FD-Bench: A Modular and Fair Benchmark for Data-driven Fluid Simulation

TL;DR

FD-Bench tackles fragmented evaluation in data-driven fluid dynamics by providing a fair, modular benchmark that decouples spatial, temporal, and loss components. It systematically re-implements 85 baselines across 10 flow scenarios, and benchmarks neural solvers against traditional pseudo-spectral solvers to illuminate accuracy–efficiency trade-offs. The work reveals that self-attention and temporal bundling offer strong performance, while Eulerian discretizations favor rollout accuracy and Lagrangian methods require multi-scale design to avoid error accumulation. By offering an open-source, extensible framework and a comprehensive generalization analysis, FD-Bench lays groundwork for reproducible, robust evaluation and future advances in data-driven fluid simulation.

Abstract

Data-driven modeling of fluid dynamics has advanced rapidly with neural PDE solvers, yet a fair and strong benchmark remains fragmented due to the absence of unified PDE datasets and standardized evaluation protocols. Although architectural innovations are abundant, fair assessment is further impeded by the lack of clear disentanglement between spatial, temporal and loss modules. In this paper, we introduce FD-Bench, the first fair, modular, comprehensive and reproducible benchmark for data-driven fluid simulation. FD-Bench systematically evaluates 85 baseline models across 10 representative flow scenarios under a unified experimental setup. It provides four key contributions: (1) a modular design enabling fair comparisons across spatial, temporal, and loss function modules; (2) the first systematic framework for direct comparison with traditional numerical solvers; (3) fine-grained generalization analysis across resolutions, initial conditions, and temporal windows; and (4) a user-friendly, extensible codebase to support future research. Through rigorous empirical studies, FD-Bench establishes the most comprehensive leaderboard to date, resolving long-standing issues in reproducibility and comparability, and laying a foundation for robust evaluation of future data-driven fluid models. The code is open-sourced at https://anonymous.4open.science/r/FD-Bench-15BC.

Figures (19)

• Figure 1: Limitations identified in three key areas.
• Figure 2: A schematic illustration of common approaches for each key module in data-driven neural PDE solvers. Note that self-attention is also applicable to temporal representation.
• Figure 2: Statistics of flows. The "shape” column represents, in order: the trajectory sample number, time steps, feature channels, and the flow field resolution.
• Figure 3: We collect and generate 10 representative fluid flow scenarios that span a diverse range of physical conditions. We also present the corresponding visualizations.
• Figure 4: We compare FNO against traditional solver (i.e., pseudo-spectral solver) operating at lower resolutions on N-S equation across five Reynolds numbers, spanning laminar to turbulent regimes. Solver $x$ implies a pseudo-spectral solver operating at $x \times x$ resolution.