Review of the Elbrus-4C microprocessor and the Russian Elbrus-401 computer (UPD: launched for sale)

Story

Elbrus is a line of Russian microprocessors. Historically, the first prototype of modern Elbrus processors was the Elbrus-3 supercomputer. Unfortunately, it was not fully assembled, but only in the form of several cabinets. The adjustment was not completed, and these cabinets were dismantled.

But the ideas that were put into Elbrus-3 did not fade into oblivion and were revived in the first Russian microprocessor Elbrus. Its release fell on 2007.

Over the next decade, CPUs improved, their gigaflops power increased, and the nanometers in the process technology decreased. Significant achievements were the development of the Elbrus-2C+, Elbrus-4C, and Elbrus-8C processors.

How Russian are they?

Stories about Russian electronics manufacturers are often punctuated by criticism about the country where processors are made. Thus, in Russia, the production of crystals is currently not very developed - even Elbrus and Baikal are produced at the facilities of TSMC in Taiwan

. Exclusively Russian production is still limited to the 65 nm process technology – there is no more precise production in Russia yet.

TSMC FAB 15 Factory Building, Taiwan

But, as mentioned above, manufacturing the physical processor core is only the least of the tasks. Moreover, TSMC produces processor wafers for almost all major electronics manufacturers, including Intel (not all, but, for example, Atom models), Qualcomm, AMD, NVIDIA, ARM, Apple and many others, and occupies more than 55% of the market contract manufacturing

microchips in the world.

The rest of the processors are exclusively of Russian origin. The most important thing is that the rights to the chip topology (that is, the processor design itself) remain in Russia, which is why they are called Russian manufacturers. For example, the architecture, circuit design and topology of the Elbrus-8S processor are being developed

specialists from the Russian Institute of Electronic Control Machines, so this processor meets the requirements of government contracts.

Initially, the very idea of ​​producing processors in Russia was related to security issues - some data simply cannot be processed on computers with Intel or AMD processors, which remain at risk of data leakage

.

So even processors with Taiwanese-made cores are already a big step towards import substitution

. But, alas, Russian achievements have not yet reached household use and are unlikely to reach it in the foreseeable future.

MCST developments

A chain of developments led to the development in 2020 of a highly modified version of the latest Elbrus-8SV processor. Its serial production is planned for 2020. The main indicator of the Russian CPU is computing power - 580 gigaflops of single precision and an advanced chip. The power of 8SV is more than 100 times higher than that of the first CPU from this line.

Elbrus is five generations of microprocessors. Of these, 4 generations are in mass production.

The chips contain from one to 8 cores, including a chip with an integrated graphics core. The technologies used are from 130 nm to 28 nm.

The specificity of these microprocessors is that the instruction system is Russian. It was developed by specialists. Has much in common with the Elbrus-3 command system.

A modern processor produces up to 25 operations per clock (8С and 8СВ) and is universal (works with almost any software).

Company's achievements

Since MCST themselves developed the command system and use a non-standard approach, they also developed the processor logic itself. This is not some licensed version of ARM, Intel and others processors. This is an independent development in Russia.

Developed in Russia:

• the logic of the kernel itself;
• chip topology;
• functional blocks that are in the MPC in addition to the core itself (cache memory, memory controller, peripheral controllers to provide I/O channels);

Some blocks have to be taken from other factories, licensed and applied to their own developments. Unfortunately, this is how the microprocessor design chain works, and the MCST company cannot develop everything. But this is not an obstacle to consider the processor not Russian.

Construction process

CPU development is a fairly similar process to software development. The description of the main logical blocks is carried out in a high-level language:

• Verilog;
• HDL;
• Others similar in form to the C programming language.

When a developer has described a logical node, he translates this description in a high-level language into logical chains. Next comes the integration into a single project of the developments of the MCST company and developments from other companies (for example, logical and physical blocks that implement an external controller). Compared to software programming, this is an analogue of external plug-in libraries.

Next, a topology is created - the final placement of all logical circuits along with transistors and licensed blocks (issued in the form of ready-made pieces of the topology) on the chip, and ensuring the full functioning of this crystal. This is similar to compiling a program along with all the libraries and assembling it into one executable binary file.

