Any idea that MHD Dynamo Benchmark while it runs perfectly (scalability) on Nehalem X5570 nodes, while its performance is miserable in Sandy Bridge (E5-2450, E5-2670 etc.)
It performs upto 4-cores per socket.
We're planning for EPSD products (Swift Current Pass [S2400SC2]) and 2xE5-2450 with Intel/QLogic QDR dual-port to provide ~128 compute cores. On the software side, Intel Fortran Composer XE 2013 with Intel MKL , Intel MPI library 4.1.1, UniData NetCDF 4.0, Fast Fourier Transformation W Library 3.3.3 on a CentOS 6.2x64
My advanced thanks for your view.
email@example.com (girish at njgroup dot net)