Released in 1964, the Control Data Corporation (CDC) 6600 was considered to be the first successful supercomputer, capable of executing 3 million floating-point operations per second (FLOPS). Fast forward to 2020, and the PlayStation 5 has hardware capable of up to 10.28 teraFLOPS. For reference, 1 teraFLOP is 1 trillion FLOPS.
The term “supercomputer” refers to a class of extremely powerful computers, but what is considered powerful changes as years pass and technology evolves – and the specific workloads towards which they are deployed change as needs of enterprises and governments develop. The Frontier Enterprise staff has put together a list of today’s fastest supercomputers and their use cases. Most of it is based on the 57th edition of the TOP500, a project that tracks the world’s most powerful supercomputers, with one new addition.
Of the supercomputers featured here, half of them are used in healthcare, while nearly a third delve into drug-related research. Two of the supercomputers are used primarily in an enterprise setting, with one of them specialising in oil- and gas-related research. Three have propelled weather and climate research, and one is dedicated to nuclear testing (although it has been made available for studying COVID-19).
One thing is certain: these machines have made it possible for us to solve our most challenging and computationally intensive problems.
Fugaku
Fugaku is a supercomputer developed jointly by the Riken Center for Computational Science and IT giant Fujitsu. Its specifications include a 512-bit ARM architecture with 7,630,848 cores and 5,087,232 GB of memory. Its benchmark scores show why Fugaku is on the top spot:
- On the TOP500, it achieved a LINPACK score of 442.01 petaFLOPS.
- On the High-Performance Conjugate Gradient benchmark, it scored 16.00 petaFLOPS.
- On the High-Performance LINPACK for Acceleration Introspection, it scored 2.004 exaFLOPS.
Fugaku’s current projects are focused mainly on COVID-19-related fields, including exploring drug candidates, predicting the conformational dynamics of proteins on the SARS-Cov-2 virus, and carrying out fragment molecular orbital (FMO) calculations for COVID-19 proteins.
According to Satoshi Matsuoka, director of Riken, Fugaku has also contributed to the establishment of COVID-19 safety guidelines for the government and private sector, and has helped lead a digital transformation in the area of infectious diseases.”
Matsuoka also highlighted the supercomputer’s speed in a real-world use case. Referencing research on COVID-19’s spike protein, Matsuoka said that on the K computer, Fugaku’s predecessor, FMO calculations to investigate it would have taken “days, weeks, multiple weeks.” It took Fugaku three hours.
Naoki Shinjo, Corporate Executive Officer at Fujitsu, revealed that Fugaku has started shared use with other researchers starting March 2021. “In the future, we anticipate that Fugaku’s use will continue to play an important role in contributing to the realisation of the Japanese government’s plan for Society 5.0 (a smart society initiative), in areas like drug discovery, as well as ensuring a secure and safe society.”
Cambridge-1
Cambri dge-1 is said to be the United Kingdom’s fastest supercomputer. It was built by technology company Nvidia and is capable of cranking out 400-petaFLOPS. All this power is used mainly for medical research. Some of its projects include the following:
- Nvidia and pharmaceutical company AstraZeneca have teamed up for drug discovery research, creating a transformer-based generative artificial intelligence (AI) model for chemical structures. They are also planning to train AI algorithms on Cambridge-1 for digital pathology.
- King’s College London and Guy’s and St Thomas’ NHS Foundation Trust are using Cambridge-1 to teach AI models to generate synthetic brain images and learn more about diseases such as dementia, stroke, brain cancer, and multiple sclerosis for earlier diagnosis and treatment.
Summit
Summit, built by tech giant IBM at the Oak Ridge National Laboratory (ORNL) in Tennessee, is the fastest supercomputer in the United States, capable of performing 148.8 petaFLOPS. It presently has 4,356 nodes, with each one housing two 22-core Power9 central processing units (CPUs) and six Nvidia Tesla V100 graphics processing units (GPUs).
Weighing over 340 tons and occupying 5,600 square feet of floor space, Summit can perform 200 quadrillion FLOPS, store 250 petabytes of data. According to ORNL computational scientist Wayne Joubert, Summit is so fast that in just one hour using it, researchers can solve a problem that would take 30 years on a desktop computer.
