In 1966 M.Flynn the convenient approach to architecture classification of computing systems has been offered. In a basis concept of a stream was necessary. The stream is understood as sequence of elements, commands or data, the central processor processable. This system's classification is based on consideration number of streams of instructions and dataflows and describes four architectural classes:

• SISD = the Unique Instruction Unique Data
• MISD = the Repeated Instruction Unique Data
• SIMD = the Unique Instruction Repeated Data
• MIMD = the Repeated Instruction Repeated Data

SISD - single instruction stream / single data stream. Consecutive computer systems which have one central processor concern to this class, capable to process only one stream of consistently executed instructions. Now practically high-efficiency systems have all more than one central processor, however, each of them is carried out with untied streams of instructions that does such systems by complexes of the SIMD-systems acting different spaces of data. Conveyor processing can be applied to increase in speed of processing of commands and speed of performance of arithmetic operations. In case of vector systems the vector data stream should be considered as a stream from single indivisible vectors. Examples of computers with architecture SISD are the majority of workstations, for example: Compaq, Hewlett Packard and Sun Microsystems.

MISD - multiple instruction stream / single data stream. In this type of machines the set of instructions should will be executed above the unique dataflow. Till now any real PC getting in the given class, it has not been created. As analogue of such system work, apparently, it is possible to consider work of bank. From any terminal is possible to submit a command and something to make with an available databank. As the database one, and is a lot of commands, we deal with a plural stream of commands and the single data stream.

SIMD - single instruction stream / multiple data stream. These systems have a plenty of processors, within the limits from 1024 up to 16384 which can carry out the same instruction concerning different data in a rigid configuration. The unique instruction is in parallel carried out above many elements of data. Examples SIMD of machines are systems CPP DAP, Scale II Quadrics and Apemille. Other subclass of SIMD-systems - vector computers. Vector computers manipulate files of similar data. It just as scalar servers process separate elements of such files. It is done due to use of specially designed vector central processors. When data are processed by means of vector modules, results can be given out on one, two or three steps of the generator of frequency (the step of the generator of frequency is the basic time parameter of system). At work in a vector mode vector processors are processed by data practically in parallel, that does them faster, than at work in a scalar mode.

MIMD - multiple instruction stream / multiple data stream. These servers in parallel carry out some streams of instructions above various streams. Unlike the multiprocessing SISD-servers mentioned above, a command and data are connected, because they represent various parts of the same carried out problems. For example, MIMD-systems can carry out in parallel set of subtasks, with the purpose of time reduction of the primary goal performance. Presence of the big variety of systems getting in a given class, does classification Flynn not completely adequate. Really and four processor SX-5 companies NEC and one thousand processor Cray T3E both get in this class. It forces to use other approach to the classification, differently describing classes of computer systems. The basic idea of such approach can consist, for example, in the following. We consider, that the plural stream of commands can be processed by two ways: Or one conveyor device of the processing, working in a mode of division of time for separate streams, or each stream is processed by the own device. The first opportunity is used in MIMD servers which usually name conveyor or vector, the second - in parallel servers. In a basis of vector computers the concept of conveyoring - obvious segmentation of the arithmetic device on separate parts lays, each of which carries out the subtask for pair operands. In a basis of a parallel computer idea of use for the decision one problem of the several processors working together lays, and processors can be both scalar, and vector.