Performance study of graphic processors usage in computational nanotechnology problems

Computational nanotechnology is strongly connected with modern high-performance computing. With fast evolution of graphical processing units (GPU), general-purpose computing (GPGPU) became a popular choice for computationally demanding tasks. Tasks in nanomaterial researches with mass-spectrometers usage for analysis and material creation are good examples of such kind of tasks. Paper is devoted to ions behavior modeling in traps of mass-spectrometers based on Fourier transform. We use heterogeneous computational systems with GPU inside for calculation. Particle-in-cell model is used for direct modeling of ions behavior. We also use two GPU libraries: CULA and cuFFT.
Paper shows, that some GPU-oriented libraries could significantly ease the development of parallel algorithms for GPU and allow to get good performance of parallel applications. Calculations were performed on several systems including “Lomonosov” supercomputer in Moscow State University. Paper shows, that different strategies of mapping parallel processes to nodes could significantly effects on performance because of parallel access to an every single GPU from multiple processes.
Results obtained in this work could be useful for big molecular structure modeling, for solving computational nanotechnology problems on modern high-performance parallel computational systems with GPU.