Speed AND Accuracy: First-Of-Its-Kind Broad-Spectrum CFD Solver Built Natively on GPUs
Since the advent of computational methods to solve physics problems, especially in the realm of fluid dynamics, scientists and engineers have had to balance the need for accurate simulations with faster times to solution — with available computing resources affecting this balance.
Now, we introduce to you a new broad-spectrum native GPU solver created by developers at Ansys. They’ve brought the same numerics that have made Ansys Fluent a trusted and market-leading CFD package across industries and employed ground-breaking parallelization techniques on GPUs to provide results at lightning speed.
This is a new beginning, but stands on the shoulders of decades of CFD experience and expertise.
The current document is intended to report accuracy of the Ansys native-GPU solver for computational fluid dynamics (CFD). We report results for canonical problems in the literature typically used to validate/verify CFD codes. We want to bring you along on this journey that Ansys is taking to build the fastest broad-spectrum CFD code while still preserving accuracy. We trust that our methodical approach will extract the most out of fast-evolving GPU architectures and provide a strong platform to incorporate progressively more complex physics. The speed of the GPU solver on industry-strength problems is reported in a sequence of blogs.
We will begin with details on a laminar flow case followed by several turbulent flow cases. For turbulent flows, we will show the performance of Reynolds-averaged Navier-Stokes (RANS) models, as well as scale-resolving stress-blended eddy simulation (SBES) and wall-adapting local eddy (WALE) turbulence models that have been implemented. Comparisons with experimental data when available are also presented.