[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Could eig() be outperformed by MKL numpy?
From: |
Klonuo Umom |
Subject: |
Could eig() be outperformed by MKL numpy? |
Date: |
Tue, 17 May 2011 22:30:09 +0200 |
Hi,
I was comparing performance of ATLAS vs MKL NumPy and then Octave, through very
simple linalg tests, and found some interesting to me results:
Tests were run on low-end 2400 MHz P4 with 768 MB RAM running Windows XP SP3 32b
Octave 3.2.4, Python 2.6, NumPy 1.6
a = rand(1000, 1000)
b = rand(1000, 1000)
1. Simple dot product - dot(a,b)
======================
Octave : 0.031
Cygwin/ATLAS : 0.843
Windows/ATLAS* : 1.29
Windows/MKL : 2.37
----------------------
*from official numpy binary installer
2. Eigenvalues - eig(a)
=======================
Octave : 45.05
Cygwin/ATLAS : 116
Windows/MKL : 25.2
-----------------------
Results are in seconds.
Windows/MKL is Python 2.6.6 and numpy 1.6.0 build with Intel MKL library
Cygwin/ATLAS is Python 2.6.5. and numpy 2.0.dev build with ATLAS library
Comments are welcomed
- Could eig() be outperformed by MKL numpy?,
Klonuo Umom <=