PI: John J. Low
Objective: Atomistic calculations consume a large fraction of the computer time at High Performance Computer Centers. The next generation of high performance computers will be using the GPUs and MICs to reach higher levels of performance. Intel has published recipes for building many of the community software packages for atomistic simulations, such as, NAMD, LAMMPS, Quantum Espresso, CP2K and BigDFT. In this project, will we evaluate how efficiently these packages run on new architectures and learn how to build and run this software efficiently on GPUs and MICs. This will prepare us to support computational chemistry, materials science and biology applications on the next generation of high performance servers at Argonne.
Testbed description: The systems Jenny, Ruth and Lucie are Supermicro Superservers with dual Xeon processors and Phi coprocessors. Intel Parallel Studio and Intel MPI will be required to build and run the computational chemistry applications.