Schematics of the 1D decomposition showing a cubic domain divided into 4 tiles identified by color code and process IDs 0,1,2,3. my = mz = N/P denotes the subdivided range of grid points present in the respective directions of each tile. Credit: Computational Physics Communications (2024). DOI: 10.1016/j.cpc.2024.109364
From the water that comes out of the tap to the chemical reactions in the jet engines that power planes, turbulence affects our daily lives. Georgia Tech researchers study the complex physics of turbulence in simplified contexts that could help us better understand nature and engineering.
At its core, turbulence comprises disordered fluctuations over a wide range of scales, both in time and in three-dimensional space. These complexities mean that many fundamental aspects are still not understood. Computers can help unravel the mystery, but direct numerical simulations based on exact physical laws have always been very resource-intensive. Their challenges are greatest when it comes to studying rare and very large fluctuations.
Today, Frontier, the world’s first – and still fastest – exascale computer, capable of performing a quintillion operations per second, is helping researchers better understand turbulence.
“Turbulence is very complex, theories are incomplete and laboratory measurements are arduous,” said PK Yeung, professor in the Daniel Guggenheim School of Aerospace Engineering with a joint courtesy appointment in the School of Mechanical Engineering George W. Woodruff.
“State-of-the-art resolution of more than 5 trillion grid points on Frontier is expected to lead to new discoveries, which in turn can facilitate advances in modeling where hypotheses and predictions can be tested numerically.”
Yeung and his team accessed Frontier, located at Oak Ridge National Laboratory, when it first came online and also received significant allocations of machine time from the INCITE program, managed by the Office of Science of US Department of Energy. The power of Frontier lies mainly in powerful graphics processing units (GPUs), which calculate quickly.
Yeung’s group has published a journal article describing a high-performance algorithm, specifically designed to make the most of Frontier’s features to make extremely high-resolution simulations feasible and efficient. The research is published in the journal Computational Physics Communications.
“In many scientific fields, people thought calculations of this magnitude were not possible, but now we are there, perhaps sooner than expected,” Yeung said.
“Our work on turbulence simulations also demonstrates several advanced GPU programming principles of interest in other areas, particularly those where so-called pseudo-spectral methods are important. The scientific impacts of our extreme-scale simulations are expected to be further enhanced by public data. -shared in partnership with the Johns Hopkins Turbulence Database Project supported by the National Science Foundation.
More information:
PK Yeung et al, GPU-compatible extreme-scale turbulence simulations: pseudo-spectral Fourier algorithms at exascale using OpenMP offloading, Computational Physics Communications (2024). DOI: 10.1016/j.cpc.2024.109364
Provided by Georgia Institute of Technology
Quote: Researchers achieve world record resolution in turbulence simulations (October 10, 2024) retrieved October 10, 2024 from
This document is subject to copyright. Except for fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for informational purposes only.
