For most of us, a computer probably seems fast enough if it can run 8K videos or the latest version of Far Cry in 60 fps without slowing down. However, there are many complicated tasks that require billions of calculations per second – something a desktop with i9 processor can’t do.
That’s where supercomputers come in handy. They offer a high level of performance that allows governments and organizations to solve problems that wouldn’t be possible with conventional computers.
Today’s supercomputers are built with AI (artificial intelligence) workloads in mind. In addition to weather forecasting, climate research, physical simulations, and oil and gas exploration, supercomputers help scientists discover more resilient building materials and study humans proteins and cellular systems at an extreme level of detail.
Usually, the supercomputer’s performance is measured in floating-point operations per second (FLOPS). In the field of scientific computations, FLOPS is a more accurate figure than measuring instructions per second.
Did you know the first supercomputer — Livermore Atomic Research Computer — was built for the US Navy Research and Development Centre in 1960.
To show you how far we have come since then, we have curated a detailed list of fastest supercomputers in the world. They all are non-distributed computer systems running on Linux.
12. Sequoia
Speed: 17.1 petaFLOPS
Cores: 1,572,864
Vendor: IBM
Location: Lawrence Livermore National Laboratory, United States
Sequoia uses IBM’s BlueGene/Q servers to deliver a theoretical peak performance of 20 petaFLOPS. It has 123% more cores and is 37% more energy efficient than its predecessor K computer.
Although the machine is mostly used for nuclear weapons simulation, it is also available for many scientific purposes such as climate change and human genome analysis. It has also demonstrated its great scalability with a 3D simulation of the human heart’s electrophysiology.
11. PANGEA III
Credit: Total S.A.
Speed: 17.8 petaFLOPS
Cores: 291,024
Vendor: IBM
Location: CSTJF technical and scientific research center in Pau, France
Pangea III relies on IBM’s AI-optimized, high-performance architecture. IBM and NVIDIA worked together to build the industry’s only CPU-to-GPU NVLink connection, which enables over 5 times faster memory bandwidth between the IBM POWER9 CPU and NVIDIA Tesla V100 Tensor Core GPUs than the conventional x86-based systems.
The architecture not only improves computing performance but also enhances energy efficiency. The new system uses less than 10% of the energy consumption per petaFLOP as its predecessor, Pangea I and II.
Pangea III has various applications, especially in three different fields – exploration and development seismic imaging, development and production models, and asset valuation and selectivity.
10. Lassen
Speed: 18.2 petaFLOPS
Cores: 288,288
Vendor: IBM
Location: Lawrence Livermore National Laboratory, United States
Lassen is designated for unclassified simulation and analysis. It is installed in the same lab and using the same building components as Sierra (#2 fastest supercomputer).
Although Sierra is a big system, Lassen is a decent size in its own right: it is exactly 1/6th of the size of its larger brother. Lassen system is contained in 40 racks, while Sierra hogs up 240 racks.
IBM Power9 processors and 253 terabytes of main memory help Lassen to achieve to a perk performance of 23 petaFLOPS.
9. SuperMUC-NG
Speed: 19.4 petaFLOPS
Cores: 305,856
Vendor: Lenovo
Location: Leibniz Supercomputing Centre, Germany
SuperMUC-NG features 6,400 Lenovo ThinkSystem SD650 direct-water-cooled computing nodes with over 700 terabytes of main memory and 70 petabytes of disk storage.
It is connected to powerful visualization systems that contain a large 4K stereoscopic powerwall and a 5-sided CAVE artificial virtual reality environment.
The supercomputer servers European scientists of many fields, including genome analysis, fluid dynamics, quantum chromodynamics, life sciences, medicine, and astrophysics.
8. AI Bridging Cloud Infrastructure
Credit: ABCI
Speed: 19.8 petaFLOPS
Cores: 391,680
Vendor: Fujitsu
Location: National Institute of Advanced Industrial Science and Technology, Japan
This is the world’s first large-scale Open AI Computing Infrastructure that delivers 32.577 petaFLOPS of peak performance. It has a total of 1,088 nodes, each containing 2 Intel Xenon Gold Scalable processors, 4 NVIDIA Tesla V100 GPU, 2 InfiniBand EDR HCAs, and 1 NVMe SSD.
Fujitsu Limited claims that the supercomputer can achieve 20 times the thermal density of conventional data centers, and a cooling capacity of 70 kW Rack by using hot water and air cooling.
7. Trinity
Speed: 21.2 petaFLOPS
Cores: 979,072
Vendor: Cray
Location: Los Alamos National Laboratory, United States
Trinity is built to provide an extraordinary computational capability for the NNSA Nuclear Security Enterprise. It aims to improve geometric and physics fidelities in nuclear weapons simulation code, while ensuring that the nuclear stockpile is safe, secure, and effective.
