NERSC Researcher Part of Gordon Bell Prize Finalist Collaboration

Rahulkumar Gayatri, an application performance specialist at the National Energy Research Scientific Computing Center who originally joined the facility in 2017 as a post-doctoral researcher, is co-author on a paper that is a finalist for the Gordon Bell Prize to be awarded at the SC21 conference in November.

Since 2019, Gayatri has been involved with the Exascale Atomistic Capability for Accuracy, Length, and Time (EXAALT) project, which is part of the NERSC Exascale Science Applications Program (NESAP) and the Exascale Computing Project. The EXAALT team is developing an exascale-scalable molecular dynamics simulation platform that uses classical models such as the Spectral Neighbor Analysis Method (SNAP), which is a computationally intensive interatomic potential in the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS), a molecular dynamics code used in materials modeling. A key focus of EXAALT is to improve the performance of SNAP and LAMMPS on newer generation CPUs and GPUs, and the team has up-streamed their work to the LAMMPS library that includes SNAP and other potentials based on a variety of science and algorithms.

For the SC21 paper, a collaboration that included the University of South Florida, Sandia National Laboratories, Berkeley Lab, the KTH Royal Institute of Technology, and NVIDIA used components of this library to help produce billion atom molecular dynamics simulations of carbon at extreme conditions and experimental time and length scales. Running a 24-hour, 4,650-node production simulation on the Oak Ridge Leadership Computing Facility’s Summit system, the team applied a combination of a SNAP force kernel and the Kokkos CUDA backend to achieve a 20-billion-atom molecular dynamics simulation on the full Summit machine (27,900 GPUs).

“SNAP is now gaining quite a bit of popularity,” Gayatri said. “It is computationally intensive, but it has a bit more accuracy and allows us to change the parameters to fit a broad range of simulations. In addition, we have much more powerful machines today, so being computationally intensive is not an issue because what we are interested in is accuracy. And because we have powerful machines, we can trade off computationally intensive simulations in favor of accuracy.”

“This work represents a big step forward in terms of combined scale and fidelity in molecular dynamics,” said Jack Deslippe, who leads the application performance group at NERSC. It is also the third time in four years that a NESAP-related effort has been selected as a Gordon Bell finalist, he noted.

The authors on the SC21 Gordon Bell Prize finalist paper are: Kien Nguyen Cong, Jonathan T. Willman, and Ivan I. Oleynik of the University of South Florida; Stan G. Moore, Mitchell A. Wood, and Aidan Pl. Thompson of Sandia National Laboratories; Rahulkumar Gayatri of Lawrence Berkeley National Laboratory; Anatoly B. Belonoshko of the KTH Royal Institute of Technology; and Evan Weinberg of NVIDIA.

The Gordon Bell Prize recognizes outstanding achievement in high performance computing. The purpose of the award is to track the progress over time of parallel computing, with particular emphasis on rewarding innovation in applying high performance computing to applications in science, engineering, and large-scale data analytics.