[SciPy-user] Advantages of Intel's MKL

Albert Strasheim fullung at gmail.com
Sat Dec 9 10:47:22 CST 2006


Hello all

On Sat, 09 Dec 2006, John Travers wrote:

> On 09/12/06, Robert Kern <robert.kern at gmail.com> wrote:
> > Nils Wagner wrote:
> > > A final question...
> > > Do I need Intel's C compiler to build numpy/scipy
> > > with MKL support ?
> >
> > I have no idea. Consult the MKL's documentation.
> >
> > --
> > Robert Kern
> 
> No, I use Intel's MKL with numpy/scipy on Linux and do not use intels
> C compiler. In fact I couldn't get it to work with Intel's compiler.
> You may need the fortran compiler though - at least that is what works
> with me.

As far as the C libraries go, MKL works with GCC without problems. I 
haven't experimented with MKL and GCC FORTRAN yet. If you have a few 
hours, I can really recommend that you play with Intel's compiler a 
bit. In some cases the code it produces is much faster than GCC's or 
MSVC's.
 
> I find that MKL is the fastest I've seen for ffts. Not sure about how
> it squares up for Lapack type stuff though.

Makes sense. MKL 9 which was released recently apparently has even 
faster FFT code for the new Xeons. The free for non-commercial use 
version of MKL 9 will be available in early 2007.

Cheers,

Albert


More information about the SciPy-user mailing list