Ultimately, the topology is like a set of drawings - actual drawings of the tracks. These are drawings of the layers from which transistors and other active and passive elements will be obtained. Later, these drawings are sent to the factory.

The factory performs a logically simple, but at the same time subtle and complex operation:

• transfers these patterns to the surface of the crystal;
• carries out etching;
• carries out alloying.

That is, it launches a whole cycle of technological operations to transfer layers to a silicon wafer, which together will create a working microcircuit on a chip.

But the crystal itself cannot be used, so a substrate is created together with it - this is actually part of the chip body. The plate is a miniature printed circuit board.

A crystal is soldered onto this printed circuit board. The latter has special contacts that are soldered to the contact pad on the substrate. The substrate extends the pads to the pads on the underside of the chip.

All operations are done using specialized computer-aided design (CAD) development tools.

The controller for the processor and the printed circuit board are also domestic developments, because motherboards from foreign companies are not suitable for this line of processors.

The motherboard for Elbrus processors uses both domestic developments for installing the CPU itself, and developments from other companies for connecting hard drives, RAM, graphics cards and other devices required by the customer.

History of the appearance and features of domestic processor architecture

"Elbrus"

is primarily the name of the processor architecture and supercomputers developed on its basis.
Initially, they were created as part of missile defense systems commissioned by the military. The processor architecture
describes the set of supported instruction sets, the method of their execution, the registers used and addressing.

Development began in 1973 at the Lebedev Institute of Precision Mechanics and Computer Technology (ITMiVT) under the leadership of Academician Vsevolod Sergeevich Burtsev, a scientist in the field of control systems and the theory of design of universal computers.

When creating the machine, we were guided by advanced technologies at that time: superscalarity and multiprocessing.

Superscalar architecture

– a processor architecture that uses multiple instruction decoders that transmit executing instructions simultaneously to multiple execution units. Scheduling the execution of the instruction stream is dynamic and is carried out by the CPU itself.

First Multiprocessor Computing Complex (MCC) “Elbrus-1”

was put into operation in 1980. It could contain up to 10 processors and showed a performance of 12 million operations per second. The amount of RAM was 64 MB (or 220 machine words).

Multiprocessor computing complex “Elbrus-1”

But computer technology developed by leaps and bounds, and already in 1985 the next modification of the MVK appeared. It was named “Elbrus-2”

and due to the use of a new element base, productivity increased to 125 million op/s when combining 10 processors (2 of them were backup). Also, the amount of RAM has increased to 144 MB. “Elbrus-2” has found its application in the following projects:

• Radar “Don-2N” - a stationary all-round radar station, the main center of the Moscow missile defense system;
• Space Flight Control Center;
• Nuclear Center Arzamas-16 (now the closed city of Sarov) is the first nuclear research center in the USSR, part of the structure of the Rosatom State Corporation;
• Nuclear center Chelyabinsk-70 is the now closed city of Snezhinsk in the structure of enterprises of the Rosatom State Corporation.

At the same time, a simplified version of the MVK was produced - “Elbrus-1K2”

and
“Elbrus-B”
, which were used to smoothly replace outdated BESM-6 computing systems.

After the successful commissioning of Elbrus-2, the development of the next modification, which received the expected name Elbrus-3, was actively underway.

. It was planned to have many architectural improvements and the use of 16 processors. However, due to a number of historical events and financial difficulties, this project was not completed.

Specifications

The table shows the characteristics of processors in use and soon to be put into operation.

What is the Elbrus-4S microprocessor?

Elbrus-4C is a quad-core processor operating at 800 MHz, which supports three memory channels. There is also a cache memory with a total capacity of 8 Megabytes. The processor is manufactured using 65 nanometer technology, its average power consumption is 45 watts.

“This is a universal microprocessor that is characterized by unique features of its architecture. Depending on the purpose of a particular technique, it can be used in harsh conditions. For example, some equipment can be immersed in water, some can be used at the North Pole, or used at temperatures below 40 degrees,” Vasily Vorobushkov, chief designer of the Elbrus 401-PC ,

Question answer

Why are there many hackers among Russians?

