Philippe Grosjean
2002-Oct-09 08:35 UTC
[R] R 1.6.0 benchmark with and without optimized ATLAS
Hello,
I am updating my benchmark (http://www.sciviews.org/other/benchmark.htm) to
recent versions of data analysis software (including R 1.6.0 and Splus 6.1),
and I now run it on a Pentium IV instead of the old Celeron 500 Mhz that
candidates for retirement. I test R under Windows Xp pro with and without
optimized BLAS. I use the optimized Rblas.dll for P4 found on CRAN. Here are
the results. Beside the discussion of limited role of the benchmark and
cautions on its interpretation (see the benchmark web page where this is
developed), does anyone want to comment it. I would more particularly
appreciate comments on the differences between results with and without the
optimized BLAS library (cross-product is 3 times faster, but linear
regression and inverse are slower with the optimized BLAS).
Best,
Philippe
...........]<(({?<...............<?}))><...............................
( ( ( ( (
) ) ) ) ) Philippe Grosjean
( ( ( ( (
) ) ) ) ) IFREMER Nantes - DEL/AO
( ( ( ( ( rue de l'Ile d'Yeu, BP 21105, 44311 Nantes Cedex 3
) ) ) ) ) tel: (33) 02.40.37.42.29, fax: (33) 02.40.37.42.41
( ( ( ( ( e-mail: philippe.grosjean at ifremer.fr
) ) ) ) )
( ( ( ( ( "I'm 100% confident that p is between 0 and 1"
) ) ) ) ) L. Gonick & W. Smith (1993)
.......................................................................
Benchmark on PIV 1.6 Ghz, 512 Mb Ram, Win Xp pro :
=================================================note the code for the test is
available on the web page
(see http://www.sciviews.org/other/benchmark.htm).
- R 1.6.0 with standard Rblas.dll
R Benchmark
==========Number of times each test is run__________________________: 3
I. Matrix calculation
---------------------
Creation, transp., deformation of a 1200x1200 matrix (sec):
0.973333333333333
1250x1250 normal distributed random matrix ^1000____ (sec):
2.25333333333333
Sorting of 1,100,000 random values__________________ (sec):
0.463333333333335
550x550 cross-product matrix (b = a' * a)___________ (sec):
0.463333333333331
Linear regression over a 700x700 matrix (c = a \ b') (sec): 1.75
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
0.924125732935107
II. Matrix functions
--------------------
FFT over 900,000 random values______________________ (sec):
1.53666666666667
Eigenvalues of a 220x220 random matrix______________ (sec):
0.466666666666669
Determinant of a 750x750 random matrix______________ (sec): 1.65
Cholesky decomposition of a 1000x1000 matrix________ (sec):
1.30333333333334
Inverse of a 500x500 random matrix__________________ (sec):
1.76000000000000
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
1.48949725041955
III. Programmation
------------------
225,000 Fibonacci numbers calculation (vector calc)_ (sec): 0.25
Creation of a 1500x1500 Hilbert matrix (matrix calc) (sec):
0.49333333333333
Grand common divisors of 35,000 pairs (recursion)___ (sec):
0.539999999999997
Creation of a 220x220 Toeplitz matrix (loops)_______ (sec):
0.896666666666666
Escoufier's method on a 22x22 matrix (mixed)________ (sec):
0.159999999999997
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
0.405344927915484
Total time for all 15 tests_________________________ (sec): 14.96
Overall mean (sum of I, II and III trimmed means/3)_ (sec):
0.823250186027012
--- End of test ---
- R 1.6.0 with optimized Rblas.dll (P4 version)
R Benchmark
==========Number of times each test is run__________________________: 3
I. Matrix calculation
---------------------
Creation, transp., deformation of a 1200x1200 matrix (sec): 0.98
1250x1250 normal distributed random matrix ^1000____ (sec):
2.25666666666667
Sorting of 1,100,000 random values__________________ (sec):
0.466666666666664
550x550 cross-product matrix (b = a' * a)___________ (sec):
0.156666666666669
Linear regression over a 700x700 matrix (c = a \ b') (sec): 2.03
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
0.975535245596796
II. Matrix functions
--------------------
FFT over 900,000 random values______________________ (sec): 1.49
Eigenvalues of a 220x220 random matrix______________ (sec):
0.463333333333334
Determinant of a 750x750 random matrix______________ (sec):
1.32333333333333
Cholesky decomposition of a 1000x1000 matrix________ (sec):
1.15333333333333
Inverse of a 500x500 random matrix__________________ (sec): 1.98
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
1.31503341435566
III. Programmation
------------------
225,000 Fibonacci numbers calculation (vector calc)_ (sec): 0.25
Creation of a 1500x1500 Hilbert matrix (matrix calc) (sec):
0.470000000000004
Grand common divisors of 35,000 pairs (recursion)___ (sec):
0.533333333333336
Creation of a 220x220 Toeplitz matrix (loops)_______ (sec):
0.906666666666666
Escoufier's method on a 22x22 matrix (mixed)________ (sec):
0.170000000000002
--------------------------------------------
Trimmed geom. mean (2 extremes eliminated):
0.397202705684386
Total time for all 15 tests_________________________ (sec): 14.63
Overall mean (sum of I, II and III trimmed means/3)_ (sec):
0.798725071833925
--- End of test ---
- Splus 6.1 on the same computer
Splus Benchmark
==============Number of times each test is run__________________________: 3
I. Matrix calculation
---------------------
Creation, transp., deformation of a 1200x1200 matrix (sec):
2.02266666666666
1250x1250 normal distributed random matrix ^1000____ (sec):
3.31466666666667
Sorting of 1,100,000 random values__________________ (sec):
1.22866666666666
550x550 cross-product matrix (b = a' * a)___________ (sec):
0.427666666666667
Linear regression over a 700x700 matrix (c = a \ b') (sec):
2.18966666666667
--------------------------------------------
Trimmed mean (2 extremes eliminated):
1.81366666666667
II. Matrix functions
--------------------
FFT over 900,000 random values______________________ (sec): 1.963
Eigenvalues of a 220x220 random matrix______________ (sec):
0.330333333333333
Determinant of a 750x750 random matrix______________ (sec):
0.670999999999997
Cholesky decomposition of a 1000x1000 matrix________ (sec):
2.76366666666667
Inverse of a 500x500 random matrix__________________ (sec): 2.644
--------------------------------------------
Trimmed mean (2 extremes eliminated):
1.75933333333333
III. Programmation
------------------
225,000 Fibonacci numbers calculation (vector calc)_ (sec):
0.501000000000005
Creation of a 1500x1500 Hilbert matrix (matrix calc) (sec):
0.514000000000001
Grand common divisors of 35,000 pairs (recursion)___ (sec):
0.333666666666659
Creation of a 220x220 Toeplitz matrix (loops)_______ (sec):
6.71633333333334
Escoufier's method on a 22x22 matrix (mixed)________ (sec): 5.108
--------------------------------------------
Trimmed mean (2 extremes eliminated): 2.041
Total time for all 15 tests_________________________ (sec):
30.7283333333333
Overall mean (sum of I, II and III trimmed means/3)_ (sec):
1.87133333333333
--- End of test ---
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Possibly Parallel Threads
Wisdom of the Ancients
