Implementation of molecular dynamics method using NVIDIA CUDA technology

Simulation of systems represented by a large number of atoms poses an exceptional interest. However, the simulation time grows proportionally to the square of the atoms’ number. Accordingly, the time needed for the calculation of a large system on a Central Processing Unit(CPU) becomes unacceptably high. We studied the possibilities that can provide acceleration of molecular dynamics simulations. Since the use of the parallel program model is the best way to increase computing performance we used the most convenient and accessible technology - NVIDIA CUDA. This technology allows speeding up calculation with the use of NVIDIA Graphics Processing Unit (GPU). We present the results of molecular dynamics simulations using NVIDIA GF114 GPU. The study tested GPU and CPU implementations of the molecular dynamics method. We ran series simulations with a different number of atoms (up to 18000 atoms) to determine the performance gains. The study showed that there is a significant fluency of parallel organization of calculations on its execution time. Moreover, performance reaches its maximum in case of the well-optimized data access to high latency global memory of a GPU. Due to the use of a relatively cheap graphics’ card for a molecular dynamics simulation the computing performance can be improved up to 60 times for double precision and more than 200 times for single precision. The designed scheme of parallel computation is quite efficient and can be easily applied for similar problems.