Boosting materials science simulations by high performance computing

Abstract

Technology development is often limited by knowledge of materials engineering and manufacturing processes. This scenario spans across scales and disciplines, from aerospace engineering to MicroElectroMechanical Systems (MEMS) and NanoElectroMechanical Systems (NEMS). The mechanical response of materials is dictated by atomic/nanometric scale processes that can be explored by molecular dynamics (MD) simulations. In this work we employ atomistic simulations to prove indentation as a prototypical deformation process showing the advantage of High Performance Computing (HPC) implementations for speeding up research. Selecting the right HPC hardware for executing simulations is a process that usually involves testing different hardware architectures and software configurations. Currently, there are several alternatives, using HPC cluster facilities shared between several researchers, as provided by Universities or Government Institutions, owning a small cluster, acquiring a local workstation with a high-end microprocessor, and using accelerators such as Graphics Processing Units (GPU), Field Programmable Gate Arrays (FPGA), or Intel Many Integrated Cores (MIC). Given this broad set of alternatives, we run several benchmarks using various University HPC clusters, a former TOP500 cluster in a foreign computing center, two high-end workstations and several accelerators. A number of different metrics are proposed to compare the performance and aid in the selection of the best hardware architecture according to the needs and budget of researchers. Amongst several results, we find that the Titan X Pascal GPU has a ∼3 x speedup against 64 AMD Opteron CPU cores.