Supercomputer performance is typically measured in floating-point operations per second, and the current generation of top machines has crossed into the exascale era, capable of more than a quintillion calculations every second. These systems are used for tasks ranging from climate modeling and nuclear weapons simulation to training the largest artificial intelligence models, and national governments now treat leadership in this hardware as a matter of strategic competition. The nine machines below rank among the fastest supercomputers built, spanning current record holders and the recent former champions they replaced.
1. Frontier
Frontier, housed at Oak Ridge National Laboratory in Tennessee, became the first supercomputer confirmed to exceed one exaflop of sustained performance, a milestone reached in 2022 that formally opened the exascale era of computing. Built on HPE Cray hardware using AMD processors and graphics accelerators, it supports major scientific research across fields including materials science, climate modeling, and fusion energy. The Frontier supercomputer’s debut ended nearly a decade of anticipation for machines capable of exascale calculation. Researchers across dozens of institutions now compete for allocated time on its processors each year.
2. Aurora
Aurora, deployed at Argonne National Laboratory outside Chicago, is one of the United States Department of Energy’s flagship exascale systems, built using Intel and Hewlett Packard Enterprise hardware after a development process that ran years behind its original schedule. It is used heavily for large-scale artificial intelligence research alongside traditional scientific simulation work. The Aurora supercomputer represents one of the country’s major investments in AI-capable exascale infrastructure, joining Frontier as one of the Department of Energy’s two exascale flagships, though its delayed rollout drew scrutiny over cost overruns during construction.
3. El Capitan
El Capitan, at Lawrence Livermore National Laboratory in California, became the world’s fastest supercomputer according to the TOP500 rankings in late 2024, built on AMD-based HPE Cray hardware and dedicated primarily to United States nuclear weapons stockpile stewardship simulations. Its performance substantially exceeded Frontier’s, reasserting Livermore’s long history hosting record-setting machines. The El Capitan supercomputer’s classified and unclassified computing partitions support both national security work and open scientific research, a dual role Livermore’s earlier flagship systems, including Sierra before it, have also filled.
4. Fugaku
Fugaku, developed by Japan’s RIKEN research institute together with Fujitsu, held the title of the world’s fastest supercomputer through much of 2020 to 2022, notable for using ARM-based processors rather than the x86 chips more common in earlier record holders. Its architecture demonstrated that ARM-based designs could compete at the very top of high-performance computing, a result that influenced chip design decisions well beyond Japan. The Fugaku supercomputer also played a widely publicized role modeling airborne virus transmission during the COVID-19 pandemic, work that shaped public health guidance in Japan.
5. MareNostrum
MareNostrum, operated by the Barcelona Supercomputing Center, is one of Europe’s most powerful research computers and famously housed, in earlier generations, inside a deconsecrated chapel on a university campus, a striking juxtaposition that made it one of the most photographed supercomputers in the world. Its current generation continues to support European scientific research across climate, medical, and engineering fields. The MareNostrum supercomputer remains a flagship symbol of Spain’s high-performance computing capability, drawing visiting researchers and students from across Europe.
6. Leonardo
Leonardo, operated by Italy’s CINECA consortium, is one of the European Union’s pre-exascale supercomputers funded through the EuroHPC Joint Undertaking, an initiative aimed at keeping European research computing competitive with the United States and China. It supports large-scale scientific projects across European universities and research institutions, from climate modeling to pharmaceutical simulation work that would otherwise take far longer on national-scale machines. The Leonardo supercomputer’s construction reflected a broader European push to reduce reliance on computing infrastructure hosted outside the continent, a strategic priority that has grown since its launch.
7. Summit
Summit, also at Oak Ridge National Laboratory, was the world’s fastest supercomputer from 2018 until it was surpassed by Fugaku in 2020, built using IBM Power9 processors paired with Nvidia graphics accelerators. It was among the first supercomputers to demonstrate that GPU acceleration, originally developed for graphics and gaming, could dramatically boost scientific computing performance, a shift that reshaped how later systems were designed. The Summit supercomputer’s success helped establish the GPU-heavy architecture that later exascale machines, including Frontier, would build on. Its research output spans thousands of published scientific papers across thousands of collaborating institutions.
8. Sierra
Sierra, Summit’s sister system at Lawrence Livermore National Laboratory, uses similar IBM Power9 and Nvidia hardware but is dedicated primarily to nuclear weapons stockpile simulation rather than open scientific research. The two machines were developed together under the same Department of Energy program, sharing much of their underlying architecture. The Sierra supercomputer has since been succeeded at Livermore by the newer El Capitan system, though it continues to run lower-priority workloads.
9. Sunway TaihuLight
Sunway TaihuLight, based at the National Supercomputing Center in Wuxi, China, was the world’s fastest supercomputer from 2016 to 2018 and notable for using entirely Chinese-designed SW26010 processors rather than the American chips most rival machines relied on. Its debut signaled China’s growing self-sufficiency in high-performance computing hardware, reducing reliance on imported processor designs for its most demanding workloads. The Sunway TaihuLight supercomputer remains one of the most powerful machines built entirely on domestically designed Chinese processors, and it continues to support Chinese scientific and engineering research.
This article was produced with the assistance of AI and reviewed by Morning Overview editors.
More from Morning Overview
