According to one definition, a supercomputer is a computer that, when completed, will be among the most powerful computers. Supercomputers are used in high-performance computing (HPC). The computing power of a supercomputer is expressed in flops (floating point operations per second), which tells how many floating-point operations the computer performs per second. . Generally, the theoretical computing power (Rpeak) and the highest computing power (Rmax) achieved in practice with the test software are reported for the supercomputer. For example, in Finland, the theoretical computing power of CSC's supercomputer is about 2.2 teraflops, and the highest computing power obtained in practice is about 1.2 teraflops.
In order to achieve high computing power, an expensive specially made supercomputer is not necessarily needed, but a cluster can be assembled from several ordinary computers available to consumers. For example, in November 2004, the world's seventh most powerful supercomputer was a System X cluster assembled from 1,100 Apple Xserve servers. A supercomputer developed for the purpose, such as Earth Simulator, can be many times faster than one using clustering technology. The speed of a supercomputer does not depend only on CPU power. Since a supercomputer usually has hundreds or thousands of processors, the speed of their mutual communication (the speed of the switching network) easily becomes a bottleneck. There are several communication bus technologies, which are usually called Remote DMA.
The first actual supercomputers entered the market in the 1960s. The first were the IBM 7030 and the University of Manchester's Atlas. Control Data Corporation's (CDC) CDC 6600 computer is considered the first commercial supercomputer, for which the term supercomputer was born. Many of the well-known supercomputers in the 1960s were designed by Seymour Cray at Control Data Corporation (CDC). CDC led the supercomputer market until the 1970s, when Cray left the company and founded his own company, Cray Research. Cray conquered the supercomputer market with its innovative solutions, holding the top spot for 25 years (1965–1990).
The 1980s saw the emergence of a large number of smaller competitors, as was the case with minicomputers a decade earlier, but many of these disappeared in the mid-1990s "supercomputer market collapse". In the 2010s, supercomputers are typically custom jobs manufactured by traditional IT companies such as IBM and HP. These companies bought several companies in the 1980s to gain experience with supercomputers. Cray Inc. continues to specialize in manufacturing supercomputers.
In 2004, the world's most powerful supercomputer was the American IBM Blue Gene/L, which surpassed the Earth Simulator with a performance of about 70 teraflops. BlueGene held the first place among supercomputers until in 2008 the IBM Roadrunner supercomputer surpassed it in calculation speed, being the first to reach petaflops performance.
In November 2010, the Chinese supercomputer Tianhe-1 was the most powerful in the world with a power of about 2.5 petaflops. The record was broken in June 2011 by the Japanese K computer with 10.51 petaflops and in June 2012 by IBM Sequoia with 16.32 petaflops.
In 2016, the first place in the TOP500 list was the Chinese Sunway TaihuLight (93 PFLOPS). In November 2018, the first place was Summit (200 PFLOPS). In June 2020, the first place was taken by Fugaku (415.5 PFLOPS). In May 2022, the first supercomputer Frontier became the fastest. In Finland, in 2021, the LUMI computer will be installed in Kajaani as a joint project of the EU and ten EU countries. Its computing power is reported to be 552 petaflops. The computer rose to third place on the TOP500 list and the fastest in Europe in May 2022. Some supercomputers use GPGPU technology to increase computing capacity. Ex