The International Conference for High Performance Computing, Networking, Storage, and Analysis
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Workshop: Workshop on Artificial Intelligence and Machine Learning for Scientific Applications (AI4S)
Authors: Nariman Piroozan and Nalini Kumar (Intel Corporation)
Abstract: The ability to retain DFT-level accuracy and reduce the high computational costs has been made possible using Deep Potential models which allow accurate prediction of interatomic force and energy distributions, when trained on DFT data. DeePMD-kit is a Python/C++ package which implements such a model. In this paper, we extend DeePMD to accurately predict the thermal conductivity for crystalline Au and Ag systems of up to 2 million atoms. We demonstrate that both DeePMD training and DeePMD inference with LAMMPS can be run efficiently on CPU-based systems. On a single node of 4th generation Intel® Xeon® Scalable 9480 processors, we can train the model in less than 5 minutes. Using this trained model with LAMMPS on 128 dual-socket nodes with Intel® Xeon® Scalable 8480+ processors, we can accurately determine the thermal conductivity of crystalline Au and Ag systems, within 5% of experimental results, in under one hour.