A supercomputer is called such a computer, which in performance and other technical characteristics is much superior to others that currently exist. The composition of such a computer includes several processors. Another distinguishing characteristic of such computing devices is the use of vector arithmetic, that is, they can perform arithmetic operations simultaneously on several pairs of numbers. For example, a typical supercomputer can simultaneously calculate the wages of several employees, while a regular computer at the same time will calculate the salary of only one employee.
The history of supercomputers: the advent of supercomputers in the 1960s.
The first supercomputer was created at Control Data Corporation (CDC) under the direction of Seymour Crey. One of the first computers developed in this company was Cray CDC 1604. Vacuum electron tubes were replaced by transistors in it, and it quickly gained popularity in scientific laboratories. CDC later developed the CDC 7600 supercomputer and began work on the CDC 8600. In 1964, the fastest computer on Earth was Stretch, which could perform three million floating point operations per second (FLOPS).
One of the advantages of computers developed under the direction of Seymour Crey was the tight packaging of electronic components, which increased the performance of computers. All Seymour Crey computers have been optimized for demanding scientific applications, for example, solving differential equations, matrix calculations, seismic analysis, linear programming, and other similar tasks.
Cray supercomputers in the 1970s
Seymour Cray left CDC and founded Cray Research, Inc. in 1972. In 1975, Cray Research released the Cray-1 computer, which belongs to the 4th generation of computers. In total, more than 80 such machines were sold, which for that time was a great success. Cray-1 was one of the first computers on which time-consuming operations could occur on several processor devices at once, and thus was one of the first "multiprocessor" devices.
One of the pioneers of multiprocessing was Cray X-MP, introduced in 1982, which linked two Cray-1 computers. He was also the first computer to implement vector computing.
In addition, in the 1970s. the first 32-bit super-mini-computers appeared.
The development of supercomputers in the 1980s.
In 1985, Cray Research introduced the Cray-2 quad-core computer. It was the first computing device to deliver more than one billion FLOPS.
In 1983, Daniel Hillis, a graduate student at the Massachusetts Institute of Technology, came up with how to improve the performance of multiprocessor systems belonging to the 4th generation of computers. And in the same year, he co-founded Thinking Machines Corporation. In 1985, this company developed its first CM-1 computer. He used 65,536 inexpensive single-bit processors, which were grouped in 16 pieces. on one chip. The performance of the CM-1 computer in some operations reached several billion FLOPS and was comparable to the fastest Cray supercomputer at that time.
Further development of supercomputers in the 1990s - early 2000s.
Important customers for supercomputers were the military. After the United States signed the Comprehensive Nuclear Test Ban Treaty in 1996, an alternative certification program for nuclear warheads became necessary. Therefore, the US Department of Energy allocated money for a new program for the development of supercomputers, the aim of which was to develop a computer capable of simulating nuclear tests by 2004. This computer should have a performance of more than 100 trillion FLOPS, and the fastest existing computer at that time was Cray T3E, with a capacity of up to 150 billion FLOPS. The ASCI Red supercomputer, built at Sandia's National Laboratories in Albuquerque, in collaboration with Intel, was the first to reach 1 TFLOPS. It involved 9,072 standard Pentium Pro processors.
Japanese supercomputer
While the multiprocessor approach prevailed in the United States, in Japan, NEC has reverted to an older approach — the individual design of a computer chip. Made by this corporation, Earth Simulator computer took first place in the list of the most productive computers in 2002.
Modern computers
In 2004, the fastest supercomputer was Blue Gene / L, released by IBM. Its performance was approximately equal to 36 TFLOPS. After two doubles in the number of processors, Blue Gene / L, installed in 2005 at Sandia National Laboratories in Livermore, California, was the first machine to break through the 100 TFLOPS performance barrier.
The first computer, whose performance exceeded 1000 TFLOPS or 1 petaflop, was built by IBM in 2008.
