Scalable eVTOL Aerodynamics Simulations on Heterogeneous HPC Platforms with Minimal-Invasive GPU Porting

Urban air-mobility systems such as electric vertical take-off and landing (eVTOL) aircraft need full-vehicle, six-degree-of-freedom (6-DOF) aerodynamic simulations that can handle rotor-rotor and rotor-airframe interactions. Such simulations demand extremely high computational resources necessitating the need for harnessing the power of emerging heterogeneous HPC systems. We present a scalable eVTOL aerodynamic simulation pipeline developed in CUBE, a Fortran computational mechanics solver built on Building-Cube Method (BCM). We also show how minimally intrusive OpenMP offloading can effectively utilize heterogeneous GPU-based systems without significant rewrite of mature production codes like CUBE. The framework combines a robust CAD-to-BCM mesh process with lightweight loop restructure and data-layout changes to effectively scale on both CPUs and GPUs. We evaluate performance of CUBE on both leadership-class and commodity platforms. On Fugaku’s A64FX nodes, CUBE maintains over 90\(\%\) parallel efficiency in weak-scaling tests with more than one million processes. On a traditional x8664 cluster, we see similarly stable scaling. On GPU-equipped systems, we achieve similar throughput when BCM cube granularity matches device occupancy, which makes the trade-offs between MPI offload and throughput clearer. Tests of self-propelled eVTOL configurations show that unsteady aerodynamic simulations can be carried out quickly and accurately enough for design iteration and certification-by-simulation. The findings provide practical direction for adapting large-scale CFD programs to heterogeneous systems while ensuring codebase continuity and performance portability.