An Adaptive Explicit 3D Discontinuous Galerkin Solver for Unsteady Problems&lt;br /&gt;

Kirby, Andrew C.; Mavriplis, Dimitri J.; Wissink, Andrew M.

doi:10.2514/6.2015-3046

Cited by 9 publications

(4 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…W 2 A 2 KE3D utilizes a variable order DG FEM implemented in a dynamic adaptive mesh refinement (AMR) framework. In previous work, W 2 A 2 KE3D used an off-body solver known as SAMCartDG based on the open-source SAMRAI (Wissink et al) AMR package provided by Lawrence Livermore National Laboratory with the standalone DG solver CartDG (Kirby et al, 2015). SAMRAI is a patch-based AMR system.…”

Section: Off-body Flow Solvermentioning

confidence: 99%

“…Through this approach, the numerical complexity is reduced. The details of the algorithm are outlined in Kirby et al (2015) which follows the derivation in Diosady and Muman (2014) and Hindenlang et al (2012). The off-body solver dg4est has been demonstrated on various problems such as flow over the three-dimensional wing NACA0015 at moderate Reynolds number and on the Taylor-Green Vortex flow problem (Brazell et al, 2017).…”

Section: Off-body Flow Solvermentioning

confidence: 99%

See 1 more Smart Citation

Wind farm simulations using an overset hp-adaptive approach with blade-resolved turbine models

Kirby

Brazell

Yang

et al. 2019

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

Blade-resolved numerical simulations of wind energy applications using full blade and tower models are presented. The computational methodology combines solution technologies in a multi-mesh, multi-solver paradigm through a dynamic overset framework. The coupling of a finite volume solver and a high-order, hp-adaptive finite element solver is utilized. Additional technologies including in-situ visualization and atmospheric microscale modeling are incorporated into the analysis environment. Validation of the computational framework is performed on the National Renewable Energy Laboratory (NREL) 5MW baseline wind turbine, the unsteady aerodynamics experimental NREL Phase VI turbine, and the Siemens SWT-2.3-93 wind turbine. The power and thrust results of all single turbine simulations agree well with low-fidelity model simulation results and field experiments when available. Scalability of the computational framework is demonstrated using 6, 12, 24, 48, and 96 wind turbine setups including the 48 turbine wind plant known as Lillgrund. The largest case consisting of 96 wind turbines and a total of 385 overset grids are run on 44,928 cores at a weak scaling efficiency of 86%. Demonstration of the coupling of atmospheric microscale and Computational Fluid Dynamics (CFD) solvers is presented using the National Center for Atmospheric Research (NCAR) Weather Research and Forecasting Model (WRF) solver and the NREL Simulator fOr Wind Farm Applications (SOWFA) solver.

show abstract

Section: Off-body Flow Solvermentioning

confidence: 99%

Section: Off-body Flow Solvermentioning

confidence: 99%

Wind farm simulations using an overset hp-adaptive approach with blade-resolved turbine models

Kirby

Brazell

Yang

et al. 2019

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

show abstract

“…The computational kernel within the DG4EST component, known as CartDG [23]- [25], is the primary focus of this work. CartDG is a discontinuous Galerkin method designed for computational efficiency on Cartesian meshes utilizing tensorproduct collocation-based basis functions.…”

Section: Cpu Implementationmentioning

confidence: 99%

GPU-Accelerated Discontinuous Galerkin Methods: 30x Speedup on 345 Billion Unknowns

Kirby¹,

Mavriplis²

2020

Preprint

Self Cite

View full text Add to dashboard Cite

A discontinuous Galerkin method for the discretization of the compressible Euler equations, the governing equations of inviscid fluid dynamics, on Cartesian meshes is developed for use of Graphical Processing Units via OCCA, a unified approach to performance portability on multi-threaded hardware architectures. A 30x speedup over CPU-only implementations using non-CUDA-Aware MPI communications is demonstrated up to 1,536 NVIDIA V100 GPUs and parallel strong scalability is shown up to 6,144 NVIDIA V100 GPUs for a problem containing 345 billion unknowns. A comparison of CUDA-Aware MPI communication to non-GPU-direct communication is performed demonstrating an additional 24% speedup on eight nodes composed of 32 NVIDIA V100 GPUs.

show abstract

“…The Computational Fluid Dynamics solver is known as CartDG [20]- [22] which is a Discontinuous Galerkin method that discretizes the compressible Navier-Stokes equations. CartDG has been used in the high-fidelity simulation of wind energy applications [23], [24] and in aerospace applications including rotorcraft wakes [25].…”

Section: B Computational Fluid Dynamicsmentioning

confidence: 99%

Benchmarking network fabrics for data distributed training of deep neural networks

Samsi

Prout

Jones

et al. 2020

2020 IEEE High Performance Extreme Computing Conference (HPEC)

View full text Add to dashboard Cite

Artificial Intelligence/Machine Learning applications require the training of complex models on large amounts of labelled data. The large computational requirements for training deep models have necessitated the development of new methods for faster training. One such approach is the data parallel approach, where the training data is distributed across multiple compute nodes. This approach is simple to implement and supported by most of the commonly used machine learning frameworks. The data parallel approach leverages MPI for communicating gradients across all nodes. In this paper, we examine the effects of using different physical hardware interconnects and network-related software primitives for enabling data distributed deep learning. We compare the effect of using GPUDirect and NCCL on Ethernet and OmniPath fabrics. Our results show that using Ethernet-based networking in shared HPC systems does not have a significant effect on the training times for commonly used deep neural network architectures or traditional HPC applications such as Computational Fluid Dynamics.

show abstract

An Adaptive Explicit 3D Discontinuous Galerkin Solver for Unsteady Problems<br />

Cited by 9 publications

References 43 publications

Wind farm simulations using an overset hp-adaptive approach with blade-resolved turbine models

Wind farm simulations using an overset hp-adaptive approach with blade-resolved turbine models

GPU-Accelerated Discontinuous Galerkin Methods: 30x Speedup on 345 Billion Unknowns

Benchmarking network fabrics for data distributed training of deep neural networks

Contact Info

Product

Resources

About