Do I need Intel Math Kernel Libraries?

Hi, all:

I am planning to use an Intel Mac (Leopard) for running numerical models (WRF, CMAQ, SMOKE). I want to download the Intel Fortran Compiler for Mac but I don't know whether to blow the money getting the Professional version with Intel Math Kernel Libraries. I have already installed XCode 3.0 and I found:

/Developer/SDKs/MacOSX10.5.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A
-rwxr-xr-x 1 root wheel 185664 Oct 12 00:23 vecLib
-rwxr-xr-x 1 root wheel 154260 Oct 12 00:23 libBLAS.dylib
-rwxr-xr-x 1 root wheel 623676 Oct 12 00:23 libLAPACK.dylib
-rwxr-xr-x 1 root wheel 78760 Oct 12 00:23 libvDSP.dylib
-rwxr-xr-x 1 root wheel 77848 Oct 12 00:23 libvMisc.dylib
drwxr-xr-x 14 root wheel 476 Oct 12 00:23 Headers

Is there any advantage to using the Intel Math Library in terms of speed and perfomance for Fortran/C programs?

Intel fortran compiler is

Intel fortran compiler is clearly more powerful compared to gfortran on a MacIntel. There are some resturbs for non-commercial uses (academics, students..) around ~US$50 (I payed 45€).

RE: Do I need Intel Math Kernel Library?

Thanks. I am changing to ifort because I had problems with gfortran. You're right--there are academic discounts> I'm going to get both the Fortran Compiler and Math Kernel Library.

Debra Baker
Department of Atmospheric
and Oceanic Sciences
University of Maryland

Sparse solvers in Intel MKL also multi-core tuned

Also, most of the sparse solver in MKL are tuned for multi-core system. I am not sure how the rest of the MKL is for multi-core systems, but I believe they are also updated.

bulent

MKL bad on sparse

I had a bad experience with mkl sparse vector matrix product, 30% slower than my routine.
Did you benchmark it ?

Boolegue.

Trial Version

I was able to get a trial version and ran some benchmarks. In my case, I didn't find much improvement over the Leopard Accelerate framework, which exposes BLAS, LAPACK, etc.

On the other hand, if you're compiling numeric-intensive Fortran, the Intel compiler does seem to be a big improvement and worth the money, IMHO.