 0xfff7fffbfffffffe, 0xfff7fffbffffffff, '.
'0xfff7fffffffdfffe, 0xfff7fffffffdffff, 0xfff7fffffffffffe, 0xfff7ffffffffffff, 0xfffffffbfffdfffe, 0xfffffffbfffdffff, '.
'0xfffffffbfffffffe, 0xfffffffbffffffff, 0xfffffffffffdfffe, 0xfffffffffffdffff, 0xfffffffffffffffe, 0xffffffffffffffff',
'gpuowl', '0x0000000000000000',
'mfaktc', '',
'mfakto', ''
);
Residue64 values are not as applicable for mfaktx. The bad CUDAPm1 residues seen with lots of f's were cyclic, not monotonic repeats.

Mfaktc can return a known specious factor 38814612911305349835664385407. It's only valid for one p value, which is part of the self test. I think based on past experience with a couple declining gpus here, that this shows up for other exponents at declining hardware reliability.

Last fiddled with by kriesel on 2019-02-22 at 20:16
kriesel is offline   Reply With Quote
Old 2019-02-23, 05:40   #17
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

CCD16 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Sorry, Travis. I just looked at the save file in a text editor and it appears to be all zeroes.

Please use the latest CudaLucas (which may be marked Beta) (2.06??), it is supposed to detect this problem. I have no idea why this important fix never made it into the production version.
Ditto that.

Honestly, George and I are at the point where when we see the email notification that a possible prime was found, and we see the cudalucas version is 2.05.1, we just ignore it. WAY too many false positives.

To put it simply, if you're using cudalucas 2.05.1, update it.
Madpoo is offline   Reply With Quote
Old 2019-02-23, 05:44   #18
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

29×113 Posts
Default

Quote:
Originally Posted by travisjank View Post
Possible Prime?

M90,xxx,xxx might be a new prime! The admins have been notified and will investigate this result before it can be checked in.

Lucas-Lehmer test: !!!PRIME!!!
I apologize, Travis, because I saw the multiple emails that the server generated when you submitted the result. I meant to email you to let you know it was almost certainly because of a bug in that version of code, but I didn't get a chance.
Madpoo is offline   Reply With Quote
Old 2019-02-23, 05:49   #19
dcheuk
 
dcheuk's Avatar
 
Jan 2019
Pittsburgh, PA

E716 Posts
Default

Greetings!

I had the same problem running cudaLucas 2.05 on my 2070 and 2080, returning zero residues from the start. I was able to solve the problem by running 2.06 beta.

Maybe it's a driver or cuda software issue I assume? As it does not seem to affect pascal cards.
dcheuk is offline   Reply With Quote
Old 2019-02-23, 15:56   #20
Stargate38
 
Stargate38's Avatar
 
"Daniel Jackson"
May 2011
14285714285714285714

2·3·101 Posts
Default

So what was the exact exponent that the OP posted? Is it really a false positive? If not, could you please PM me the exponent?
Stargate38 is offline   Reply With Quote
Old 2019-02-23, 20:59   #21
Mark Rose
 
Mark Rose's Avatar
 
"/X\(‘-‘)/X\"
Jan 2013

23×359 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Ditto that.

Honestly, George and I are at the point where when we see the email notification that a possible prime was found, and we see the cudalucas version is 2.05.1, we just ignore it. WAY too many false positives.

To put it simply, if you're using cudalucas 2.05.1, update it.
Maybe the manual submission form should tell people about the bug and link them to the newer version, if 2.05.1 is detected?
Mark Rose is offline   Reply With Quote
Old 2019-02-23, 22:40   #22
dcheuk
 
dcheuk's Avatar
 
Jan 2019
Pittsburgh, PA

23110 Posts
Default

I'm curious, what is the probability that a residue from an LL test is bad? Is it dependent on the software/hardware/exponent/FFT size/etc?

If the residue is not guaranteed to be 100% correct (well I guess that's why we have double checks), what can we do overall to improve the accuracy (... beside getting ecc memory and quadros/tesla graphics) and should we be expecting bad results in some kind of frequency? And how unlikely is it for both the original LL test and the double-check test for an exponent to return the bad residue that is identical?

I should probably read the wikipedia page on FFT or something to understand it, but procrastinates. How much prerequisites do I need to understand the mathematical concept and proof behind fft assuming i have the standard algebra/analysis background as a grad student? Any recommendations on any resources that introduces and proves this topic?

Thanks guys. I appreciate it.
dcheuk is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
(M48) NEW MERSENNE PRIME! LARGEST PRIME NUMBER DISCOVERED! dabaichi News 571 2020-10-26 11:02
How does one prove that a mersenne prime found with CUDALucas is really prime? ICWiener Software 38 2018-06-09 13:59
Twin Prime Days, Prime Day Clusters cuBerBruce Puzzles 3 2014-12-01 18:15
disk died, prime work lost forever? where to put prime? on SSD or HDD? emily PrimeNet 3 2013-03-01 05:49
How do I determine the xth-highest prime on prime pages? jasong Data 7 2005-09-13 20:41

All times are UTC. The time now is 11:18.

Wed Oct 28 11:18:15 UTC 2020 up 48 days, 8:29, 2 users, load averages: 1.23, 1.51, 1.65

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.