Skip to content

Latest commit

 

History

History
75 lines (67 loc) · 3.67 KB

README.md

File metadata and controls

75 lines (67 loc) · 3.67 KB
Raw

DOI

Swirl-LM: Computational Fluid Dynamics in TensorFlow

This is not an official Google product

Swirl-LM is a computational fluid dynamics (CFD) simulation framework that is accelerated by the Tensor Processing Unit (TPU). It solves the three dimensional 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 is implemented in TensorFlow.

Installation

To use Swirl-LM, you will need access to TPUs on Google Cloud. For small simulations, the easiest way to access TPUs is to use Google Colab. To see a demo, you can open one of the example notebooks and follow the notebook's instructions.

To run large simulations, you will need to create TPU Nodes or VMs in your Google Cloud project. See the instructions for the stand-alone demo on how to set up TPU Nodes and the docs about Cloud TPUs to set up TPM VMs.

Citation

If you extend or use this package in your work (except the components in the ext subpackage, in which case please reference the information within that subpackage), please cite the paper as

@ARTICLE{Wang2022-ln,
  title     = "A {TensorFlow} simulation framework for scientific computing of
               fluid flows on tensor processing units",
  author    = "Wang, Qing and Ihme, Matthias and Chen, Yi-Fan and Anderson,
               John",
  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.",
  journal   = "Comput. Phys. Commun.",
  publisher = "North-Holland",
  volume    =  274,
  pages     = "108292",
  month     =  may,
  year      =  2022,
  issn = {0010-4655},
  doi = {https://doi.org/10.1016/j.cpc.2022.108292},
  url = {https://www.sciencedirect.com/science/article/pii/S0010465522000108},
  keywords  = "Tensor processing unit; TensorFlow; Computational fluid
               dynamics; High performance computing"
}