Possibilities

To initialize all components of a finished computer on Elbrus, an analogue of the BIOS is used, called the initial start program. It is capable of performing preparatory work at startup and transferring control to loading the operating system.

Windows or Linux cannot be used in this version of the computer. We use our own OS from MCST. Several of them have appeared over the years, and they are all united under the name “Elbrus Operating Systems”. Soon, along with the serial release of Elbrus-8SV, a new Elbrus-linux OS will be released - based on the Linux kernel. In terms of package composition, the latest OS is close to Debian 9 version.

Open source applications can be ported to Elbrus-linux without any problems. Games from Open Source were also ported and launched on the 4th generation Elbrus processor (4 cores, 800 MHz). For example:

1. Console Doom 3.
2. The Elder Scrolls III: Morrowind.
3. Counter strike 1.6.

Such games were ported by the MCST team. It is not yet profitable for developers to port games to the Elbrus system, because they will not receive any profit from this at this stage of development of the ecosystem. But it is possible that if the sources of modern popular games (Dota 2, GTA 5, PUBG) are in the hands of developers, they will be able to compile them and run them on a PC without any problems.

At a Crossroads: Collaboration with Sun Microsystems

After the collapse of the USSR, on the basis of the ITM&VT team in 1992, the Moscow Center for SPARC Technologies (MCST) LLP (now JSC MCST) was created. Until 1996, the new enterprise collaborated with the American company Sun Microsystems, which promoted its computers with the SPARC architecture (which is reflected in its name).

SPARC

(
Scalable Processor ARChitecture
- scalable processor architecture) - 32- and 64-bit open microprocessor architecture, which is based on a reduced instruction set (RISC).

Joint work with a large company allowed MCST to gain access to advanced technologies in processor design, writing compilers, creating operating systems, etc. As a result, until 2007, only microprocessors with the SPARC architecture and computing systems based on them were produced: MCST-R100, MCST-R150 , MCST-R500 and MCST-R500S.

MCST-R500 processor based on SPARC architecture

Nevertheless, this period allowed the MCST to stay afloat, preserve and supplement the scientific and technical base, and the native architecture was not forgotten.

Comparison with Intel

In terms of capitalization and developments, Intel is many times superior to MCST. But this does not prevent us from comparing fundamentally and technologically different processors according to different indicators.

Download. Free does not mean accessible

Not a bad “machine” for work

In May 2020, MSTC published the original installation files of several versions of the operating system on its website.

The distribution kit, list of packages and documentation for the junior open version, compatible with x86 processors, have been opened for download. A more modern version of the system is not yet available.

Perhaps the most interesting are the versions for running on MCST processors of the Elbrus and R lines (SPARC architecture). But they are only available upon request and require a computer with the appropriate architecture.

Therefore, today we will limit ourselves to the regular version of the Elbrus OS , compatible with ordinary desktop computers.

However, like everything in Russia, open does not mean accessible. The download process took me a total of almost a month of rare attempts: the developer did not bother to upload the distribution kit to a normal exchanger.

The dial-up speed and constant interruptions due to the “habra effect” delayed the process until the download mirror appeared on Yandex.Disk.

Future plans

MCST is not going to stop there, so in 2021 mass production of the Elbrus-16C processor currently under development will be launched. This will be a sixteen-core CPU with performance of 750 Gflops/s double precision and 1.5 Tflops/s single precision.

Like 8SV, this will be a fifth-generation CPU architecture. Features include:

• Using a system-on-chip to introduce a “south bridge” into the CPU;
• Support for virtualization at the hardware level, including Intel x86/64 codes;
• Increased kernel performance by supporting dynamic optimization.

It promises to use a 16 nm process technology with about 6 billion transistors used on a chip area of ​​400 square meters. mm.

The price for mass production will become affordable for the consumer. The adequate cost of Russian Elbrus processors and the availability of personal computers based on them will be able to attract many users to work with the domestic ecosystem.

Russian processors Baikal-T1 and Baikal-M

If Elbrus processors are intended purely for computers and are ready to compete with other manufacturers of PC processors, then Baikal processors are intended more for the industrial segment and will not face such tough competition. However, Baikal-M processors are already being developed, which can be used for desktop PCs.

