Xeon Phi automatic offload in matlab

Hello All, I am fighting to make the MKL AO feature to work with Matlab 2016a. I have 1 Xeon Phi running on Windows 10 pro (+2 Xeon E5-2690). The automatic offload works in the LINPACK examples supplied with MKL and in R (!, Microsoft version). I have followed the instructions at this page

https://software.intel.com/en-us/articles/using-intel-math-kernel-library-with-mathworks-matlab-on-intel-xeon-phi-coprocessor-system

without much success -- there is no error message, but micsmc shows zero load on Xeon Phi and, obviously, all the job is done by the CPU's. I have tried it with Matlab MKL (11.2.3) and with new 2017 MKL (recognized by matlab, version -BLAS shows the 2017 version) with the same result --- Matlab does not seem to see the coprocessor.

Looks like I am missing something stupid. If anybody has a working offload of Matlab MKL, please do not hesitate to provide a hint.