A groundbreaking achievement in supercomputing performance has been demonstrated using AMD's Instinct MI250X GPUs, setting a new world record for computational fluid dynamics (CFD) simulation speed.
The record-setting run took place at Oak Ridge National Laboratory's Frontier supercomputer, where engineers completed a massive 2.2-billion-cell turbine simulation in just 1.5 hours - a dramatic improvement from the previous 38.5-hour runtime. This represents a more than 25-fold speed increase in processing time.
The simulation, conducted using Ansys Fluent software for Baker Hughes' next-generation gas turbine development, leveraged 1,024 AMD Instinct MI250X accelerators working alongside AMD EPYC processors. This unprecedented processing speed allows engineers to rapidly iterate designs and analyze performance impacts.
"This collaboration showcases how cutting-edge supercomputing can tackle the toughest engineering challenges," noted Brad McCredie, AMD Senior Vice President for Data Center Engineering. The breakthrough enables advances in efficiency and sustainability for industrial applications.
The achievement highlights AMD's growing presence in high-performance computing, with its processors powering some of the world's fastest supercomputers including Frontier and El Capitan. Frontier itself contains over 37,000 AMD Instinct MI250X accelerators paired with more than 9,000 EPYC processors.
While the record-setting simulation used only a fraction of Frontier's total computing capacity, it demonstrates the immense potential of AMD's hardware architecture for scientific and engineering applications. The company continues to compete with market leader Nvidia, particularly in specialized computing tasks requiring massive parallel processing power.
The breakthrough simulation results showcase AMD's capabilities in delivering exceptional performance for complex computational workloads, reinforcing its position as a leading innovator in high-performance computing technology.