Frontier establishes a new fluid dynamics speed record, reducing simulation time from 38.5 to 1.5 hours on AMD GPUs

Frontier slashes CFD simulation time using AMD GPUs from 38.5 to 1.5 hours.

: Frontier supercomputer dramatically reduces computational fluid dynamics simulation time from 38.5 to 1.5 hours using AMD Instinct GPUs, compared to traditional CPU methods. Ansys, in collaboration with Baker Hughes, demonstrates this speed-up on a 2.2-billion-cell turbine simulation, indicating potential time savings and design efficiency. The deployment of 1,024 AMD Instinct MI250X GPUs showcases a record-breaking CFD application scaling, opening possibilities for diverse industrial applications. These advancements promise significant benefits even for smaller businesses, enhancing simulation capabilities without large infrastructures.

In an impressive leap forward for computational fluid dynamics (CFD), the Frontier supercomputer has broken new ground by slashing simulation times from 38.5 hours to just 1.5 hours. This achievement pivots on utilizing AMD Instinct GPUs instead of conventional CPU cores, demonstrating the high-efficiency benefits of GPUs in handling complex CFD simulations. The collaborative effort involves Ansys, a simulation software company, and Baker Hughes, an energy firm focused on next-gen turbine designs, showcasing how modern computational power can significantly enhance energy efficiency and design processes.

The specific simulation in question involves a complex analysis of a turbine system, executed on the world's foremost exascale supercomputer, Frontier. Located at Oak Ridge National Laboratory, Frontier is powered by AMD Epyc CPUs and Instinct GPUs, marking a noteworthy uptick in supercomputing capabilities by sidestepping Nvidia's dominant presence in data centers. Ansys used its Fluent software to tackle a 2.2-billion-cell axial turbine stator simulation, using 1,024 AMD Instinct MI250X GPUs, epitomizing a substantial scaling record for CFD applications.

By shifting away from CPU-dominant approaches, which previously required over 3,700 cores, the switch to GPU technology not only cut simulation times dramatically but also opened doors for faster project iterations and more precise performance forecasting across various industrial sectors. This breakthrough highlights not only the prowess of Frontier's architecture but an exciting new chapter for CFD-related applications in achieving unprecedented efficiency, even for small and medium-sized businesses with limited computational resources.

Ansys posits that utilizing smaller-scale GPU setups can significantly impact enterprises without access to massive high-performance computing systems. Simulations that once demanded extensive infrastructure can now be carried out with relatively fewer resources, resulting in faster and more scalable CFD workloads. Such enhancements enable businesses of varying scales to compete more effectively, innovate rapidly, and deliver products or solutions with superior precision and reliability.

Looking ahead, Baker Hughes continues to leverage Frontier and Ansys Fluent for refining its turbine engines. Ansys remains optimistic that the blend of modern GPUs and sophisticated CFD software can revolutionize design and innovation across diverse applications—from industrial machinery to energy systems. The achievement underscores a powerful synergy between cutting-edge technological hardware and advanced software methodologies, promising a revitalized era in simulation efficiency.

Sources: TechSpot, Oak Ridge National Laboratory

TechnologySoftware