Supercomputers Application
Super-computers are used in the scientific field to perform labor-intensive calculations and process a large amount of information in real time. In addition, advances in computer technology have enabled scientists to use accurate models of ongoing processes, instead of the simplified ones used previously.
In mathematics, with the help of supercomputers, problems of cryptography and statistics are solved. In physics, they help to understand the processes occurring inside the atom. Supercomputers are helped to decrypt DNA. They are also indispensable in the preparation of weather forecasts, the study of Earth's climate changes and the search for oil and gas deposits. Supercomputers are also used to perform military calculations related to nuclear weapons.
The use of powerful computers allowed us to make a number of breakthroughs in areas such as meteorology, global climate analysis, the creation of new medicines and aerospace technology.
Supercomputer Review
When talking about heavy duty computers, the question often arises: "Which computer is the fastest?" The answer to this question can give a rating of the 10 most powerful supercomputers. This rating presents the latest computers.
- The fastest computer at the moment is the Summit Power System AC922 computer. Its performance according to the data obtained using the LINPACK test system is 122.3 PFLOPS. The maximum theoretical performance of this computing device is 187.659 PFLOPS. The Summit Power System AC922 supercomputer was made by IBM specifically for use in the Oak Ridge National Laboratory.
- In second place in terms of performance is the Chinese supercomputer Sunway TaihuLight. The computational speed of this computer, which was measured using the LINPACK test system, is 93 PFLOPS. This supercomputer was the most productive in the world from June 2016 to June 2018. This supercomputer is located in the People's Republic of China, at the computer center in Wuxi, and is used for weather forecasting, medical research, and various complex calculations.
- The next highest performance is the Sierra Power System S922LC computing device. This supercomputer has a performance of 71.61 PFLOPS, according to LINPACK tests. This device is located in the Livermore Laboratory. E. Lawrence, part of the University of California.
- The Tianhe-2 supercomputer was the most powerful computing device from 2013 to 2016. Its name is translated from Chinese as “Milky Way - 2”. According to the LINPACK standard test, its performance is 61.445 PFLOPS, and the theoretical peak is 100.679. This device is located in the National Computer Center in Guangzhou (People's Republic of China).
- In fifth place in terms of performance at the moment is the Japanese AI Bridging Cloud Infrastructure supercomputer. Its LINPACK benchmark performance is 19.88 and its maximum theoretical performance is 32.577 PFLOPS.
- The Piz Daint supercomputer is located in the Swiss computer center and is the most powerful computing device in Europe. Its peak theoretical performance is 25.326 PFLOPS, and real, recorded using LINPACK tests, is 19.59 PFLOPS. Developed by the American company Cray.
- The seventh place in performance is taken by the Titan supercomputer, released in 2012 based on the Cray XK7 architecture. The real performance of this device, measured using the LINPACK test suite, is 17.59 PFLOPS, and the maximum theoretical performance is 27.113 PFLOPS. He works in the laboratory of the US Department of Energy, Tennessee. It was the most powerful computing device from November 2012 to July 2013.
- The Sequoia supercomputer was developed by IBM on the Blue Gene / Q platform. Its real performance is 17,173 PFLOPS, and theoretically possible is 20,133 PFLOPS. Located in the Livermore Laboratory.
- The Trinity computer is based on the Cray XC40 platform. The measured performance of this computing device is 14,137 PFLOPS. Installed at the Los Alamos Laboratory.
- The Cori supercomputer, like the previous one, is based on the Cray XC40 architecture. Its LINPACK benchmark performance is 14.015 PFLOPS.
Conclusion
The development of supercomputers has had a great influence on many fields of science and industry. At the moment, the biggest obstacle hindering the disclosure of the entire computing potential of such devices is the difficulty of writing programs that could simultaneously load all the processors in the supercomputer at full capacity. This happens because writing a program that effectively splits a computational task into several threads is much more difficult than writing one that runs sequentially on a single processor. And not every task lends itself to such parallelization. That's all you need to know about super-computers, the purpose, capabilities and principles of building these computers.