Website bug: wrong units on the CPU TIME CERT

Message boards : Bug reports : Website bug: wrong units on the CPU TIME CERT

To post messages, you must log in.

AuthorMessage
HireMe.geek.nz

Send message
Joined: 23 Sep 07
Posts: 14
Credit: 221,822
RAC: 0
Message 723 - Posted: 11 Dec 2008, 23:00:49 UTC

Website bug: Wrong units on the CPU TIME CERT!

"has participated in Enigma@Home since 23 September 2007, and has contributed 2,079 Cobblestones of computation (1.80 quadrillion floating-point operations) to Enigma@Home."

BUT THE APPLICATION DOES NOT DO ANY FLOATING POINT MATH!
ID: 723 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 725 - Posted: 18 Dec 2008, 16:20:14 UTC - in response to Message 723.  

Actually, it does some fp calculations, but I'll change the certificate text anyway.

M4 Project homepage
M4 Project wiki
ID: 725 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
HireMe.geek.nz

Send message
Joined: 23 Sep 07
Posts: 14
Credit: 221,822
RAC: 0
Message 1968 - Posted: 5 Mar 2011, 11:25:50 UTC - in response to Message 725.  
Last modified: 5 Mar 2011, 11:40:04 UTC

Actually, it does some fp calculations, but I'll change the certificate text anyway.


Your Hill Climbing Method seems to use only Integer scoring.

Your Index of Coincidence used inside the Hill Climbing method uses limited floating point math.

However, the floating point matmatical requirements would not tax

-- http://en.wikipedia.org/wiki/Half_precision_floating-point_format

because only very small floats are used, and all are positive as far as I can tell.

That said, the certificate is still busted.

Certificate of Computation

This certifies that HireMe.geek.nz has participated in Enigma@Home since 23 September 2007, and has contributed 9,391 Cobblestones of computation (8.11 quadrillion floating-point operations) to Enigma@Home.

5 March 2011


Recommended fix :

Use the back end scripting that displays the page (and the cross project one too) and display both integer and floating point numbers.

I suggest subtracting 55% from the base floating point number to get the quasi correct Integer number. Display the number.

Then subtract 45% from the base floating point number again to get the proper floating point number. Display the other number.

This is a BOINC server problem and should be reported as a bug. I have submitted a few bugs but have not gotten around to this one.
ID: 1968 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 1971 - Posted: 10 Mar 2011, 20:25:02 UTC - in response to Message 1968.  

It was already fixed here, but when I upgraded the server software, the file was overwritten.
I set the fp_benchmark_weight to 0.1 because I believe that around 10% or less operations are floating point. It varies between apps, as far as I remember one of the optimized apps was pure integer.

M4 Project homepage
M4 Project wiki
ID: 1971 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Bug reports : Website bug: wrong units on the CPU TIME CERT




Copyright © 2024 TJM