Implementation properties of many-body potentials on heteroheneous computing systems

One way to improve the performance of atomistic modeling algorithms is to transform them for running on heterogeneous computing systems that combine the CPU and GPU. Adequate calculation of the mechanical and transport properties of nanostructures using many-body interatomic potentials require significantly more processing power. The article describes the features of the implementation of GPU algorithm for calculating many-body Tersoff and embedded atom potential by means of the OpenCL technology. The performance of different algorithms for GPU computing is compared.