Image 1 of 26
Summit: #1 on the TOP500 supercomputers list
The latest TOP500 supercomputer list is out, and they all have one thing in common: They all run on Linux.
The new TOP500 list is led for the second time by Summit, a US Department of Energy supercomputer capable of performing 200 quadrillion floating-point operations per second.
Another US-based supercomputer, Lawrence Livermore National Laboratory’s Sierra, maintains its number two position, while the Chinese National Computing Center’s Sunway TaihuLight is number three.
Here are the 25 most powerful machines in the TOP500 list of supercomputers.
Editor’s note: This gallery was updated by Brandon Vigliarolo to reflect the current TOP500 supercomputer list.
Summit is housed at Oak Ridge National Laboratory (ORNL), where it aids in research into subjects like material science, cancer, and fusion energy for the US Department of Energy (DoE).
Summit has held the number one spot on the TOP500 list since June 2018, when it became the first US-based supercomputer to head the list since 2012.
The machine has approximately 4,600 nodes, each equipped with two 22-core Power9 CPUs, and six Nvidia Tesla V100 GPUs.
Processor cores: 2,414,592
Max performance (Linpack benchmark): 148.6PFlop/s
Power consumption: 10,096kW
Sierra received a hardware overhaul recently, giving it some new capabilities that pushed it from number three up a place to the second place spot.
Tasked with the crucial role of simulating tests of nuclear weapons in the US stockpile, this new machine is based at the Lawrence Livermore National Laboratory in California.
The supercomputer relies on a mix of IBM Power9 CPUs and Nvidia Volta GPUs and is significantly more capable than the lab’s existing Sequoia supercomputer, with Sierra able to sustain four-to-six times the performance and five-to-seven times the workload of the older machine.
Processor cores: 1,572,480
Max performance (Linpack benchmark): 94,640TFlop/s
3. Sunway TaihuLight
The Sunway TaihuLight once set the standard for supercomputing speed but has since slipped to number three.
Based in the National Supercomputing Center in the city of Wuxi, the Chinese system performs calculations to aid research and engineering work, ranging from climate modelling to advanced manufacturing.
Unlike most other supercomputers, the TaihuLight doesn’t rely on Intel CPUs but instead utilises a custom ShenWei processor, a RISC CPU with 260 cores, and custom interconnects made in Wuxi.
Processor cores: 10,649,600
Max performance (Linpack benchmark): 93PFLOPS – quadrillion floating point operations per second
Power consumption: 15,371kW
Tianhe-2, capable of more than 33 quadrillion calculations per second, holds steady in the number four spot, where it sat in the June list as well.
Otherwise known as the Milky Way 2, the Tianhe-2 supercomputer memory is based in the National Supercomputer Center in Guangzho, China.
The machine is capable of carrying out a massive number of operations in parallel, spreading tasks between its millions of cores. Each of the machine’s nodes has two Intel Xeon E5 Ivy Bridge processors and custom-built Matrix-2000 coprocessors.
Processor cores: 4,981,760
Max performance (Linpack benchmark): 61.4PFLOPS – quadrillion floating point operations per second
Power consumption: 18,482kW
At number five, the Frontera supercomputer at the Texas Advanced Computing Center (TACC) of the University of Austin is the fastest supercomputer at a university in the world.
It uses Dell EMC PowerEdge servers with second-generation Intel Xeon processors.
Processor cores: 448,448
Max performance (Linpack benchmark): 23.5PFLOPS – quadrillion floating point operations per second
6. Piz Daint
The fastest system in Europe and number six machine in the world is Piz Daint, based at the Swiss National Supercomputing Centre.
An upgrade to the Cray XC50 research machine doubled its performance, with Nvidia Tesla P100 GPUs added to its cluster of 2.2GHz Intel Xeon E5-2692 CPUs.
The Piz Daint has come a long way, entering the TOP500 supercomputer list at number 114 in 2012, but steadily climbing thanks to repeated upgrades.
Processor cores: 387,872
Max performance (Linpack benchmark): 21.2PFLOPS – quadrillion floating point operations per second
Power consumption: 2384.24kW
The Trinity supercomputer is another machine engaged in helping the US test the effectiveness of its nuclear arsenal. While it’s lost some processing cores since the June 2018 list, Trinity has still managed to make seventh place.
Based at Los Alamos National Laboratory, Trinity simulates nuclear explosions using hundreds of thousands of processors.
Processor cores: 979,072
Max performance (Linpack benchmark): 20.2PFLOPS – quadrillion floating point operations per second
Based at the National Institute of Advanced Industrial Science and Technology in Japan, ABCI (AI Bridging Cloud Infrastructure) is engaged in research and development of artificial intelligence technologies.
As well as being a processing powerhouse, the system also ranks highly for energy efficiency, claiming eighth place in the Green500 list for its 12.054 gigaflops per watt operating performance.
The Fujitsu-built supercomputer is powered by 20-core Intel Xeon Gold processors, along with Nvidia Tesla V100 GPUs.
Processor cores: 391,680
Max performance (Linpack benchmark): 19,880TFlop/s
The fairly new SuperMUC-NG from the Leibniz Supercomputing Centre enters the list at number nine.
The stated purpose of the SuperMUC-NG is to be available to “all European researchers to expand the frontiers of science and engineering,” which it can do with surprising speed.
Processor cores: 305,856
Max performance (Linpack benchmark): 19,476.6 Tflop/s
Lawrence Livermore National Laboratory’s Lassen supercomputer is similar to Sierra, but is much smaller in size.
