A TensorFlow simulation framework for scientific computing of fluid flows on tensor processing units

DOI10.1016/J.CPC.2022.108292zbMATH Open1518.76041arXiv2108.11076OpenAlexW4226276902WikidataQ114192706 ScholiaQ114192706MaRDI QIDQ6159635FDOQ6159635

Author name not available (Why is that?)

Publication date: 20 June 2023

Published in: Computer Physics Communications (Search for Journal in Brave)

Abstract: A computational fluid dynamics (CFD) simulation framework for fluid-flow prediction is developed on the Tensor Processing Unit (TPU) platform. The TPU architecture is featured with accelerated dense matrix multiplication, large high bandwidth memory, and a fast inter-chip interconnect, making it attractive for high-performance scientific computing. The CFD framework solves the variable-density Navier-Stokes equation using a low-Mach approximation, and the governing equations are discretized by a finite-difference method on a collocated structured mesh. It uses the graph-based TensorFlow as the programming paradigm. The accuracy and performance of this framework is studied both numerically and analytically, specifically focusing on effects of TPU-native single precision floating point arithmetic. The algorithm and implementation are validated with canonical 2D and 3D Taylor-Green vortex simulations. To demonstrate the capability for simulating turbulent flows, simulations are conducted for two configurations, namely decaying homogeneous isotropic turbulence and a turbulent planar jet. Both simulations show good statistical agreement with reference solutions. The performance analysis shows a linear weak scaling and a superlinear strong scaling up to a full TPU v3 pod with 2048 cores.

Full work available at URL: https://arxiv.org/abs/2108.11076

Recommendations

• A stable tensor-based method for controlled fluid simulations
• Flow simulation and high performance computing
• FluTAS: a GPU-accelerated finite difference code for multiphase flows
• Efficient flow simulation on high performance computers
• scientific article; zbMATH DE number 2121322
• Deep-learning accelerated calculation of real-fluid properties in numerical simulation of complex flowfields

zbMATH Keywords

finite difference methodvariable-density Navier-Stokes equationslow-Mach approximationaccelerated dense matrix multiplicationcollocated structured meshlarge high bandwidth memory

Mathematics Subject Classification ID

Finite difference methods applied to problems in fluid mechanics (76M20) Compressible Navier-Stokes equations (76N06)

Cites Work

• A stable and accurate convective modelling procedure based on quadratic upstream interpolation
• Numerical study of the turbulent flow past an airfoil with trailing edge separation
• Turbulent Flows
• Application of a fractional-step method to incompressible Navier-Stokes equations
• Consistent initial conditions for lattice Boltzmann simulations
• The Accuracy of Floating Point Summation
• Large calculation of the flow over a hypersonic vehicle using a GPU
• A comparison of vortex and pseudo-spectral methods for the simulation of periodic vortical flows at high Reynolds numbers

Cited In (3)

• Solving the discretised neutron diffusion equations using neural networks
• JAX-fluids 2.0: towards HPC for differentiable CFD of compressible two-phase flows
• Solving the discretised multiphase flow equations with interface capturing on structured grids using machine learning libraries

This page was built for publication: A TensorFlow simulation framework for scientific computing of fluid flows on tensor processing units

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6159635)