benchmark 1.4 and results

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

benchmark 1.4 and results

Francesco Potorti`-9
There is a new version (bm 1.4) of the octave benchmark at
ftp://fly.cnuce.cnr.it/pub/benchmark.m.  The tests are the same as
before, only some minor changes have been made, and the Sun Sparc 10
is now the reference machine, as it is the platform used for Spec95
numbers, as Rick Niles says.

I collected the numbers some of you sent to me and organised them in a
table.  The numbers referred to Sparc2, I just scaled them in order to
match the new reference, which is Sparc 10.

Unfortunely I miss some interesting platforms, for which I have
numbers only for the very first version of the benchmark, which are
not scalable to the new tests.  Here are the missing platforms:

Evan Thomas <[hidden email]>
  Linux/Pentium 133
Jarle Brinchmann <[hidden email]>
  Linux/486dx4-100
John L Daschbach <[hidden email]>
  Linux/Pentium 120
  OS2/486
  IBMrisc6000

And here is the table of results:

Octave benchmark bm 1.4
                                 inv   schur lsode fft2  for   index

[11] Linux/486dx-33              0.27  0.36  0.33  0.26  0.20   0.28
[8] Sun Ultra 1                  0.23  0.95  0.37  0.48  0.13   0.35
[5] Linux/486dx2-50              0.83  0.52  0.40  0.37  0.25   0.37
[2] Sun Sparc 2                  0.49  0.81  0.33  0.73  0.17   0.44
[9] Linux/486dx2-66              0.46  0.65  0.48  0.46  0.29   0.45
[8] HP 9000/735/99               0.33  0.94  0.91  0.58  0.40   0.58
[8] SGI Indigo 2 XL              0.47  2.75  0.59  1.12  0.23   0.72
[9] Sun Sparc 10/30              1.21  1.90  0.64  1.32  0.42   0.96
[10] Sun Sparc 10                1.00  1.00  1.00  1.00  1.00   1.0
[6] Linux/Pentium 90             1.52  2.42  1.28  1.24  0.77   1.4
[5] Linux/Pentium 100            1.84  2.57  1.23  1.30  0.79   1.4
[9] Sun Sparc 20                 1.87  2.94  0.85  2.02  0.60   1.4
[7] Irix 5.3/R8000 75/90         2.18  2.51  1.21  2.55  0.69   1.6
[1] Alpha 2100                   4.55  6.83  1.78  4.97  1.07   3.1

[1] Francesco Potorti` <[hidden email]>
[2] Evan Thomas <[hidden email]>
[3] Jarle Brinchmann <[hidden email]>
[4] John L Daschbach <[hidden email]>
[5] Michael Smolsky <[hidden email]>
[6] Jim Van Zandt <[hidden email]>
[7] Przemek Klosowski <[hidden email]>
[8] Matthias Roessler <[hidden email]>
[9] Krzysztof Gozdziewski <[hidden email]>
[10] Rick Niles <[hidden email]>
[11] Carl Wilhelm Welin <[hidden email]>

Reply | Threaded
Open this post in threaded view
|

Re: benchmark 1.4 and results

niles-5

This for a DEC Alpha station 400 -> 21064A @ 233MHz
Running OSF/1 (DEC UNIX) 3.2

Octave benchmark version bm 1.4
Speed of octave ss-960323 on alpha-dec-osf3.2 relative to Sun Sparc 10
Matrix inversion (LAPACK)         3.51 +/- 3.7% (42 runs)
Schur decomposition (LAPACK)      6.53 +/- 3.2% (14 runs)
Differential equation (LSODE)     1.74 +/- 4.6% (14 runs)
Fourier transforms (FFTPACK)      4.26 +/- 1.6% (28 runs)
for loop                          1.50 +/- 4.9% (32 runs)
-- Performance index (bm 1.4): 3

Reply | Threaded
Open this post in threaded view
|

Re: benchmark 1.4 and results

Matthias Roessler
In reply to this post by Francesco Potorti`-9

Some of my results (with bm 1.3) on SPARCstations 10 and 20
and a benchmark 1.4 done on an Ultra 1.
 
