I am experimenting with an Intel Xeon Phi 3120A card and automatic offloading using Python. I got it running using Intel Python 2017 with the help of this post. By that I found out that the card unfortunately is only supported by the 2017 version of MKL for automatic offloading. As Intel Python 2017 is based on Python 3.5 I am wondering if there is some way to use, e.g., Python 3.6 with MKL 2017 to preserve the automatic offloading functionality of the Xeon Phi card. Can I somehow set up a Miniconda environment with Python 3.6 still using MKL 2017? Or is it possible to compile numpy and scipy with the old MKL library?
Thanks!