Using Python for Mapping Molecular Dynamics Simulation Openmm Onto the Xeon Phi Architecture

In seeking to accelerate the mathematically intense engineering and scientific applications, parallel computing is used to bypass the physical limitations of traditional stand-alone CPU based systems. This has led to the creation of new hardware and software paradigms developed in tandem for future performance gains. The thrust of this paper is to investigate the performance of applications and code run in two distinct hardware configurations: standalone 16-Core 2.3 GHz Intel Xeon E5-2968 CPU and the Xeon E5 coupled with dual Xeon Phi 7120 61-Core 1.25GHz Intel Many Integrated Core (MIC) Architecture. The hardware configurations were implemented under different software environments: standard Python, Intel's distribution of Python and both Python versions with Automatic Offloading enabled. The results revealed a significant increase in application performance when implemented under Intel's distribution of Python combined with the offloading of computationally demanding portions of the code to the host co-processors.