mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2013-11-03, 00:25   #1
fairsky
 
Sep 2013

41 Posts
Default It seemed I got problem from CUDALucas test?

Hi guys,

Did anyone of you receive an email like below

Quote:
Hi,

There seems to be a problem with your CUDALucas results.
Please try asking for help at mersenneforum.org. There are many
there that can help you set up CUDALucas properly.


Regards,
George Woltman
It was sent from, apparently, woltman@... .

I reply to the sender, GW, for confirmation.
Is that possible he, GW, will send email to us if our result had wrong?

Thank you all.

Last fiddled with by Batalov on 2013-11-03 at 01:39 Reason: (email redacted)
fairsky is offline   Reply With Quote
Old 2013-11-03, 00:44   #2
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

976110 Posts
Default

Quote:
Originally Posted by fairsky View Post
Thank you all.
LOL... Quite possibly a bit of social engineering.

Care to drill down on the SMTP headers to determine the actual sender?
chalsall is online now   Reply With Quote
Old 2013-11-03, 00:59   #3
fairsky
 
Sep 2013

41 Posts
Default

Quote:
Originally Posted by chalsall View Post
LOL... Quite possibly a bit of social engineering.

Care to drill down on the SMTP headers to determine the actual sender?
I had this understanding.
But when I check my credits from GIMPS, I saw the rank was largely decreased from 119 to 280.
That implies all my cudaLucas credits (each is more than 5500) are deducted, in turn suggests my cudaLucas jobs are not kicked out off database....

Now, i just checked the header, then the sender is gwoltman@... .
Previously I just saw the sender's name where the GW@... was showed.

Regardless of the correct email address, it is possibly my cudaLucas jobs were failed, in the sense of my credits and ranks from GIMPS records.

So sad....

Last fiddled with by Batalov on 2013-11-03 at 01:40 Reason: (emails redacted)
fairsky is offline   Reply With Quote
Old 2013-11-03, 02:12   #4
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

165548 Posts
Default

Fairsky is the user having the problem with CUDALucas returning 0xFFFFFFFF80000000 -- presumably from an FFT length that is too small.
Prime95 is offline   Reply With Quote
Old 2013-11-03, 02:16   #5
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

9,491 Posts
Default

@fairsky :
This is a google-indexed forum. What you wrote will show up in google - you can search for it. If you wanted to repay George for his answers by giving away his email to spambots, you couldn't have done better. Don't post anyone's email addresses in the open, neither on this nor other forums. This will be looked upon unfavorably, everywhere. This is point number 1.

Point number 2: it is unethical to re-post private emails (and private messages) without the other respondent's approval. Of course, George didn't write anything special in this email, but next time you may repost some sensitive message and the result won't be pretty. So, don't.

Point number 3: yes, your results are patently wrong (they have the final resulting LL residue = 0xffffffff80000000) and as such the credit for them is zero. Do not override FFT size, unless you know what it is and what it does. (Goes without saying that a well-written program wouldn't have allowed you to set up meaningless FFT sizes, but hey, CUDAlucas is in eternal alpha stage.)

Anyway, the short version is: if you don't know what an option does, don't use it; use default. "One does not simply walk into Mordor," or in other words, one does not simply run the program 50 times faster and get the right result. It is like expecting that if you use ZIP 50 times, you will get compression every time and the final file will become really small. ;-) it won't! Similarly, the necessary computation to perform a certain task cannot be compressed beyond certain running time. The given credit is proportional to this necessary running time.
Batalov is offline   Reply With Quote
Old 2013-11-03, 02:23   #6
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

22×7×269 Posts
Default

Don't worry about my email address - every spambot already has it :)

Can the CUDALucas author(s) add a sanity check to the FFT length so another user doesn't stumble into the same problem?
Prime95 is offline   Reply With Quote
Old 2013-11-03, 02:28   #7
fairsky
 
Sep 2013

518 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Fairsky is the user having the problem with CUDALucas returning 0xFFFFFFFF80000000 -- presumably from an FFT length that is too small.
Wow, thank you for response here.

