The Oak Ridge National Laboratory’s Frontier supercomputer has won the top spot on the Top500 list of the world’s fastest supercomputers.

The latest edition of the Top500 list, which was released today, ranks Frontier as the world’s fastest supercomputer with performance of 1.102 exaflops per second. An exaflop equals 1 quintillion, or a million trillion, computing operations per second. Frontier’s performance was measured using a popular benchmark test known as HPL.

The top spot on the Top500 list was previously held by the Fugaku supercomputer in Kobe, Japan. Fugaku achieved performance of 442 petaflops in a test conducted using the HPL benchmark. However, the system’s theoretical top speed is believed to be in excess of one exaflop. 

The Frontier supercomputer was constructed as part of a technology initiative that the U.S. Department of Energy announced in May 2019. Officials entrusted the task of developing Frontier to Advanced Micro Devices Inc. and Cray Inc., a major supercomputer builder. 

A few days after the initiative to build Frontier was announced, Hewlett Packard Enterprise Co. inked a deal to acquire Cray for $1.3 billion. The acquisition established HPE as one of the supercomputing market’s largest players. The company has since incorporated Cray’s products into its supercomputing portfolio. 

Frontier comprises 74 HPE Cray EX cabinets that each accommodate 128 blades, a type of compact server. A blade usually has fewer components than a standard data center server, which makes it easier to cool. Certain tasks related to hardware maintenance are simpler as well.

Each blade in Frontier includes one central processing unit from AMD’s Epyc processor line and four Instinct MI250x accelerators. The MI250x accelerator is a computing module sold by AMD that features two graphics processing units made using a six-nanometer manufacturing process. The GPUs are each supported by a significant amount of HBM2E memory, a type of high-speed memory used to store the data being processed by a chip for rapid access.

Overall, Frontier includes some 9,408 CPUs and 37,632 GPUs. The heat produced by the chips is dissipated using a specialized cooling system that moves 5,900 gallons of water through Frontier every minute. 

Water conducts heat better than air, which generally makes water cooling systems more efficient than fans. HPE says that Frontier is not only the world’s fastest supercomputer but also the most power-efficient. Frontier provides 52.23 gigaflops of performance per watt, with 1 gigaflop equal to a billion computing operations per second.

The chips, cabinets and other components that comprise Frontier are linked together by 90 miles of cables. The system’s CPUs and GPUs are supported by a high-speed storage system dubbed Orion that provides 700 petabytes of capacity. Orion can write more than 35 terabytes of data per second to storage, as well as carry out north of 15 billion random-read input/output operations per second.

“Today’s debut of the Frontier exascale supercomputer delivers a breakthrough of speed and performance, and will give us the opportunity to answer questions we never knew to ask,” said Justin Hotard, executive vice president and general manager of HPE’s HPC and AI unit. “Frontier is a first-of-its-kind system that was envisioned by technologists, scientists and researchers to unleash a new level of capability to deliver open science, AI and other breakthroughs, that will benefit humanity.”

According to HPE, Frontier will enable researchers to develop neural networks eight times larger than those that can be created using earlier hardware. Additionally, the company said, AI training will be 4.5 times faster. 

“Scientists and engineers from around the world will put these extraordinary computing speeds to work to solve some of the most challenging questions of our era, and many will begin their exploration on Day One,” said Jeff Nichols, associate lab director for computing and computational sciences at Oak Ridge National Laboratory.

HPE and AMD are currently developing another exascale supercomputer called El Capitan for the Department of Energy. The system is expected to be even faster than Frontier. When it launches in 2023, El Capitan will provide more than 2 exaflops of performance. 

Photo: HPE