One of the undisputed highlights of SC each November is the acclaimed ACM Gordon Bell Prize in High Performance Computing, which since 1987 has recognized the most innovative applications of parallel computing to challenges in science, engineering, and large-scale data analytics. This year, five outstanding research efforts competed for the honor, which comes with a $10,000 prize, but just one was crowned champion at the awards ceremony on Thursday in Austin.
This year’s prize went to a team of scientists from the University of Texas at Austin, IBM Research, New York University and the California Institute of Technology for their design of an implicit solver that scaled to nearly 1.6 million cores on “Sequoia,” the IBM BlueGene/Q located at Lawrence Livermore National Laboratory. This breakthrough opens up a new era for understanding mantle convection and associated plate tectonics — which control earthquakes, volcanoes, mountain formation and long-term sea level change.
“Here we present a new implicit solver that exhibits optimal algorithmic performance and is capable of extreme scaling for hard PDE problems, such as mantle convection,” they write in their paper, “An Extreme-Scale Implicit Solver Complex PDEs: Highly Heterogeneous Flow in Earth’s Mantle.” The advanced feature set posed an immense scalability challenge, and still the team “demonstrate[d] that—contrary to conventional wisdom—algorithmically optimal implicit solvers can be designed that scale out to 1.5 million cores for severely nonlinear, ill-conditioned, heterogeneous, and anisotropic PDEs.”
The team’s solver scaled to all of Sequoia’s 1,572,864 cores with 97 percent parallel efficiency, which they reported in Figure 5 of their paper. The method enabled the team to simulate the “global instantaneous mantel flow in the entire earth with unprecedented accuracy.”
In addition to their seminal achievement on “Sequoia,” the IBM BlueGene/Q, the team also ran their simulation on another BG/Q system, JUQUEEN at Jülich Research Centre, which has nearly a half-million cores. They developed the algorithm, solvers, and science result visualizations on “Stampede.” Housed at the Texas Advanced Computing Center (TACC), the Dell PowerEdge Xeon Phi-boosted cluster is one of the most powerful machines in the world for open science research.
Now in a brand-new podcast, hosted by TACC Science Writer Jorge Salazar, two of the Gordon Bell prize winners — Johann Rudi and Omar Ghattas of the Institute of Computational Engineering and Sciences at UT Austin — discuss their work utilizing Stampede and Sequoia to carry out this groundbreaking research. They also comment on how it fits in with President Obama’s NSCI initiative, which is in the process of getting feedback from the community.
The podcast in its entirety can be found at https://www.tacc.utexas.edu/-/sc15-acm-gordon-bell-prize-winners-supercompute-deep-earth.