Requesting tutorial on integrating ACML with Scipy/Numpy and Matlab

Discussion created by afl22 on Jul 17, 2014
Latest reply on Apr 7, 2015 by atcl

I've noticed that AMD does not provide tutorials on using ACML with Numpy/Scipy and Matlab, in stark contrast to intel which provides clear instructions for both (Google "MKL with numpy", then google "ACML with numpy". One is a page with a useful tutorial for MKL, the other is a page full of issues with ACML).


Nor does it provide performance comparisons, giving me zero reason to use it. If it exists, and I missed it, please point it out. This subpar documentation is certainly one of the reasons why amd processors and accelerators are nonexistant in scientific computing these days. Could you guys provide documentation, and or provide ACML integrated Numpy/Scipy license permitting. I would also like to point out I intensely dislike the AMD dev forum software and layout, it is far more difficult to use than the intel dev zone.


You guys might get away with non existent documentation in the consumer world, but try that in fields that actually run their processors 24/7 for months on end, and every little optimization can cut days of processor time, that approach doesn't work.


We need more documentation!

Edit: Numpy binaries with Intel MKL integrated can be found here. There is no reason AMD can't do the same.