This article is more than 1 year old
All-AMD US Frontier supercomputer ousts Japan's Fugaku as No. 1 in Top500
Exascale beast's test system also claims top spot in the Green500
The land of the rising sun has fallen to the United States’ supercomputing might. Oak Ridge National Laboratory’s (ORNL) newly minted Frontier supercomputer has ousted Japan’s Arm-based Fugaku for the top spot on the Top500 rankings of the world's most-powerful publicly known systems.
Frontier’s lead over Japan’s A64X-based Fujitsu machine is by no means a narrow one either. The cluster achieved peak performance of 1.1 exaflops according to the Linpack benchmark, which has been the standard by which supercomputers have been ranked since the mid-1990s.
Frontier marks the first publicly benchmarked exascale computer by quite a margin. The ORNL system is well ahead of Fugaku’s 442 petaflops of performance, which was a strong enough showing to keep Fugaku in the top spot for two years.
Reaching exascale status is one thing, but many expected the efficiency-geared Fugaku system to hang onto its green rankings, even if it slipped on the performance front. But Frontier isn’t just the most powerful known supercomputer, its technology is now also the most efficient. At 52.23 gigaflops per watt, the system's test-and-development machine Crusher also outperformed Japan’s Preferred Networks MN-3 system to take the lead spot on the Green500.
“The fact that the world’s fastest machine is also the most energy efficient is just simply amazing,” Thomas Zacharia, lab director at ORNL, said during a press conference Sunday.
The results mark a turning point for US supercomputing, which has waned in recent years. American systems now capture the first, fourth, fifth, seventh and eighth slots in the top ten of the Top500 list just unveiled at the International Supercomputing Conference.
Frontier performance and efficiency balances were struck via a Hewlett Packard Enterprise’s (HPE) Cray EX platform. The direct liquid-cooled systems were equipped with AMD’s third-gen Epyc processors, Instinct MI250X accelerators, and 200 Gbit/second Slingshot-11 interconnect fabric.
- US fears China may have ten exascale systems by 2025
- US to rev fastest supercomputer with powerful test system
- Another AI supercomputer from HPE: Champollion lands in France
- HPE, Cerebras build AI supercomputer for scientific research
Europe rises
The US wasn’t the only region to make notable gains. Europe also faired well with Finland’s newly deployed LUMI supercomputer narrowly outpacing America’s Summit to claim the No. 3 spot with 151.9 petaflops of FP64 performance.
Meanwhile, France’s Adastra at GENCI-CINES narrowly claimed the number 10 spot at 46.1 petaflops of performance. While nowhere near as powerful as LUMI, Adastra still holds the distinction of being the second most powerful supercomputer in Europe.
Europe’s relative strength in the latest Top500 results underscores growing momentum around high-performance computing in the region. The EuroHPC Joint Undertaking has already taken steps to acquire and deploy Europe’s first exascale-class system by 2023.
Additionally, Europe will be home to several upcoming supercomputing projects, including Switzerland’s long-awaited Alps supercomputer, the Champollion supercomputer in France, and Cineca's Leonardo system in Italy to name just a few.
Systems like LUMI and Adastra also illustrate another trend. Like Frontier, they’re based on HPE’s all-AMD Cray EX platform using the chipmaker’s third-gen Epyc CPUs and Instinct GPUs.
HPE built four of the top 10 in this spring’s ranking, Justin Hotard, EVP and GM of HPC and AI at HPE, said, touting the strength of the company’s AMD-based Cray systems.
Despite only launching its MI250X GPUs last fall, AMD’s chips dominated the Top500, with its CPUs at the heart of half of the 10 most powerful systems.
Don't forget about China
Intel powered just one of the top 10 systems. China’s Tianhe-2A used the chipmaker’s ancient Xeon E5-2692v2 Ivy Bridge CPUs running alongside its TH Express-2 and Matrix-2000 accelerators to claim the number 9 spot with 61.4 petaflops of peak performance.
While Intel-based systems are sparse in the top 10, China faired better, with its Sunway TaihuLight supercomputer nabbing the number six slot.
Like Fugaku, you won’t find any commodity hardware under the lid. TaihuLight is based on an entirely custom architecture, which enabled it to achieve 93 petaflops of peak performance.
Despite America’s lead in the Top500, it should be noted that China has already exceeded the exaflop barrier. Two systems to keep an eye on are the Sunway Oceanlite — the successor to TaihuLight — and Tianhe-3, both of which have reportedly achieved 1.3 exaflops of performance in the Linpack benchmark.
Should those systems make their way into the rankings, this fall’s Top500 could look very different.
But while exascale systems will continue to grab headlines for years to come, it’s just another goal post on the way to zettascale, and even that won’t be the final frontier.
Checkout the full list of systems of this spring's Top500 here. ®