Upcoming Event: Oden Institute Seminar

Simulation of fluid flow: From exascale to cell phone

Spencer Bryngelson, Professor, UT Austin, Department of Mechanical Engineering

Abstract

Fluid flows require modeling, and that modeling can be complex. Consider multiple species—water, air, hydrocarbons, and so on—fast and slow characteristic speeds, and small and large features. The equations one uses to describe these flows, often of the Navier–-Stokes family, require numerical treatment to admit simulation. Further, they beg numerical taming for efficient simulation. Computational engineers and mathematicians have spent many decades developing the necessary tools. The effort has been largely successful; inspect your prestigious colleagues should you have doubt.

CFD practitioners consume a lion's share of the computational resources at the world's largest supercomputing centers, perhaps in accordance with Jevons' paradox. It is unclear that high-fidelity simulation needs remain synonymous with such machines or their traditional use. This talk is about some of the field's shortcomings and possible avenues to something better. Or, at least, something different.

I will discuss PDE regularization as a means of eliding complex numerical treatments. I will consider the viability of reduced-precision computation, reframing some classical numerics for emerging architectures, including the tensor (matrix) units that occupy ever greater real estate on datacenter hardware. At striking odds with kilowatt GPUs, I consider the energy-efficient neural accelerators sitting in our cell phones, tablets, and laptops. Are these useful? I will opine on quantum architectures and the computational odyssey they entail.

Perhaps somewhere among these excursions lies a more fortuitous computational future.

[All software discussed in this talk is openly available on GitHub and under the MIT license.]

Biography

Spencer Bryngelson is core Oden faculty, homed in Mechanical Engineering. His group works to resolve grand challenge engineering problems via computation. They were a 2025 Gordon Bell Prize finalist. He is the institutional PI for two DOE PSAAP IV centers. Spencer has been a visitor at Pacific Northwest National Lab, Stanford, and MIT. He was a postdoc at Caltech and earned his Ph.D. from the University of Illinois at Urbana-Champaign. His honors include the ORAU Powe Award for Junior Faculty, the Illinois Outstanding Dissertation Award, and the Hassan Aref Prize, among others. He serves on the executive board of the Oak Ridge Computing Facility.