It was previously known as uSierra, meaning “unclassified Sierra,” because its purpose is to be a similar machine without the classified nature of its bigger brother.
Processor cores: 288,288
Max performance (Linpack benchmark): 18,200 Tflop/s
11. Pangea III
Developed by IBM, Total’s Pangea III is located at the Centre Scientifique et Technique Jean Fu00e9ger (CSTJF) in Pau, France. Its capacity adds to Pangea I and II, and it has multiplied Total’s computing power by five to 31.7PF.
Processor cores: 291,024
Max performance (Linpack benchmark): 17.9PFLOPS – quadrillion floating point operations per second
Another former frontrunner, Sequoia was not only the most powerful supercomputer in its day but also one of the most efficient. It might be aging, but that doesn’t mean it still isn’t one of the fastest machines in the world.
The IBM Blue Gene/Q system remains a processing powerhouse, and is used by Lawrence Livermore National Laboratory in the US to model massively complex processes, ranging from approximating the universe to the beating of a human heart. It relies on more than one million Power BQC cores.
Processor cores: 1,572,864
Max performance (Linpack benchmark): 17.2PFLOPS – quadrillion floating point operations per second
Power consumption: 7,890kW
The flagship machine at the National Energy Research Scientific Computing Center in Berkeley, CA, Cori entered the TOP500 list in November 2016, making this its third anniversary of being near the top.
Named after the pioneering biochemist Gerty Cori, the system runs on a mix of Intel Xeon Haswell processors and Intel Xeon Phi many-core CPUs.
Processor cores: 622,336
Max performance (Linpack benchmark): 14PFLOPS – quadrillion floating point operations per second
Power consumption: 3,939kW
The research system is a type of Cray CS500 supercomputer, seen above, and relies on a mix of 68-core Intel Xeon Phi co-processors and Intel Omni-Path interconnects.
Processor cores: 570,020
Max performance (Linpack benchmark): 13,929.3TFlop/s
Formerly Japan’s fastest supercomputer, the machine is based at Joint Center for Advanced High Performance Computing in Kashiwa.
Dropping slightly down the rankings this year, the Oakforest-PACS relies on an array of 68-core Intel Xeron Phi 7250 processors.
Processor cores: 556,104
Max performance (Linpack benchmark): 13.6PFLOPS – quadrillion floating point operations per second
Power consumption: 2,719kW
Hosted outside of Milan, this Spanish machine will be used for seismic and petroleum-system modelling.
Noted for its energy efficiency, the supercomputer uses a mix of 24-core Intel Platinum 8160 CPUs and Nvidia Tesla P100 GPUs.
Processor cores: 253,600
Max performance (Linpack benchmark): 12,210TFlop/s
The Bull Sequana X1000 supercomputer has been installed at the French Alternative Energies and Atomic Energy Commission. Its installation is part of the Tera 1000 project to eventually develop an exascale supercomputer.
Processor cores: 561,408
Max performance (Linpack benchmark): 11,965.5 TFlop/s
Power consumption: 3,178kW
The Texas Advanced Computing Center’s latest supercomputer is the Stampede2. It, like the original Stampede built in 2012, is a Dell machine.
Processor cores: 367,024
Max performance (Linpack benchmark): 10.7 TFlop/s
The Marconi supercomputer used by the Italian research consortium Cineca maintained its position in the charts at number 19.
Processor cores: 348,000
Max performance (Linpack benchmark): 10,384.9 TFlop/s
20. DGX SuperPOD
NVIDIA’s DGX SuperPOD comes in at number 20. It has 96 NVIDIA DGX-2H servers containing 1,536 NVIDIA Tesla V100 SXM3 GPUs.
Processor cores: 127,488
Max performance (Linpack benchmark): 9,444 TFlop/s
21. Taiwania 2
Taiwan’s National Center for High Performance Computing was quick to outdo itself, having only entered into the TOP500 in 2017 with the Taiwania supercomputer. Now it’s showing off the Taiwania 2 at number 21, while the original Taiwania barely managed to crack the top 500.
Processor cores: 170,352
Max performance (Linpack benchmark): 9,000 TFlop/s
Power: 797.54 kW
Capable of some 8.6 quadrillion calculations per second, Mira plays a pivotal role in research at the Argonne National Laboratory in Illinois.
The IBM BlueGene supercomputer helps US researchers model everything from the performance of jet engines to the inner workings of the human body.
Driving this machine are a host of 16-core, 1.6GHz Power BQC processors.
Processor cores: 786,432
Max performance (Linpack benchmark): 8.6PFLOPS – quadrillion floating point operations per second
Power consumption: 3,945kW
23. TSUBAME 3.0
Used by the Tokyo Institute of Technology for computational modelling and simulations, work on TSUBAME 3.0 was started in August 2017.
Processor cores: 135,828
Max performance (Linpack benchmark): 8,125 TFlop/s
Power consumption: 792.08kW
Supercomputer AMOS (Advanced Multiprocessing Optimized System) joins the list in 2019 for the first time at number 24. It is located at the Rensselaer Center for Computational Innovations (CCI) and became active in October 2013.
Processor cores: 130,000
Max performance (Linpack benchmark): 8,045 TFlop/s
Memory: 128,000 GB
Power consumption: 510.10kW
PupMaya – Apollo 2000 is manufactured by HPE (Hewlett Packard Enterprise)
Processor cores: 169,920
Max performance (Linpack benchmark): 7,483.73 TFlop/s
Memory: 815,616 GB
Editor’s note: This is a stock data center image because HPE has not released a photo of the PupMaya supercomputer.