RAPIDUS: A Unified, Performance-Portable CFD Framework on Unstructured, Curvilinear, and Strand Grids for Advanced Air Vehicles
The high-performance computing landscape is increasingly defined by heterogeneous architectures that combine Central Processing Units (CPUs) and Graphics Processing Units (GPUs). To fully leverage the computational potential of these platforms, performance portability is essential. It allows simulation software to efficiently exploit diverse hardware without requiring architecture-specific rewrites. However, adapting legacy CFD codes for GPU execution often demands extensive refactoring to expose parallelism and optimize memory access. Maintaining separate code paths for CPUs and GPUs further complicates development, increasing both maintenance overhead and the risk of code divergence. A single-source codebase that executes efficiently across architectures and evolves with hardware generations is therefore critical for long-term sustainability and efficiency. 

This work presents an overview of RAPIDUS, a high-performance, parallel CFD framework designed for performance portability across modern heterogeneous platforms. Supporting multiple mesh paradigms–including unstructured, curvilinear structured, and strand grids–RAPIDUS is well suited for both fixed-wing and rotary-wing simulations. It operates as a standalone solver or within an overset grid framework for complex, multi-body configurations. 

We will highlight recent applications of the unstructured-grid module of RAPIDUS for high-fidelity simulations of complex aerodynamic configurations, including cases from the AIAA High Lift and Drag Prediction Workshops. We also demonstrate enhanced accuracy and performance for the strand- and structured-grid modules. Comparative performance results across CPU and GPU platforms–including multiple generations of GPU hardware–will be discussed, highlighting the significant speedups enabled by GPU acceleration. Finally, we discuss the performance-portable strategies underpinning RAPIDUS’s unified execution model, emphasizing how it maintains efficiency, scalability, and code integrity across different grid topologies and architectures.
IMPACT
Accomplishment: Led the development of the RAPIDUS framework—a single-source, massively parallel, performance-portable CFD code—designed for both standalone simulations and complex multi-body overset configurations.  Result and Impact: On the DoD Nautilus HPC system, GPU-based runs using four NVIDIA A100-80GB GPUs achieved performance equivalent to approximately 28 CPU nodes (3584 cores), demonstrating a 7× reduction in computing footprint. This advancement significantly enhances the scalability and efficiency of high-fidelity CFD for the design and analysis of both rotary- and fixed-wing aircraft, directly supporting DoD objectives for the rapid development and fielding of advanced air platforms.
PRESENTER
Hosseinverdi, Shirzad
shirzad.hosseinverdi.ctr@army.mil
520-275-7453
Analytical Mechanics Associates, U.S. Army DEVCOM, Moffett Field, CA
CO-AUTHOR(S)
Jay Sitaraman
jayanarayanan.sitaraman.civ@army.mil
Andrew Bodling
andrew.l.bodling.ctr@army.mil
Vinod k. Lakshminarayan
vinod.k.lakshminarayan.ctr@mail.mil
Pavithra Premaratne
pavithra.d.premaratne.ctr@army.mil
Dylan Jude
dylan.p.jude.ctr@army.mil
CATEGORY
GPU usage for HPC
SECONDARY CATEGORY
Comp Fluid Dynamics
SYSTEM(S) USED
Nautilus, Coral, Raider