Page 20 - Kotov

Supercomputers: Testing Limits
Time Frame
Each computer generation had its high-performance leaders. But the 1970s witnessed
appearance of supercomputers, one-of-the-kind custom designed computers with a
top-tier performance. To some extend, they were similar to the racing cars in the auto-
motive industry. Seymour Cray pioneered the supercomputer design and was followed
by other ambitious computer architects. Though by that time the Soviet Union lagged
in the hardware technology and in the computer production capacity with relation to
the USA and Japan, the hope was that Russian designers could outsmart the competi-
tion by developing novel architectural solutions and smart algorithms for solving com-
plex calculation-intensive mathematical problems that were typical supercomputer ap-
plications.All supercomputers exploit computational parallelism in its various forms The Cray-1
and contain multiple (often many) processors. supercomputer

M-13
The main architectural principles developed by Mikhail Kartsev in the
M-10 multiprocessor computer were implemented and extended in his
last project, the VLSI-based supercomputer M-13. Its production started
in 1984, one year after his death, and around 20 computers have been
made. M-13 was intended for using in complex control computer sys-
tems and for processing large amount of information in real time.
Each M-13 processor performed operations on one, two or four pairs
of, respectively, 32 -, 16 - or 8-bit operands. Depending on the version,
M-13 could contain 8,5, 17 or 34 MB of RAM and 4, 8 or 16 processors, The M-13 supercomputer
respectively, achieving performance of 12, 24 or 48 million op /sec. The
maximum speed was equivalent to 2.4 billion instructions per second.

Elbrus
The original Elbrus computer family was developed at the Institute of
Precision Mechanics and Computer Engineering in Moscow in the
1970s. It was patterned after the Burroughs B6700 and B770 stack ar-
chitecture, but it was not a clone of those computers: the instruction
set and data structures were significantly different and included more
advanced mechanisms for data description, protection, and allocation.
The family included Elbrus-1, the first Soviet integrated circuit comput-
er of the fourth generation, with the speed from 1.5 MIPS (mln instruc-
tion per second) to 10 MIPS and a higher-performance Elbrus-2 with
the maximal speed of over 100 MIPS. Both Elbrus-1 and Elbrus-2 were The Elbrus-2 supercomputer
built on the same structural principles, their components were function-
ally identical, and their processors had the same instruction sets and their operating systems were functionally
equivalent.
The Elbrus computers were used in the Soviet space program, nuclear weapons research, and defense systems.
The main components of Elbrus were:

• from 1 to 10 central processors
• from 4 to 32 memory modules
• from 1 to 4 input-output processors
• from 1 to 16 communication processors
• control modules drums and discs, forming the storage management system

Each processor can access any memory module via a switch, which also blocks corrupted modules and provides
backup modules. The reliability was guaranteed by advanced hardware monitoring and controlling processors
and by sharing of information at all levels of the system. The central processor’s instruction set was based on

– 18 –