Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

Please note that all times are shown in the time zone of the conference. The current conference time is: 18th Aug 2022, 09:19:32pm CEST

Session Overview
High Performance Computing LBM Implementation
Wednesday, 29/June/2022:
9:30am - 10:50am

Session Chair: Martin Geier, TU Braunschweig
Location: Michel Crépeau's Lecture Hall, Pôle Communication, La Rochelle University

Pôle Communication Multimédia Reseaux, La Rochelle University, 44 Avenue Albert Einstein, La Rochelle.

External Resource:
Show help for 'Increase or decrease the abstract text size'
9:30am - 9:50am

GPU implementation of a discrete unified gas kinetic scheme for low-speed isothermal flows

Liu, Peiyao1; Huang, Changsheng2; Guo, Zhaoli1

1State Key Laboratory of Coal Combustion, Huazhong University of Science and Technology, Wuhan, China; 2School of Mathematics and Physics, China University of Geosciences (Wuhan), Wuhan, China

A general purpose discrete unified gas kinetic scheme (Guo et al., 2013 Phys. Rev. E) implementation for low-speed isothermal flows is developed on a single Graphics Processing Unit (GPU). The implementation has been validated by simulating lid-driven cavity flow. The parallel performance improves significantly using the newly implemented approach. Compared with CPU serial calculation method, a speedup of 30 times of magnitudes has been found. The results show that the parallel GPU provides an affordable alternative to large clusters for fluid dynamics prediction.

9:50am - 10:10am

GPU on non-uniform meshes: a modern implementation approach and detailed decomposition of performance

Latt, Jonas; Coreixas, Christophe

University of Geneva, Switzerland

We revisit the topic of mesh refinement on GPU in the lattice Boltzmann (LB) method by adopting a straightforward implementation approach based on C++17 Parallel Algorithms. The general usefulness of this approach for GPGPU implementation of LB codes had previously been explored using the open-source demo code STLBM [1] ( More recently, modifications were made to the open-source library Palabos [2] so that simulations based on single and multiple GPUs could be possible thanks to its already existing MPI backend ( While the latter works were restricted to uniform meshes, this talk further discusses how to extend the above methodology to non-uniform grids based on quad-tree refinement algorithms. To establish a baseline for the analysis of the performance to be expected from a non-uniform GPU LB code, the straightforward grid-refinement approach by Rohde et al [3] is used which, at large Reynolds numbers, can lead to results of comparatively good quality. This is illustrated in the case of a flow around a high-lift airfoil and in the case of an air flow over a cavity, with acoustic effects.

Very interestingly, we show that through application of a memory saving neighbor-link implementation to incompressible flow simulations, a non-uniform mesh performs only slightly less well than a uniform one in terms of raw cell-processing performance, with a performance loss (in lattice-site updates per second) of 25% at single precision and 10% at double precision. In a second stage, we explore the extension of the approach to multi-speed, compressible flow simulations.

[1] Latt et al. Cross-platform programming model for many-

core lattice Boltzmann simulations. Plos One (2021).

[2] Latt et al. Palabos: Parallel Lattice Boltzmann Solver. CAMWA (2021).

[3] Rohde et al. A generic, mass-conservative local grid refinement technique for lattice Boltzmann schemes. Int. J. Num. Meth. Fluids (2006).

10:10am - 10:30am

HPC Performance study of different collision models using the Lattice Boltzmann solver Musubi

Spinelli, Gregorio Gerardo1; Masilamani, Kannan1; Horstmann, Tobias2; Soni, Malav3; Klimach, Harald Günther1; Stück, Arthur1; Roller, Sabine1

1DLR, German Aerospace Center, Institute of Software Methods for Product Virtualization, Dep. of Simulation Frameworks, Zwickauer Straße 46, 01069 Dresden, Germany; 2DLR, German Aerospace Center, Institute of Propulsion Technology, Dep. of Engine Acoustics, Bismarckstraße 101, 10625 Berlin, Germany; 3DLR, German Aerospace Center, Institute for Aerodynamics and Flow Technology, Dep. of Technical Acoustics, Lilienthalplatz 7, 38108 Braunschweig, Germany

Lattice Boltzmann (LB) solvers have gained much attention in the last decades as an alternative approach to the classical Navier-Stokes solvers. The open-source LB solver “Musubi” is designed for scale resolving CFD. Key features are good scalability on HPC systems, static mesh refinement and advanced collision models such as Cumulant and Hybrid Recursive Regularized BGK. We will showcase the application of Musubi to solve aerodynamics and aeroacoustics fields using two test cases: the flow around the cylinder at Reynolds number (Re) = 3900 and the Boeing rudimentary landing gear (RLG) defined on the 2010 BANC-I workshop at a wheel Re = 10^6, for which we could achieve a stable simulation at Re = 10^5. While the cylinder case is used to validate different collision models, the RLG case will provide insights into performance and scalability characteristics of the different collision models in conjunction with a typical target configuration on state-of-the-art HPC systems. Additionally, we will use the RLG case to present its competitiveness in terms of performance and scalability.

