Ansys accelerates CFD simulations by 110x with NVIDIA tech

Ansys has announced a substantial acceleration in computational fluid dynamics (CFD) simulations using NVIDIA GH200 Grace Hopper Superchips, achieving an impressive 110 times speed increase.

The simulations, usually complex and lengthy, have seen their run time reduced dramatically from four weeks to just six hours. This technological feat underscores Ansys' efforts in enhancing capabilities across a range of industries, including automotive, aerospace, gas turbine combustion, chemical mixing processes, and semiconductor manufacturing.

CFD simulations traditionally consume considerable time due to the multiphysics interactions they model, intricate geometries, and the necessity to align simulations with real-world data accurately. These factors often result in simulations spanning days or weeks on conventional CPU cores. Moreover, enhancing the model's accuracy typically extends processing time further and necessitates increased computational resources. However, Ansys' approach harnesses GPU power to deliver quick, accurate insights without the need for extensive computational resources.

Ansys' achievement was highlighted through a collaboration with NVIDIA, utilising advanced computing capabilities at the Texas Advanced Computing Center (TACC) for running a 2.4-billion-cell automotive external aerodynamics simulation. This allowed Ansys to achieve two vital outcomes: maintaining predictive accuracy while speeding up the CFD simulation, and providing room for additional parameters to refine accuracy without slowing down the process.

NVIDIA's contribution involved the use of 320 GH200 Grace Hopper Superchips with scaling through NVIDIA Quantum-2 InfiniBand, which facilitated the 110x speed-up compared to a setup with 2,048 CPU cores. The results equated to a performance match of approximately 225,390 CPU cores. For typical deployments of GPUs, data indicated that scaling to 32 GPUs allowed one NVIDIA GH200 Grace Hopper Superchip to equal the performance of nearly 1,408 CPU cores.

Shane Emswiler, Senior Vice President of Products at Ansys, commented: "Ansys is committed to delivering increased performance and capability to provide our customers with higher levels of simulation fidelity and engineering insight to accelerate innovation. Upgrading to the latest GPU technology can enable our customers to save hours of engineering and product development time, where time to market is essential. Moreover, the energy consumption is much lower across the development cycle, which saves our customers significant costs and resources."

In addition to these advances, Ansys has become the first to implement an Omniverse Blueprint, a reference workflow comprising NVIDIA acceleration libraries, artificial intelligence frameworks, and Omniverse technologies to facilitate real-time, interactive physics visualisation in its applications.

Tim Costa, Senior Director of CAE, EDA and Quantum at NVIDIA, stated: "The power of NVIDIA GH200 Grace Hopper Superchips enables customers to push the limits of simulation model sizes and complexity. The combination of NVIDIA accelerated computing and Ansys software provides engineers with powerful simulation tools to tackle complex engineering problems and reduce time-to-market across industries such as automotive, aerospace, manufacturing and more."