high performance computing on graphics processing units: hgpu.org

Currently, medical research for the discovery of new drugs is increasingly using Virtual Screening (VS) methods. In these methods, the calculation of the non-bonded interactions, such as electrostatic or van der Waals, plays an important role, representing up to 80% of the total execution time. These are computationally intensive operations, and massively parallel in nature, so they perfectly fit in the new landscape of high performance computing, dominated by massively parallel architectures. Among those architectures, the latest releases by Intel and Nvidia – Xeon Phi and K20x (Kepler), respectively – are extremely interesting in terms of both performance and complexity. In this work, we discuss the effective parallelization of the non-bonded electrostatic computation for VS, and evaluate its performance on these two architectures. We empirically demonstrate that both GPUs and Intel Xeon Phi are well suited architectures for the acceleration of non-bonded interaction kernels. Further, we observe that single precision calculations for relatively small sized systems are more suitable for GPUs (K20x completely outperforms Xeon Phi), while for large systems, they achieve a similar order of magnitude performance.