Molecular dynamics (MD) is an important research tool extensively applied inmaterials science. Running MD on a graphics processing unit (GPU) is anattractive new approach for accelerating MD simulations. Currently, GPUimplementations of MD usually run in a one-host-process-one-GPU (OHPOG) scheme.This scheme may pose a limitation on the system size that an implementation canhandle due to the small device memory relative to the host memory. In thispaper, we present a one-host-process-multiple-GPU (OHPMG) implementation of MDwith embedded-atom-model or semi-empirical tight-binding many-body potentials.Because more device memory is available in an OHPMG process, the system sizethat can be handled is increased to a few million or more atoms. In comparisonwith the CPU implementation, in which Newton's third law is applied to improvethe computational efficiency, our OHPMG implementation has achieved a28.9x~86.0x speedup in double precision, depending on the system size, thecut-off ranges and the number of GPUs. The implementation can also handle agroup of small boxes in one run by combining the small boxes into a large box.This approach greatly improves the GPU computing efficiency when a large numberof MD simulations for small boxes are needed for statistical purposes.
展开▼