##########################################################
# The following benchmarks are done with
# Octave benchmark version bm 1.3a
# WARNING: if other processes are running the figures will
# be inaccurate
# Speed of octave 1.1.1 on sparc-sun-solaris2.4 relative
# to Sun Sparc 2
# Performance
# SPARCstation 10
#  Model 20 (33MHz) 2
#  Model 30 (36MHz) 2.2
#  Model 40 2.2
#  Model 41 2.4
#  Model 51 2.3
# SPARCstation 20
#  2x Model 50 3
#  2x Model 61 3.9
#  125MHz hyperSPARC 4.6
#  
# Matthias Roessler <[hidden email]
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 10
#  Model 20 SuperSPARC SPARCmodule
#  64 MB Memory, 125 MB Swap
# Matrix inversion (LAPACK)         2.25 +/- 1.9% (14 runs)
# Schur decomposition (LAPACK)      2.11 +/- 1.8% (7 runs)
# Differential equation (LSODE)     1.71 +/- 2.8% (7 runs)
# Fourier transforms (FFTPACK)      1.67 +/- 4.8% (8 runs)
# for loop                          2.21 +/- 2.2% (7 runs)
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 10
#  Model 30 SuperSPARC SPARCmodule
#  64 MB Memory, 206 MB Swap
# Matrix inversion (LAPACK)         2.52 +/- 2.3% (14 runs)
# Schur decomposition (LAPACK)      2.31 +/- 2.4% (7 runs)
# Differential equation (LSODE)     1.85 +/- 2.5% (7 runs)
# Fourier transforms (FFTPACK)      1.93 +/- 1.3% (7 runs)
# for loop                          2.37 +/- 2.8% (7 runs)
# -- Performance index (bm 1.3a): 2.2
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 10
#  Model 40 SuperSPARC SPARCmodule
#  64 MB Memory, 103 MB Swap
# Matrix inversion (LAPACK)         2.53 +/- 10.0% (40 runs)
# Schur decomposition (LAPACK)      2.42 +/- 2.4% (7 runs)
# Differential equation (LSODE)     2.05 +/- 7.2% (20 runs)
# Fourier transforms (FFTPACK)      1.96 +/- 0.7% (7 runs)
# for loop                          2.68 +/- 3.8% (7 runs)
# -- Performance index (bm 1.3a): 2.3
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 10
#  Model 41 SuperSPARC SPARCmodule
#  64 MB Memory, 133 MB Swap
# Matrix inversion (LAPACK)         2.72 +/- 4.9% (30 runs)
# Schur decomposition (LAPACK)      2.73 +/- 1.8% (7 runs)
# Differential equation (LSODE)     2.20 +/- 2.4% (7 runs)
# Fourier transforms (FFTPACK)      1.95 +/- 8.3% (20 runs)
# for loop                          2.76 +/- 3.6% (7 runs)
# -- Performance index (bm 1.3a): 2.4
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 10
#  Model 51 SuperSPARC SPARCmodule
#  64 MB Memory, 212 MB Swap
# Matrix inversion (LAPACK)         1.74 +/- 34.0% (40 runs)
# Schur decomposition (LAPACK)      2.12 +/- 31.1% (20 runs)
# Differential equation (LSODE)     2.21 +/- 1.6% (7 runs)
# Fourier transforms (FFTPACK)      2.21 +/- 3.1% (14 runs)
# for loop                          3.20 +/- 4.8% (7 runs)
# -- Performance index (bm 1.3a): 2.3
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 20
#  2 x Model 61 SuperSPARC SPARCmodule
#  64 MB Memory, 155 MB Swap
# Matrix inversion (LAPACK)         4.59 +/- 0.7% (21 runs)
# Schur decomposition (LAPACK)      4.23 +/- 0.4% (7 runs)
# Differential equation (LSODE)     3.31 +/- 1.4% (7 runs)
# Fourier transforms (FFTPACK)      3.33 +/- 0.6% (14 runs)
# for loop                          4.22 +/- 2.2% (14 runs)
# -- Performance index (bm 1.3a): 3.9
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 20
#  125 MHz hyperSPARC
#  128 MB Memory, 208 MB Swap
# Matrix inversion (LAPACK)         4.61 +/- 1.1% (21 runs)
# Schur decomposition (LAPACK)      6.68 +/- 1.4% (14 runs)
# Differential equation (LSODE)     3.69 +/- 1.1% (7 runs)
# Fourier transforms (FFTPACK)      3.85 +/- 1.2% (21 runs)
# for loop                          4.78 +/- 4.9% (34 runs)
# -- Performance index (bm 1.3a): 4.6
#---------------------------------------------------------
# Sun (Sun Microsystems) SPARCstation 20
#  2 x Model 50 SuperSPARC SPARCmodule
#  128 MB Memory, 349 MB Swap
# Matrix inversion (LAPACK)         3.47 +/- 0.2% (21 runs)
# Schur decomposition (LAPACK)      3.28 +/- 0.3% (7 runs)
# Differential equation (LSODE)     2.65 +/- 0.5% (7 runs)
# Fourier transforms (FFTPACK)      2.48 +/- 0.4% (14 runs)
# for loop                          3.39 +/- 0.6% (7 runs)
# -- Performance index (bm 1.3a): 3
##########################################################



