Mkl is included in intel compiler installers, and also available from the mkl community download site. It comes with full documentation. If you have further questions, the companion mkl forum site might be appropriate.
From what little you have said, one might question whether you gave attention to your settings of omp_num_threads, omp_places, and omp_proc_bind. Mkl default for Xeon is omp_places=cores.
If you compare avx2 vs. Avx, evidently you might expect a large difference in performance of some blas functions.
