Thursday, February 22, 2018
Dr. Freddie Witherden
Stanford University
2:00 - 3:00pm
SEH, B1220
Abstract
Accurate simulation of unsteady turbulent flow is critical for the improved design of greener aircraft that are quieter and more fuel-efficient. In this presentation, we will show how recent advances in high-order numerical methods and high-performance computing are enabling scale resolving simulations of hitherto intractable flows within the vicinity of complex engineering geometries. Our vehicle for this will be PyFR: an open source Python-based framework for solving the compressible Navier-Stokes equations using the high-order flux reconstruction approach. Unlike traditional high-performance computing codes, PyFR is written almost entirely in Python and is one of the first extreme-scale codes to make extensive use of run-time code generation. This enables PyFR to achieve unrivaled performance across a range of modern hardware platforms. Results will be presented for various benchmark and real-world flow problems. Comparisons with existing results in the literature will be made. Additionally, we will also showcase the scalability of PyFR, including its ability to sustain in excess of 13 PFLOP/s of computation on the Titan supercomputer. Further, we will also outline recent developments we have made in the fields of in-situ visualization and post-processing. Current challenges and future directions within computational fluid dynamics, and computational mechanics, in general, will also be discussed.