Yes, I was noticing this result pattern in my result.txt, but I did not get aware of the causes till today.
fairsky is offline   Reply With Quote
Old 2013-11-03, 02:31   #8
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

9,491 Posts
Default

Bill "Dubslow" Winslow was adding some options upon options, bells upon whistles, and never got to implement the shifted initial value (and corresponding changes to the savefile format ...or better yet some convergence to the existing format used by P95 binaries, or some compromise format). Now, that he has left this project, maybe owftheevil could address this?
Batalov is offline   Reply With Quote
Old 2013-11-03, 02:34   #9
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

9,491 Posts
Default

Quote:
Originally Posted by fairsky View Post
Wow, thank you for response here.

Yes, I was noticing this result pattern in my result.txt, but I did not get aware of the causes till today.
There's the whole thread devoted to this issue. It is important to read the existing threads before creating four new ones, ok?

This forum also has a search functionality.
Batalov is offline   Reply With Quote
Old 2013-11-03, 02:35   #10
fairsky
 
Sep 2013

2916 Posts
Default

Quote:
Originally Posted by Batalov View Post
@fairsky :
This is a google-indexed forum. What you wrote will show up in google - you can search for it. If you wanted to repay George for his answers by giving away his email to spambots, you couldn't have done better. Don't post anyone's email addresses in the open, neither on this nor other forums. This will be looked upon unfavorably, everywhere. This is point number 1.

Point number 2: it is unethical to re-post private emails (and private messages) without the other respondent's approval. Of course, George didn't write anything special in this email, but next time you may repost some sensitive message and the result won't be pretty. So, don't.

Point number 3: yes, your results are patently wrong (they have the final resulting LL residue = 0xffffffff80000000) and as such the credit for them is zero. Do not override FFT size, unless you know what it is and what it does. (Goes without saying that a well-written program wouldn't have allowed you to set up meaningless FFT sizes, but hey, CUDAlucas is in eternal alpha stage.)

Anyway, the short version is: if you don't know what an option does, don't use it; use default. "One does not simply walk into Mordor," or in other words, one does not simply run the program 50 times faster and get the right result. It is like expecting that if you use ZIP 50 times, you will get compression every time and the final file will become really small. ;-) it won't! Similarly, the necessary computation to perform a certain task cannot be compressed beyond certain running time. The given credit is proportional to this necessary running time.
Great. I actually understood that my post which contains the prvt email-address would get him/her in trouble.
But well, I did that for kind of reporting a possible scam... esp. i don't know where to get the answer.

But anyway, your suggestions are welcomed. And I am the person who really care about the info-security, so I think I can do it better from today on.

All in all, your comments are rational and accepted!
Thanks a lot for the reminding.
fairsky is offline   Reply With Quote
Old 2013-11-03, 02:37   #11
fairsky
 
Sep 2013

41 Posts
Default

Quote:
Originally Posted by Batalov View Post
There's the whole thread devoted to this issue. It is important to read the existing threads before creating four new ones, ok?

This forum also has a search functionality.
Aha,great post and advice, you are so professional.

I expect to find the answer for my another post regarding to cudaLucas settings.
Thanks again.
fairsky is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Problem during LL test Lorenzo Information & Answers 0 2016-07-22 08:17
CUDALucas Residue Test (-r) Reference Table Brain GPU Computing 0 2012-04-12 20:21
Windows XP BSOD Prime95 Blend Test - Paging Problem??? john6205 Software 18 2006-11-20 17:21
strange problem with torture test on 16core machines TheJudger Hardware 5 2006-04-08 11:20
Same crash, same test, same problem. Bug or instability? SlightlyStoopit Software 2 2004-04-18 01:31

All times are UTC. The time now is 20:59.


Thu Jul 29 20:59:50 UTC 2021 up 6 days, 15:28, 1 user, load averages: 2.86, 2.69, 2.77

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.