[SciPy-User] Building NumPy 1.6.0 with MKL and MSVS9 on XP 32bit

[Charles R Harris]

On Tue, May 17, 2011 at 3:47 PM, Klonuo Umom <klonuo at gmail.com> wrote:

> > It would help if you gave the actual code you're running there. It is
> > extremely unlikely that Octave is really 27 times faster than Numpy +
> > ATLAS/MKL.
>
> I posted it in my previous mail. Here is basically same procedure:
>
> =======================================
> import numpy as np
> from numpy.random import random
>
> a = random((1000, 1000))
> b = random((1000, 1000))
> %timeit np.dot(a, b)
> ---------------------------------------
>
> Octave code:
> =======================================
> a = rand(1000, 1000);
> b = rand(1000, 1000);
> tic; dot(a,b); toc
> ---------------------------------------
>
> And it just here, where MKL build numpy, behaves unexpectedly slow.
> Or I'm making some mistake I'm not aware of
>
>
I get 67ms/loop with plain ATLAS.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20110517/408ff3f8/attachment.html>