The supercomputer was developed in two stages: the first stage incorporated the Intel Xeon Haswell processor, and the second stage included a substantial performance increase using the Intel Xeon Phi Knights Landing Processor. It can deliver a total peak performance of over 41 petaFLOPS.
6. Piz Daint
Speed: 21.2 petaFLOPS
Cores: 387,872
Vendor: Cray
Location: Swiss National Supercomputing Centre, Switzerland
This supercomputer, named after the mountain Piz Daint in the Swiss Alps, runs on Intel Xeon E5-26xx microprocessor and NVIDIA Tesla P100.
Piz Daint utilizes DataWarp’s ‘burst buffer mode’ to increase effective bandwidth to and from storage devices. This accelerates the data input/output rates, facilitating the analysis of millions of small, unstructured files.
In addition to its daily tasks, it can handle the data analysis of some of the world’s most data-intensive projects, such as data collected from experiments at the Large Hadron Collider.
Read: What Is A Particle Accelerator?
5. Frontera
A view between two rows of Frontera servers | Credit: TACC
Speed: 23.5 petaFLOPS
Cores: 448,448
Vendor: Dell EMC
Location: Texas Advanced Computing Center, United States
Frontera opens up new possibilities in engineering and research by providing extensive computational resources that make it easier for scientists to tackle many complex challenges across a wide range of domains.
Frontera features two computing subsystems: the first one focuses on double-precision performance while the second one focuses on single-precision stream-memory computing. It also has cloud interfaces and multiple application nodes for hosting virtual servers.
4. Tianhe-2A
Tianhe-2 in National Supercomputer Center in Guangzhou
Speed: 61.4 petaFLOPS
Cores: 4,981,760
Vendor: NUDT
Location: National Supercomputing Center in Guangzhou, China
With more than 16,000 computer nodes, Tianhe-2A represents the world’s largest installation of Intel Ivy Bridge and Xeon Phi processors. While each node has 88 gigabytes of memory, the total memory (CPU+coprocessor) is 1,375 tebibyte.
China spent 2.4 billion yuan (US$390 million) on building this supercomputer. It is now mostly used in simulations, analysis, and government security applications.
3. Sunway TaihuLight
Speed: 93 petaFLOPS
Cores: 10,649,600
Vendor: NRCPC
Location: National Supercomputing Center in Wuxi, China
The computing power of TaihuLight comes from a homegrown several-core SW26010 CPU that includes both computing processing elements and management processing elements.
A single SW26010 provides a peak performance of more than 3 teraFLOPS, thanks to its 260 processing elements (integrated into one CPU). Each computing processing element has a scratchpad memory that serves as a user-controlled cache, significantly reducing the memory bottleneck in most applications.
In addition to life sciences and pharmaceutical research, TaihuLight has been used to simulate the universe with 10 trillion digital particles. However, China is trying to achieve a lot more: the country has already stated its goal to be the leader in AI by 2030.
2. Sierra
Image credit: Wikimedia
Speed: 94.6 petaFLOPS
Cores: 1,572,480
Vendor: IBM
Location: Lawrence Livermore National Laboratory, United States
Sierra offers up to 6 times the sustained performance and 7 times the workload performance of its predecessor Sequoia. It combines two types of processor chips: IBM’s Power 9 processors and NVIDIA’s Volta GPUs.
Sierra is specifically designed for assessing the performance of nuclear weapon systems. It is used for predictive applications in stockpile stewardship, the US program of reliability testing and maintenance of nuclear weapons without any nuclear testing.
Read: What Is Tensor Processing Unit (TPU)? How Is It Different From GPU?
1. Summit
Image credit: ORNL
Speed: 148.6 petaFLOPS
Cores: 2,414,592
Vendor: IBM
Location: Oak Ridge National Laboratory, United States
Summit is the fastest supercomputer in the world that can deliver 200 petaFLOPS at peak. This is equivalent to 200 quadrillion floating-point operations per second.
It is also the world’s third most energy-efficient supercomputer, with a recorded power efficiency of 14.66 gigaFLOPS per watt.
Summit’s 4,600+ servers, which take up the size of two basketball courts, house more than 9,200 IBM Power9 processors and over 27,600 NVIDIA Tesla V100 GPUs. The system is connected by 185 miles of fiber optic cable, and it consumes enough power to run 8,100 homes
In 2018, Summit became the first supercomputer to break the exascale barrier. While analyzing genomic data, it achieved a peak throughput of 1.88 exaops, which is nearly 2 billion billion calculations per second.
Read: 18 Most Interesting Facts About Quantum Computers
The United States aims to develop a fully capable exascale computing ecosystem for scientific studies by 2021, and Summit is a step towards that.