sortiwa
In context: Computational fluid dynamics (CFD) exploits the power of computer technology to study and predict how fluids behave under simulated conditions. Supercomputers can do wonders with CFD applications, and AMD Instinct GPUs have been properly leveraged to reduce the time needed to run these simulations, with room for even more improvement.
Ansys, a simulation software company, used its Fluent software to simulate a complex turbine system on one of the world’s most powerful supercomputers. By running the task on AMD Instinct GPUs instead of traditional CPU cores, the company cut simulation time by 96 percent. Ansys says this is just the beginning of speeding up computational fluid dynamics workloads.
Ansys Fluent provides advanced physics modeling to analyze various fluid phenomena. The company recently partnered with energy firm Baker Hughes, which is designing next-generation gas turbines and other turbomachinery to boost energy conversion efficiency. Together, they worked with Oak Ridge National Laboratory and its Frontier supercomputer – the first HPC system to reach exascale-class performance.
Frontier runs on AMD Epyc CPUs and Instinct GPUs, which have gained traction even as Nvidia continues to semi-monopolize the data center GPU market. Ansys and Baker Hughes tapped into the supercomputer’s enormous processing power to scale its Fluent simulation across 1,024 GPUs. The result: a 2.2-billion-cell axial turbine stator simulation that helped pinpoint key flow and turbulence structures early in the design phase.
Ansys credited AMD’s Instinct GPUs for cutting the simulation time from 38.5 hours to just 1.5. The earlier method relied on more than 3,700 CPU cores, but the new approach ran on 1,024 Instinct MI250X GPUs. According to the company, this marks a new scaling record for CFD applications. This breakthrough could dramatically speed up design iterations and deliver more accurate performance forecasts for industrial systems.
Ansys also sees potential benefits for small and mid-sized businesses. Even without access to massive HPC systems, companies using smaller GPU-based setups could still see significant gains in simulation efficiency and scalability. Ansys claims enterprises can accelerate complex CFD workloads without needing large-scale infrastructure by running Fluent on GPU-based systems.
For now, Baker Hughes is using the Ansys Fluent GPU solver and Frontier supercomputer to optimize the development of its next-generation turbine engines. Ansys believes combining modern GPUs and advanced CFD software could benefit other complex design and industrial applications.
Source link