I have a problem with the current MKL version (the one shipping with the Intel C++ Composer XE 2013 Update 4). I am using the C++ interface for the ssyev function on a 16384x16384 matrix.
Everything seems to be alright when I link against the sequential MKL libraries. I have also tested to use the multithreaded version while I explicitly set the number of threads. The function seems to work as long as I don't select more than 2 threads.
In case it is is important: I am using OpenMP in my project but the call to ssyev is not in a parallelized section of the code. I am using Visual Studio 2012 on Windows 8.
Any help would be appreciated