Abstract
High-resolution flood inundation simulations impose substantial computational demands on two-dimensional hydrodynamic models, making efficient parallel computing essential for rapid flood forecasting and scenario analysis. To address this challenge, a high-performance two-dimensional flood inundation model was developed based on the Godunov-type finite-volume method and implemented on heterogeneous computing platforms. The model adopts a consistent numerical algorithm and parallel task decomposition strategy across the Hygon C86 7185 CPU, Hygon DCU Z100 accelerator, Intel Core i7-7820HQ mobile CPU, and NVIDIA GPUs (Tesla V100, RTX 3090, and RTX 5090D), using OpenMP, HIP, and CUDA programming frameworks. Three benchmark cases, including the Stoker dam-break, flood inundation at Thamesmead, and Malpasset dam-break, are employed to evaluate numerical accuracy and computational performance. The results demonstrate that the Hygon DCU Z100 achieves speedup ratios ranging from 5.24× to 29.57× over the Hygon C86 7185 CPU running with 32 OpenMP threads across the three cases and achieves higher computational efficiency than the Tesla V100 and RTX 3090 in the Thamesmead and Malpasset cases. In the Stoker case, its computational performance is slightly lower than that of the Tesla V100 and RTX 3090. An additional comparison in the Thamesmead case shows that the Hygon DCU Z100 achieves performance comparable to that of the RTX 5090D. Energy efficiency and cost-effectiveness are also evaluated using rated power and reference hardware prices. Overall, the results demonstrate the potential of the Hygon C86 7185 CPU and Hygon DCU Z100 accelerator for parallel hydrodynamic simulations and highlight the influence of hardware architecture and computational workload on computational performance.