Now bm 1.4 and relative to SPARC 10

##########################################################
Sun (Sun Microsystems) Ultra 1
 167 MHz UltraSPARC
 124 MB Memory, 262 MB Swap
WARNING: if other processes are running the figures will be inaccurate
Speed of octave 1.1.1 on sparc-sun-solaris2.4 relative to Sun Sparc 10
Matrix inversion (LAPACK)         2.34 +/- 7.4% (80 runs)
Schur decomposition (LAPACK)      4.11 +/- 5.8% (20 runs)
Differential equation (LSODE)     1.21 +/- 4.3% (7 runs)
Fourier transforms (FFTPACK)      2.65 +/- 6.9% (40 runs)
for loop                          0.67 +/- 4.2% (14 runs)
-- Performance index (bm 1.4): 1.8

Matrix inversion (LAPACK)         2.35 +/- 3.3% (21 runs)
Schur decomposition (LAPACK)      4.31 +/- 4.6% (7 runs)
Differential equation (LSODE)     1.24 +/- 6.7% (20 runs)
Fourier transforms (FFTPACK)      2.59 +/- 4.9% (33 runs)
for loop                          0.69 +/- 3.6% (14 runs)
-- Performance index (bm 1.4): 1.9
##########################################################


Matthias
------------------------------------------------------------------------------
Dipl.-Inf. Matthias Roessler                            Phone: 49 9131 85 8027
IMMD IV,  Universitaet Erlangen-Nuernberg               Fax1:  49 9131 85 8732
Martensstrasse 1; D-91058 Erlangen                      Fax2:  49 9131 39 388
                                     [hidden email]
                             http://www4.informatik.uni-erlangen.de/~roessler/
------------------------------------------------------------------------------
"Es geht auch anders, aber so geht es auch!"                                  
"You can do this differently, but you can also do it like this!" (Lotti Huber)

Reply | Threaded
Open this post in threaded view
|

Re: benchmark 1.4 and results

niles-5

The SPEC95 numbers are for a SPARC 10/40 if Matthias has access to all
these different SPARC 10 machines perhaps you should use his numbers
in place of mine since I don't know how to tell which SPARC 10 I have!

        But the numbers all seem pretty close though.

If you want to quote a source of info on this fact better than myself,
check:
http://www.specbench.org/spec/specfaq.html#cpu95

Here's a quote:

The results ("SPEC Ratio" for each individual benchmark) are expressed
as the ratio of the wall clock time to execute one single copy of the
benchmark, compared to a fixed "SPEC reference time". For the SPEC95
benchmarks, a Sun SPARCstation 10/40 was chosen as the reference
machine.

        Sorry if this messes you up,
        Rick Niles.