ANALYSIS OF URANS TURBULENCE MODEL USING A GPU FULLY-IMPLICIT FINITE VOLUME SOLVER

Resumo

© 2026, Begell House Inc. All rights reserved.Parallelized computation with Graphics Processing Unit (GPU) offers significant advantages for Computational Fluid Dynamics (CFD) problems, leveraging their massively parallel architecture to efficiently solve large-scale, computationally intensive tasks such as fluid flow simulations. This study presents a novel GPU Fully-Implicit Finite Volume Solver (GPU-FIFVS) designed to solve the Unsteady Reynolds-Averaged Navier-Stokes (URANS) equations of k-ω Shear Stress Transport (SST) turbulence model. To demonstrate the speed gained by using GPU vs CPU based finite volume computations and considering turbulent flow conditions, this research focused on two well-known cases: NACA (National Advisory Committee for Aeronautics) 0012 airfoil and Backward Facing Step (BFS). Simulations were carried out using two computational frameworks: a GPU-based solver implemented via Nvidia Compute Unified Device Architecture (CUDA) architecture, and a commercial software relying on conventional Central Processing Unit (CPU)-based processing. The study evaluates flow characteristics through wall shear stress and pressure distributions, along with velocity profiles. Simulated results are compared to well-known experimental data to assess predictive consistency across platforms. GPU-based simulations were conducted on a high-end consumer-grade GPU with 32 GB of memory, while CPU-based runs utilized a multi-core CPU operating at up to 4.3 GHz. The findings highlight that the GPU implementation delivers an improved computational time compared to the CPU-based solution, demonstrating the improvement with the application of GPU-FIFVS approach.

Palavras-chave