Processors from the MCST company under the name “Elbrus” appeared not today and not in the past year. The processor’s pedigree goes back to the times of the USSR, when the country was strong, and science was at the forefront of technological progress. “Elbrus” is a series of Soviet and Russian computer systems developed at the Institute of Precision Mechanics and Computer Engineering (IPMCE) in the early 1970s and late 1980s under the direction of Vsevolod Burtsev. Computers based on this computer were widely used in the military-industrial complex and installed on air and missile defence systems in the city of Moscow, and for 20 years, 3 generations of computers have been developed, starting from “Elbrus-1” and ending with “Elbrus-3”, which was tested, but was not put into mass production, but the systems on “Elbrus-2” work to this day at some air defence facilities.
The collapse of the 90’s and the activities of “effective” privatisers destroyed the entire radio-electronic industry of the USSR, and almost all the factories and enterprises of the industry were sold for pennies and were also disposed of for Coca-Cola and chewing gum. It seemed that we were behind forever and nothing can change, but not everything is so simple and unambiguous.
Many readers do not quite understand what a processor is, how many countries are able to produce it, and what is most important in such a complex component of a computer and a personal computer. We will not analyse in detail all the technical aspects of the processor and systems on a chip, just as we will not analyse the ARM architecture processors that are usually used in smartphones. Today, for example, we will analyse standard desktop processors.
Many believe that the leaders in this sphere are western countries, as well as Japan or Korea. But this is not so, or rather Europe and Asian countries do not have their own processors at all. There are processors from Japan or China, but as a rule, these are mobile systems on a chip using a licensed architecture, and the architecture is essentially a processor. But still, the leader among serious CPUs is only one country and that is the US. Intel and AMD have shared the market for so-called x86 processors since the 80s and are monopolists. Attempts to use the x86 architecture were also made in other countries, in particular, Taiwan’s “VIA Technologies” released a CPU on the x86, but under pressure from Intel, everything ended quickly, since x86 is the intellectual property of this company.
Thanks to the MCST company, which continued to refine the Soviet architecture, our country is closing the gap by leaps and bounds. If in 2014 the same “Elbrus-4S” with a frequency of 800 MHz for 4 cores could be compared with AMD processors in 2005-2006, then “Elbrus-8C” in 2016 could be compared with a CPU from Intel in 2012-2013. To say that “Elbrus” is just a VLIW processor is not correct, since it is still a logical continuation of the Soviet architecture, which is now called E2K, which also uses a very long machine instruction set. The question arises – what is Elbrus in our time? And why should it be taken seriously?
“Elbrus” in our time is a processor architecture and a whole family of universal VLIW-microprocessors developed by the Russian company MCST with the participation of INEUM, which, despite the 90s, continued the development of processor technologies of the Soviet “Elbrus” computing systems. The path of the MCST company was long and difficult, from the most ridiculous by modern standards processors “Elbrus 2000” with a frequency of 300 MHz (2005), to the hero of today’s article – the quite decent “Elbrus-16C”.
The newest processor in the line is “Elbrus-16C”, the processor has 16 cores with a frequency of 2 GHz (2 GHz for E2K due to the execution of 48 commands per clock, equates to 3.6-3.8 GHz x86 processor), the processor is made according to modern 16 nm standards, and this is when Intel has been stuck at 14 nm for years. The processor supports a modern eight-channel (8-channel!!!) ddr4-3200 MHz RAM with error correction (ECC).
There is support for virtualisation and support for up to 32 PCIe 3.0 lines, as well as 4 SATA 3.0 channels, while a 10-Gigabit Ethernet controller is integrated into the processor. The performance level of the new product is around 1.5 TFlops in tasks related to single-precision calculations (FP32), and reaches 0.75 TFlops for calculations with double precision (FP64), which is comparable to modern CPUs of the “Ryzen” series from AMD. The processor can work in conjunction with several other CPUs as part of a multiprocessor system that can use up to 4 CPUs and 16 TB of RAM. Moreover, the CPU has an impressive 32 MB of level 3 cache and can work with the Intel x86–64 instruction set, which in general is nonsense.
Mass production of the 16C will begin in 2021, and several dozen engineering samples have been produced for testing, while previous models including the 8C are in production and are already being used for military, government, and commercial purposes. It is unlikely to be possible to see such a CPU on the shelves, and it is not necessary for an ordinary user, since for the most part it is a server solution with an original, but still specific architecture. However, the pace of development of the project is amazing and once we were able to catch up technologically, we hope that we will not fall behind any more.
Armiya i tekhnologii
Copyright © 2022. All Rights Reserved.