Nassu

A GPU-native, high-performance LES solver for computational wind engineering.

Nassu is the lattice-Boltzmann CFD engine behind the AeroSim digital wind tunnel. It runs large eddy simulations (LES) of industrial aerodynamics natively on NVIDIA GPUs and is engineered around one idea: you never write a line of code to run a simulation. You describe the case in a single configuration file, and Nassu generates the specialised CUDA kernels, runs the simulation, and writes the fields for you.

The numerical framework, validation, and design are described in the peer-reviewed reference paper Oliveira et al.[1].

Turbulent wind flow around bluff bodies, resolved with Nassu’s LBM-LES framework.

What Nassu is built for

Nassu targets computational wind engineering (CWE) and external aerodynamics: wind loading on buildings, pedestrian wind comfort, pollutant and scalar dispersion, atmospheric boundary layer and terrain flows, and bluff-body aerodynamics in general. These problems demand the transient, peak-resolving fidelity of LES, which has traditionally been too expensive for routine engineering use. Nassu makes it practical on a single, affordable GPU by combining the lattice-Boltzmann method with a memory-efficient, recursive-regularized collision operator.

Where Nassu shines

No meshing step	The domain is a Cartesian lattice generated automatically from the configuration file: a 100-million-node grid is built in under 30 seconds, with none of the manual, error-prone meshing a body-fitted solver requires.
No watertight geometry	The immersed boundary method runs directly on raw STL surfaces or point clouds, without the clean, watertight, non-overlapping mesh that finite-volume solvers demand.
Realistic atmospheric inflow	Mean atmospheric-boundary-layer profiles with correlated synthetic turbulence (SEM) and PODFS precursor replay injected at the inlet.
High Reynolds numbers	The RR-BGK collision operator stays stable at Re > 10^5, the regime of full-scale wind engineering Oliveira et al.[].
Large domains on one GPU	Macroscopic-only storage fits roughly 10 million lattice nodes per gigabyte and cuts memory up to 50% versus a naive 3-D LBM, so an affordable 24 GB GPU holds an LES of about 240 million nodes.
GPU-native throughput	Over one billion lattice-node updates per second on a single modern GPU.
Validated	A curated portfolio of 20+ benchmark cases spanning seven physical categories, from analytical flows to wind-tunnel reproductions.

Capabilities

Physics and numerics

• Lattice-Boltzmann method with the recursive-regularized BGK (RR-BGK) collision operator (3rd-order Hermite), the default production LES operator; BGK, RBGK and hybrid HRRBGK are also available.

• Velocity sets D2Q9 (2-D) and D3Q15, D3Q19, D3Q27 (3-D), with D3Q27 as the preferred production set.

• Smagorinsky subgrid-scale model with the relaxation frequency derived analytically from the non-equilibrium stress Dong et al.[2].

Geometry and boundary conditions

• Immersed boundary method (IBM) with Lagrangian meshes generated directly from input STL files or point clouds Peskin[3].

• Wall models for the diffuse-interface IBM and equilibrium log-law / turbulent boundary-layer wall BCs, plus halfway and regularized halfway bounce-back, uniform inlet, and zero-gradient outlet conditions.

Turbulent inflow and scalar transport

• Synthetic Eddy Method (SEM) and PODFS precursor replay for realistic, correlated inlet turbulence Jarrin et al.[4].

• Passive scalar transport for dispersion studies, sharing the exact same multiblock and collision-streaming path as the fluid solver.

Grid and performance

• Static block-structured multiblock refinement on an octree grid with 2:1 level ratios Lagrava et al.[5].

• Runtime CUDA code generation: kernels are specialised per simulation configuration, with no runtime branching overhead.

• Memory-efficient macroscopic-only storage that runs a large CWE domain on a single GPU.

Tip

New to CFD or LBM? The textbook by Krüger et al.[6] is the recommended introduction and is referenced throughout the theory chapters.

---

[1]

Waine Oliveira, Rodrigo S. Romanus, and Alan Lugarini. Nassu: a high performance LES solver for computational wind engineering. Journal of Wind Engineering and Industrial Aerodynamics, 274:106465, 2026. doi:10.1016/j.jweia.2026.106465.

[2]

Yu-Hong Dong, Pierre Sagaut, and Simon Marie. Inertial consistent subgrid model for large-eddy simulation based on the lattice Boltzmann method. Physics of Fluids, 20(3):035104, 2008. doi:10.1063/1.2842379.

[3]

Charles S Peskin. The immersed boundary method. Acta numerica, 11:479–517, 2002.

[4]

Nicolas Jarrin, Sofiane Benhamadouche, Dominique Laurence, and Robert Prosser. A synthetic-eddy-method for generating inflow conditions for large-eddy simulations. International Journal of Heat and Fluid Flow, 27(4):585–593, 2006.

[5]

D. Lagrava, O. Malaspinas, J. Latt, and B. Chopard. Advances in multi-domain lattice boltzmann grid refinement. Journal of Computational Physics, 231(14):4808–4822, 2012. URL: https://www.sciencedirect.com/science/article/pii/S002199911200157X, doi:https://doi.org/10.1016/j.jcp.2012.03.015.

[6]

Timm Krüger, Halim Kusumaatmaja, Alexandr Kuzmin, Orest Shardt, Goncalo Silva, and Erlend Magnus Viggen. The lattice Boltzmann method. Graduate Texts in Physics. Springer International Publishing, Basel, Switzerland, 1 edition, November 2016.