The Association for Computing Machinery has named two teams as finalists for the prestigious ACM Gordon Bell Prize, presented each year at the conference to recognize researchers who have made significant strides in applying high-performance computing systems to scientific applications.
The prize will be presented in November at SC19, the International Conference for High Performance Computing, Networking, Storage and Analysis.
Both finalist teams used the IBM AC922 Summit supercomputer at the U.S. Department of Energy's Oak Ridge National Laboratory to solve ab initio, or first-principles, calculations. These types of approaches attempt to solve the fundamental equations accurately within a specific domain — in this case, quantum mechanics and condensed matter physics — to predict the practical phenomena of interest with that domain.
"These first-principles equations are some of the leading problems in condensed matter and materials physics and are a huge topic in high-performance computing today, consuming lots of resources in data centers around the world," says Oak Ridge Leadership Computing Facility Director of Science Jack Wells. "It's really satisfying to see these two projects pushing back the frontier of what we considered possible on Summit be selected as finalists for the Gordon Bell Prize."
Powered by over 27,648 GPUs across 4,608 nodes, Summit has consistently made waves in the computing world for the speed at which it can solve data-intensive, complex computational problems. The 200-petaflop IBM AC922 system is the most powerful supercomputer in the world for open science. Summit is a driving force behind the 2019 Gordon Bell finalists.
The 2019 Gordon Bell Prize finalists are:
"A Data-Centric Approach to Extreme-Scale Ab Initio Dissipative Quantum Transport Simulations," by Alexandros Nikolaos Ziogas, Tal Ben-Nun, Guillermo Indalecio Fernandez, Timo Schneider, Mathieu Luisier, and Torsten Hoefler. The team, a group of researchers from ETH Zurich, is using Summit to improve the computational efficiency of an ab initio quantum transport solver designed to reveal the coupled electro-thermal properties of atomically resolved nanotransistors.
"Fast, Scalable, and Accurate Finite-Element Based Ab Initio Calculations Using Mixed Precision Computing," by Sambit Das, Phani Motamarri, Vikram Gavini, Bruno Turcksin, Ying Wai Li, and Brent Leback. This study, led by researchers from the University of Michigan, in collaboration with researchers from ORNL and Los Alamos National Laboratory, works with first-principles calculations based on density functional theory in metallic systems.
By employing finite-element discretization and mixed-precision strategies, these calculations can be sped up an order of magnitude faster than previous methods. The ability to handle these large-scale metallic systems efficiently benefits numerous application areas, including the design and discovery of new catalytic materials, studies on high entropy alloys, novel energy storage materials, and organometallic complexes in biomolecular electronics. The speedup also means that molecular dynamics simulations can run for timescales longer than what was previously possible.