Benchmark on Linux

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Benchmark on Linux

Laurent Mazet-3


I've done many benchmark on diffrent PC's (486, 586 and 686). It seems that
the g77 compilator is less efficient the f2c translator. I don't no what's
happend on other Unix. Moreover, gcc optinisations aren't futile (I've test
a binary distribtion compiled for Pentium on a PPro and an other compiled
specialy for PPro). I've mirrored octave sources and some binary distribtions
for Linux on my ftp server (beijing.univ-mlv.fr). If it can help someone...

            Laurent


********************************
benchmark on a PC i486 133MH 24M
Compiled with:
 gcc 2.7.2
 f2c

Speed of octave 2.0.5 on i486-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         0.68 +/- 0.7% (7 runs)
Schur decomposition (LAPACK)      0.78 +/- 4.7% (9 runs)
Differential equation (LSODE)     1.91 +/- 3.7% (14 runs)
Fourier transforms (FFTPACK)      0.67 +/- 3.5% (7 runs)
for loop                          2.11 +/- 1.4% (28 runs)
-- Performance index (bm 1.10): 1.1


********************************
benchmark on a PC i586 133MH 16M
Compiled with:
 gcc 2.7.2
 f2c

Speed of octave 2.0.5 on i586-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         1.47 +/- 0.3% (21 runs)
Schur decomposition (LAPACK)      1.18 +/- 1.3% (7 runs)
Differential equation (LSODE)     1.79 +/- 0.6% (14 runs)
Fourier transforms (FFTPACK)      1.19 +/- 0.7% (14 runs)
for loop                          1.56 +/- 1.4% (21 runs)
-- Performance index (bm 1.10): 1.4

********************************
benchmark on a PC i586 166MH 48M
Compiled with:
 gcc 2.7.2
 f2c

Speed of octave 2.0.5 on i586-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         2.33 +/- 1.3% (28 runs)
Schur decomposition (LAPACK)      2.09 +/- 0.8% (7 runs)
Differential equation (LSODE)     3.46 +/- 1.5% (21 runs)
Fourier transforms (FFTPACK)      1.80 +/- 0.3% (21 runs)
for loop                          3.10 +/- 2.5% (42 runs)
-- Performance index (bm 1.10): 2.5

********************************
benchmark on a PC i586 166MH 48M
Compiled with:
 gcc 2.7.2.f.1
 g77

Speed of octave 2.0.5 on i586-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         1.39 +/- 5.5% (60 runs)
Schur decomposition (LAPACK)      0.99 +/- 2.1% (7 runs)
Differential equation (LSODE)     3.47 +/- 3.3% (21 runs)
Fourier transforms (FFTPACK)      0.72 +/- 4.3% (7 runs)
for loop                          3.10 +/- 5.2% (120 runs)
-- Performance index (bm 1.10): 1.6

********************************
benchmark on a PC i686 200MH 64M
Compiled with:
 gcc 2.7.2
 f2c

Speed of octave 2.0.5 on i586-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         3.15 +/- 1.5% (42 runs)
Schur decomposition (LAPACK)      2.48 +/- 5.7% (20 runs)
Differential equation (LSODE)     6.81 +/- 4.1% (49 runs)
Fourier transforms (FFTPACK)      3.91 +/- 0.6% (35 runs)
for loop                          6.38 +/- 1.2% (84 runs)
-- Performance index (bm 1.10): 4.2

********************************
benchmark on a PC i686 200MH 64M
Compiled with:
 gcc 2.7.2
 f2c

Speed of octave 2.0.5 on i686-pc-linux-gnu relative to Sun Sparc 10/40
Matrix inversion (LAPACK)         4.99 +/- 1.2% (63 runs)
Schur decomposition (LAPACK)      5.05 +/- 3.5% (14 runs)
Differential equation (LSODE)     6.90 +/- 1.3% (49 runs)
Fourier transforms (FFTPACK)      3.74 +/- 0.2% (42 runs)
for loop                          6.06 +/- 1.5% (84 runs)
-- Performance index (bm 1.10): 5.2

--
Laurent Mazet                                   PhD student in signal processing
University of Marne la Vallee                          and digital communication
FRANCE

E-mail: [hidden email]                             W-Tel: (33) 01.64.73.72.90

Reply | Threaded
Open this post in threaded view
|

Re: Benchmark on Linux

Francesco Potorti`-9
   I've done many benchmark on diffrent PC's (486, 586 and 686). It
   seems that the g77 compilator is less efficient the f2c
   translator.

This is another (or the same?) mistery about octave 2 performance.  In
particular, the test on Schur decomposition is slower for octave2 than
for octave1 on Sun and Alpha.  This seems to be not true for Intel.

--
Francesco Potorti` (researcher)        Voice:    +39-50-593203
Computer Network Division              Operator: +39-50-593211
CNUCE-CNR, Via Santa Maria 36          Fax:      +39-50-904052
56126 Pisa - Italy                     Email:    [hidden email]