Matrix benchmark

[John W. Eaton]

On 28-Jul-2003, address@hidden <address@hidden> wrote:

| Here is my machine, a Pentium 4. I've just compiled 2.1.50 with BLAS,
| atlas-SSE and libpack libraries available from Debian. These are the
| best times I got:
| 
| GNU Octave, version 2.1.50 (i686-pc-linux-gnu).
| octave-2.1.50:1> A = rand (1000,1000);
| octave-2.1.50:2> tic; B = A*A; toc
| ans = 1.9340
|  
| MATLAB Version 6.0.0.88 (R12) on GLNX86
| >> A = rand(1000,1000);
| >> tic; B=A*A; toc
| elapsed_time =
|     1.9422
| 
| 
| I declare a tie. Now watch this:
| 
| octave-2.1.50:35> A = rand(2048,2048);
| octave-2.1.50:36> tic; B=fft(A); toc
| ans = 0.84838
| 
| >> A = rand(2048,2048);
| >> tic; B=fft(A); toc  
| elapsed_time =
|     0.5116
| 
|     
| What is wrong? I compiled it with shared swfft... Souldn't I??

The tic and toc functions are not very good measurement tools because
they measure elapsed time, not CPU time.  Is there anything else going
on on your systems while you run these functions?  Even if you think
there is not, some OS processes could run which would affect the
results.  I would recommend using the cputime function instead.

Even if you use FFTW with Octave, there is no guarantee that it is
precisely the same as the version used by Matlab.

jwe



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------