10:30am - 10:50am

High-accuracy curved lattice Boltzmann boundary conditions for efficient GPU simulations

Marson, Francesco1; Ginzburg, Irina2; De Santana Neto, José Pedro1; Silva, Gonçalo3; Latt, Jonas1

1University of Geneva, Switzerland; 2University Paris-Saclay, INRAE, France; 3University of Évora, Portugal

With the lattice Boltzmann Method (LBM), a straightforward extension of the bounce-back no-slip rule leads to a vast group of directional (aka link-wise) Dirichlet boundary conditions. These methods can accurately describe the interaction of a fluid flow with a complex solid surface in a uniform Cartesian grid.

The link-wise family includes an infinite number of schemes that can be tuned to optimize their five fundamental features: (i) exactness of channel flows velocity and pressure profiles, (ii) accuracy order, (iii) linear stability, (iv) parametrization, and (v) locality.

We identify, respectively, two main groups of directional schemes. The first comprises compact schemes with linear exactness (LI) (Ginzburg et al., 2008) and its extended local (ELI) family (Ginzburg et al., 2022; Marson, 2022; Marson et al., 2021), which combines the local single-node implementation with the second-order accuracy, linear exactness, and physical consistency. The second is the Multi-reflection (MR) family, which gains parabolic exactness in the Stokes and Navier-Stokes flows but loses locality.

This presentation focuses on the new developments of ELI, which provide ELI (and LI) with parabolic exactness in Stokes flows. Additionally, it shows how these schemes allow for simple and uniform implementations, even in the coarsest simulations characterized by many narrow gaps.

One can implement ELI simply by modifying the collision model in the boundary node where the half-way bounce-back (HW) applies. Therefore, the implementation in a high-performance code is straightforward. We implement ELI in GPU-accelerated Palabos (Latt et al., 2021a, 2021b), which allows for multi-GPU simulations, showing that one can obtain outstanding numerical performances which are close to the ones of the HW for steady boundaries.

Ginzburg, I., Silva, G., Marson, F., Chopard, B., Latt, J., 2022. Unified directional parabolic-accurate Lattice Boltzmann boundary schemes for grid-rotated narrow gaps and curved walls in creeping and inertial fluid flows. Phys. Rev. E Submitted. 51.

Ginzburg, I., Verhaeghe, F., d’Humières, D., 2008. Two-Relaxation-Time Lattice Boltzmann Scheme: About Parametrization, Velocity, Pressure and Mixed Boundary Conditions. Commun Comput Phys 3, 427–478.

Latt, J., Malaspinas, O., Kontaxakis, D., Parmigiani, A., Lagrava, D., Brogi, F., Belgacem, M.B., Thorimbert, Y., Leclaire, S., Li, S., Marson, F., Lemus, J., Kotsalos, C., Conradin, R., Coreixas, C., Petkantchin, R., Raynaud, F., Beny, J., Chopard, B., 2021a. Palabos: Parallel Lattice Boltzmann Solver. Comput. Math. Appl., Development and Application of Open-source Software for Problems with Numerical PDEs 81, 334–350.

Latt, J., Marson, F., De Santana Neto, J.P., Thyagarajan, K., Coreixas, C., Chopard, B., 2021b. From CPU to GPU in 80 days.

Marson, F., 2022. Directional lattice Boltzmann boundary conditions.

Marson, F., Thorimbert, Y., Chopard, B., Ginzburg, I., Latt, J., 2021. Enhanced single-node lattice Boltzmann boundary condition for fluid flows. Phys. Rev. E 103, 053308.

Contact and Legal Notice · Contact Address:
Privacy Statement · Conference: ICMMES 2022
Conference Software - ConfTool Pro 2.6.144
© 2001–2022 by Dr. H. Weinreich, Hamburg, Germany