It is important to understand how the fastest supercomputers are measured before ranking them. Historically, computers were measured in IPS (instructions per second), but modern supercomputers are measured in FLOPS (floating point operations per second).
Before we begin this list, please remind yourself what a supercomputer is if you would like more context. From the most powerful to the least powerful, we will list supercomputers in descending order.
The frontier (United States)
In collaboration with Hewlett Packard Enterprise’s subsidiary Cray, Frontier was built in 2022. It is the world’s first exascale supercomputer, capable of computing at least one quintillion (10^18) calculations per second.
The Frontier has 8,730,112 cores and scores 1.1 EFLOPS (or exaflops) on Linpack benchmark tests. Using AMD’s 3rd generation 64-core 2GHz 7A53s CPUs and MI250X GPUs, it’s based on HPE Cray EX235a architecture.
A power efficiency rating of 52.23 gigaflops/watt makes Frontier the most efficient supercomputer in the world. With 74 cabinets weighing around 8,000 pounds each, the system costs $600 million in total.
As a successor to Fujitsu’s 2011 K computer, Fukuku was created by the Japanese IT giant in 2020. In particular, it was designed to slow down climate change, one of the world’s biggest problems.
In May 2022, Frontier dethroned Fugaku as the fastest supercomputer in the world. With 7,630,848 cores, it can do 442 quadrillion calculations per second and scores an impressive 442 PFLOPS (or petaflops).
It has a power efficiency rating of just 14.78 gigaflops/watt, making it quite inefficient by today’s standards. It is powered by Fujitsu’s A64FX 48-core 2.2GHz processor. Each cabinet weighs 1.6 tons, totaling 700 tons, and the system cost more than $1 billion.
In 2022, HPE built LUMI (Large Unified Modern Infrastructure) in Finland, making it the fastest supercomputer in Europe. As a result, LUMI clocks at a speed of 151.9 PFLOPS and has a total of 1,110,144 cores.
In terms of energy efficiency, LUMI is the second most efficient supercomputer in the world, with a power efficiency rating of 51.63 gigaflops/watt.
Summit (United States)
As part of IBM’s scientific research program, Summit was created in 2018. The facility is located at Oak Ridge National Laboratory in the US, which is also the loca148.6 PFLOPS are achieved by Summit’s 2,414,592 cores. ores 148.6 PFLOPS.
As with Fugaku, Summit uses IBM’s POWER9 22-core 3.07GHz CPUs and Nvidia Tesla V100 GPUs and has a power efficiency rating of 14.72 gigaflops/watt.
Sierra (United States)
Designed in 2018, Sierra uses the same IBM POWER9 22-core CPU architecture and Nvidia Tesla V100 GPUs as Summit. In contrast, Summit is designed for scientific research, while Sierra is designed for testing and maintaining nuclear weapons.
Using nuclear weapon simulations, the US National Nuclear Security Administration (NNSA) can test its nuclear weapons’ effectiveness without physical testing, which helps save money. With 1,572,480 cores, Sierra scores 94.64 PFLOPS and has a power efficiency of 12.72 gigaflop/watt.
Sunway TaihuLight (China)
In sixth place, we find the first Chinese supercomputer, Sunway TaihuLight. Weather forecasting, pharmaceutical research, life sciences research, and more were among the uses for which it was built in 2016.
93.01 PFLOPS are achieved by Sunway TaihuLight, which has 10,649,600 CPU cores. As the most inefficient supercomputer on this list, it runs on the Sunway SW26010 260-core 1.45GHz processor and has a power efficiency rating of just 6.05 gigaflops/watt.
Perlmutter (United States)
Built by HPE in 2021, Perlmutter is the world’s seventh most powerful supercomputer, named after Nobel Prize winner Saul Perlmutter. Similar to Sunway TaihuLight, it was designed for nuclear fusion simulations, climate projections, materials and biological research, and computational cosmology.
Currently, it is located at the National Energy Research Scientific Computing Center (NERSC) and is used by the U.S. Department of Energy. With a total of 761,856 cores and a speed of 70.87 PFLOPS, Perlmutter’s system has a power efficiency rating of 27.37 gigaflops/watt and a total of AMD 7763 64-core 2.45GHz CPUs and Nvidia A100 GPUs.
Selene (United States)
Based on the Greek goddess of the moon, Selene was built by NVIDIA in 2020. A major purpose of this device was realized in 2020 when Argonne National Laboratory, a national research lab based in Illinois, used it to study Coronavirus and possible cures for it. As part of the training process for BERT, Google also uses Selene.
During benchmarks, Selene scored 63.46 PFLOPS and has 555,520 cores. Powered by AMD 7742 64-core 2.25GHz processors, it is based on NVIDIA DGX SuperPOD reference architecture. It has a power efficiency rating of 24 gigaflops per watt.
It’s not science fiction anymore, we already have supercomputers
There is no doubt that supercomputers are incredibly powerful. A supercomputer can accomplish a task that would take humans several years and regular computers weeks. Frontier will soon be outcompeted by another supercomputer that is faster and more efficient.
Laptops, PCs, and gaming consoles can already be considered supercomputers, but they are no match for the best we’ve developed. Luckily, as is the nature of tech, it’ll continue to get smaller and eventually end up in our gadgets, helping us extend our knowledge and creating massive new industries, ready for us to mold.