According to Tom Rosamilia, Senior Vice President of IBM Cloud and Cognitive Software, Summit is built with AI workloads in mind, and seeks to address challenges like climate change, understanding genetic factors that influence opioid addiction, and predicting extreme weather before they happen.
Sierra
Built for the Lawrence Livermore National Laboratory in California, Sierra is used primarily for predictive applications in stockpile stewardship, which is a US program for testing and maintaining the reliability of its nuclear weapons.
The Summit and Sierra architectures are quite similar, with each of its 4,474 nodes equipped with two Power9 CPUs and four Nvidia Tesla V100 GPUs. This enables the Sierra computer to perform at a peak of 125 petaFLOPS.
Sunway TaihuLight
Developed by China’s National Research Centre of Parallel Computer Engineering & Technology, the Sunway TaihuLight is equipped with 40,960 nodes of SW26010 manycore 64-bit RISC processors. Each of its processors contain 256 processing cores, and four auxiliary cores for system management, giving it a total of 10,649,600 CPU cores. This enables the supercomputer to reach a processing speed of 93 petaFLOPS.
TaihuLight is currently running at the National Supercomputing Centre in Wuxi. It is used for research and engineering work, specifically in fields such as advanced manufacturing, climate, data analytics, earth systems modeling, and life science research.
Selene
Selene is a supercomputer built by Nvidia that is equipped with competitor Advanced Micro Devices’ (AMD) EPYC processors, A100 GPUs, and Mellanox HDR InfiniBand networking technology. Selene is capable of performing 63.60 petaFLOPS.
Nvidia uses Selene in various research, development, and production-related projects. This includes large-scale language modeling, autonomous vehicles, graphics rendering, and tools for quantum chemistry and genomics.
Tianhe-2A
Also called Milky Way-2A, the Tianhe-2A is a 33.86-petaFLOPS supercomputer developed by China’s National University of Defense Technology (NUDT) and deployed at the National Supercomputer Centre in Guangzhou. It has 16,000 computer nodes, each comprising two Intel Ivy Bridge Xeon processors and three Xeon Phi coprocessor chips.
As per NUDT, Tianhe-2A has a number of applications including computational fluid dynamics simulation, business opinion analysis, and government security.
JUWELS Booster Module
Installed at the Forschungszentrum Jülich research centre in Germany, the Atos-built supercomputer JUWELS Booster Module is capable of achieving 85 petaFLOPS. This is equivalent to the computing power of over 300,000 modern PCs. The JUWELS Booster Module is powered by AMD EPYC processors and Nvidia A100 GPUs, making it the most powerful supercomputer in Europe.
JUWELS is currently involved several projects such as:
- Studying drug binding over biologically relevant timescales.
- High-resolution weather forecasting based on deep learning.
- Plasma simulations for next-generation particle accelerators.
HPC5
Created by oil and gas multinational company Eni S.p.A, the HPC5 has a peak performance of 51.7 petaFLOPS.
This output is made possible through 1820 Dell EMC PowerEdge C4140 nodes, each equipped with two Intel Gold 6252 24-core CPUs and four Nvidia V100 GPUs. The nodes are connected via an Infiniband Mellanox HDR network technology with a speed of 200 Gigabits per second, using over 10 kilometres of fibre-optic cable. HPC5 comes with a 15-petabyte storage system with aggregate read/write speeds of 200 Gigabytes per second.
Epi said it uses the supercomputer for various technological, environmental and research purposes, such as processing subsoil data and developing subsoil models. In fact, this is how the company found the Zohr gas field in the Mediterranean Sea. The company has also made the HPC5 available for coronavirus research.
Frontera
Deployed at The University of Texas at Austin’s Texas Advanced Computing Center in 2020, Frontera currently achieves 23.5 petaFLOPS using 448,448 of its Intel Platinum Xeon cores. Like other supercomputers in this list, it uses Mellanox HDR to transmit data at speeds of up to 200 Gigabits per second.
Frontera is designed for academic workloads like simulation, modeling, machine learning, and big data. It has been used in various research such as binary black hole mergers, brain cancer treatment, and solving chemical problems through machine learning, modular modeling, and quantum mechanics.
