Cirrus: Adaptive Hybrid Particle-Grid Flow Maps on GPU
Mengdi Wang1 , Fan Feng2, Junlin Li1, Bo Zhu1
1Georgia Institute of Technology
2Dartmouth College
ACM Transactions on Graphics (Proceedings of SIGGRAPH 2025)
Cirrus is a GPU simulator for high-resolution fluid simulations. Its hybrid flow-map scheme combines particles and adaptive grids to preserve fine vortex details while reducing computational requirements. Compared to uniform-grid methods, the adaptive implementation achieves efficiency gains of up to 100× and supports simulations up to 512×512×2048 resolution on a single RTX 4090 GPU. The open-source code provides the simulation framework used in the results below.
Cirrus development began in 2024. The simulator was developed by Mengdi Wang. It uses adaptive octree grids, hybrid particle-grid flow maps and GPU execution.
Abstract
The adaptive hybrid particle-grid flow map method is a novel flow-map approach that uses Lagrangian particles to simultaneously transport impulse and guide grid adaptation, introducing a fully adaptive flow map-based fluid simulation framework. The core idea of the method is to maintain flow-map trajectories separately on grid nodes and particles: the grid-based representation tracks long-range flow maps at a coarse spatial resolution, while the particle-based representation tracks both long and short-range flow maps, enhanced by their gradients, at a fine resolution. This hybrid Eulerian-Lagrangian flow-map representation naturally enables adaptivity for both advection and projection steps. This method is implemented in Cirrus, a GPU-based fluid simulation framework designed for octree-like adaptive grids enhanced with particle trackers. The efficacy of the system is demonstrated through numerical tests and various simulation examples, achieving up to 512x512x2048 effective resolution on an RTX 4090 GPU. The GPU optimization achieves a 1.5 to 2x speedup over the Particle Flow Map method on the same hardware, while the adaptive grid implementation offers efficiency gains of one to two orders of magnitude by reducing computational resource requirements. The source code has been made publicly available at: https://wang-mengdi.github.io/proj/25-cirrus/.
Method
Cirrus constructs an octree-based adaptive grid on the GPU using 8×8×8 tiles as the fundamental building blocks. A particle system serves both as a medium for convective flow mapping and as an oracle to guide grid refinement, enabling fine resolution in regions with critical flow features.
Results
Racing Car
The computational domain is $1\times 1\times 2$ with effective resolution $512\times 512\times 1024$. The inflow and outflow are $\mathbf u=(0,0,1)$ and the length of the car is $0.9$ in the z-axis.
Aircraft
A WP-3D aircraft model in flow $(0,0,1)$ with $4$ rotating propellers at a 15-degree angle of attack. The effective resolution is $512\times 512\times 1024$ in a computational domain $1\times 1\times 2$, and the length of the aircraft is $0.9$. This example demonstrates that the algorithm can efficiently simulate moving small objects and reproduce physical phenomena such as wingtip vortices.
Flamingo Flock
A flying flamingo flock created with a particle system as the effective resolution spans $512\times 512\times 2048$.
Bat
A bat with complex mesh flapping its wings.
Video
