HomeWinBuzzer NewsAurora Supercomputer Achieves Exascale but Falls Behind Frontier

Aurora Supercomputer Achieves Exascale but Falls Behind Frontier

Argonne's Aurora supercomputer hits 1 exaflop, ranking second behind Frontier. Both are surpassed by the upcoming El Capitan.

-

Argonne National Laboratory and 's Aurora supercomputer has surpassed the exaflop performance threshold, a key milestone in high-performance computing (HPC). Despite this achievement, Aurora remains second to the Oak Ridge National Laboratory and AMD's Frontier in the latest Top500 supercomputer rankings. The focus now shifts to the upcoming Lawrence Livermore National Laboratory's El Capitan, which might further alter the competitive landscape.

Technical Specifications and Challenges

Aurora, equipped with 21,248 Intel Xeon Max processors and 63,744 GPU Max accelerators, was expected to lead as the most powerful supercomputer in the U.S. upon its arrival at the Department of Energy's Argonne National Laboratory. However, initial challenges in maximizing its capabilities have been evident. Aurora now boasts over an exaFLOP of performance, securing its place as the second machine to achieve exascale status in the publicly known Top500 list. Yet, its considerable power consumption of 38.6 MW to reach this milestone suggests areas for improvement, especially in comparison to Frontier's more efficient 1.2 exaFLOPS at 22.7 MW.

Performance in Benchmarks and Operational Status

Although not fully operational and utilizing only 87% of its capacity in recent benchmarks, Aurora has shown promising results. It claimed the top position in the AI-centric HPL-MxP mixed-precision benchmark with a performance of 10.6 AI Exaflops. Despite numerous hardware and cooling system issues, along with operational errors and network instability, Aurora's potential remains high. Its first submission to the Top500, using only half the system, achieved 585.34 petaflops. With a theoretical peak performance of 2 exaflops announced by Intel, Aurora's current sustained performance stands at 51% of its peak. Further tuning and completion are required for Aurora to potentially exceed Frontier's performance, as per contractual targets.

The Future of Supercomputing

Looking ahead, the El Capitan supercomputer is poised to redefine the supercomputing field with its expected 2.3 exaFLOPS peak performance, powered by AMD's MI300A APUs. This system integrates CPU and GPU capabilities with up to 128GB of HBM3 memory, setting a new standard in computational power. Additionally, global contributions like Switzerland's Alps system, featuring Nvidia's Grace-Hopper Superchips, are making significant strides in the supercomputing rankings.

The ongoing advancements in supercomputing technology underscore the continuous evolution of computational capabilities. As systems like Aurora, Frontier, and the forthcoming El Capitan push the limits of what is achievable, they facilitate advanced research and innovation across various scientific fields. Frontier is currently in place as the world's fastest supercomputer. The rivalry among these technological giants not only showcases the swift progress in the sector but also sets the stage for future breakthroughs in numerous disciplines.

SourceIntel
Luke Jones
Luke Jones
Luke has been writing about all things tech for more than five years. He is following Microsoft closely to bring you the latest news about Windows, Office, Azure, Skype, HoloLens and all the rest of their products.