Graduate Seminars

back Back to all Graduate Seminars

CS Graduate Seminar: High-Performance Computational Geophysics

Start Date: October 15, 2018
End Date: October 15, 2018

By Dr. Alexander Breuer (University of California, San Diego)   
Efficient and scalable software is in urgent demand in all of science and industry. I will summarize three examples, where the most capable solution is obtained by integrating applied mathematics, computer science, and physical science. First, I discuss Local Time Stepping (LTS) for seismic simulations, relying on the Discontinuous Galerkin Finite Element Method (DG-FEM) and unstructured meshes. By clustering elements with similar time steps, the resulting petascale solver is able to exploit theoretical LTS speedups without sacrificing hardware efficiency. The second example exploits the fact that many grand challenges in computational geophysics, e.g., inversions or seismic hazard assessment, require large ensembles of geometrically complex but similar forward simulations. This viewpoint of the simulation pipeline allows the forward solver to operate on a set of different inputs, e.g., different seismic sources, in parallel. The deep integration of inter-problem parallelism speeds up DG-FEM by a convergence-rate dependent factor of 2 to 5, while sustaining a multi-PFLOPS machine utilization. In the third example, I will present recent work that pushes DG-FEM for seismic wave propagation towards reduced precision. Supported by a comprehensive verification study, the utilization of single-precision tensor instructions illustrates the importance of inter-simulation parallelism for emerging deep learning hardware. In summary, the combined software accelerates the simulation throughput on the Intel Xeon Phi for Deep Learning by over four times over the state-of-the-art. I will conclude the presentation by discussing three system-level research goals: 1) Develop and improve extreme-scale seismic forward solvers to meet demanding engineering requirements, 2) Generalize and apply high-performance technology to other scientific domains, 3) Advance end-to-end techniques and software to high-dimensional problems in physical science.

Biography: Alexander’s research identifies and develops software and algorithms capable of solving today's and tomorrow's challenges in computational science and engineering. His work covers the vertically integrated disciplines from the bare metal of computer architectures to fully automated production workflows, by including modeling and simulation, high performance computing, software engineering, data analytics, and verification and validation. In 2014 Alexander was honored with an ACM/IEEE-CS George Michael Memorial HPC Fellowship for his Ph.D. project "High Performance Earthquake Simulations". In addition, he and his collaborators have been given the PRACE ISC Award and nominated as ACM Gordon Bell finalists. Alexander holds a doctoral degree from the Technical University of Munich. Currently, he is collaborating with the Southern California Earthquake Center as a Postdoctoral Research Scholar at the University of California, San Diego.

More Information:

For more info contact: Prof. David Keyes : email:
Date: Monday 15th Oct 2018
Time:12:00 PM - 01:00 PM
Location: Engineering Science Hall (bldg.9), Level 2, Hall 1
Refreshments: Light Lunch will be available at 11:45 am