Processor Baikal-T1

According to Baikal Electronics, Baikal-T1 can be used for routers, routers and other telecommunications equipment, for thin clients and office equipment, for multimedia centers, and CNC systems. But Baikal-M can become the heart for work PCs, for industrial automation and for building management. Already more interesting! But there is no detailed information about technical characteristics yet. We only know that it will run on 8 ARMv8-A cores and will have up to eight ARM Mali-T628 graphics cores on board and, what is also important, the manufacturers promise to make it very energy efficient. Let's see what happens.

While I was writing the article, I made a request to Baikal Electronics JSC, and the answer was not long in coming. Dear Andrey Petrovich Malafeev (public relations and corporate events manager) kindly shared with us the latest information about the Baikal-M processor .

The company plans to release the first engineering samples of the Baikal-M processor this fall. And then I quote, so as not to distort the essence of the information in any way:

— Start of quote —

The Baikal-M processor is a system on a chip that includes energy-efficient processor cores with ARMv8 architecture, a graphics subsystem and a set of high-speed interfaces. Baikal-M can be used as a trusted processor with extensive data protection capabilities in a number of devices in the B2C and B2B segments.

Areas of application of Baikal-M

• monoblock, automated workstation, graphic workstation;
• home (office) media center;
• video conference server and terminal;
• microserver;
• Small enterprise level NAS;
• router/firewall.

The high degree of integration of the Baikal-M processor allows the development of compact products in which the main share of added value comes from the domestic processor. Availability of complete information about the logical circuit and physical topology of the chip, combined with trusted software and associated hardware solutions, allows the processor to be used as part of systems designed to process confidential information.

Applicable software

The widespread use of the ARMv8 (AArch64) architecture allows the use of a huge amount of ready-made application and system software. Linux and Android operating systems are supported, including at the level of binary distributions and packages. Drivers for numerous PCIe and USB devices are available. The software package supplied by Baikal Electronics includes the Linux kernel in source and compiled form, as well as drivers for the controllers built into Baikal-M.

Main characteristics of the Baikal-M processor

• 8 ARM Cortex-A57 cores (64 bit).
• Operating frequency up to 2 GHz.
• Hardware support for virtualization and Trust Zone technology at the level of the entire SoC.
• Interface with RAM – two 64-bit DDR3/DDR4-2133 channels with ECC support
• Cache – 4 MB (L2) + 8 MB (L3).
• Eight-core Mali-T628 graphics coprocessor.
• Video path providing support for HDMI, LVDS
• Hardware video decoding
• The built-in PCI Express controller supports 16 PCIe Gen lanes. 3.
• Two 10 Gigabit Ethernet controllers, two Gigabit Ethernet controllers. The controllers support virtual VLANs and traffic prioritization.
• Two SATA 6G controllers providing data transfer speeds of up to 6 Gbit/s each.
• 2 USB v.3.0 channels and 4 USB v.2.0 channels.
• Support for trusted boot mode.
• Hardware accelerators supporting GOST 28147-89, GOST R 34.11-2012.
• Energy consumption – no more than 30 W.

— End of quote —

What do you say, friends? Did Russian processors impress you or leave you indifferent? Personally, I believe in the great future of Russian digital technologies!

Architecture

As the most important result, the company ZAO MCST developed the original Elbrus microprocessor architecture. The processor is focused on obtaining the maximum performance indicators for the given hardware resources. In the general classification, it belongs to the category of architectures that use the VLIW (Very Large Instruction Word) principle, when the compiler generates sequences of command groups (wide command words) for parallel execution, in which there are no dependencies between commands within the group and dependencies are minimized between teams in different groups.

Thus, the Russian Elbrus processor makes high use of the operation-level parallelism present in this program code. As a result, greater architectural speed is achieved by freeing the hardware from the parallelization functions inherent in superscalar architectures and transferring them to an optimizing compiler. This led to another important feature characteristic of the Elbrus architecture - the low level of energy consumption of the equipment.

Along with the effective use of parallelism of operations, the architecture of the Elbrus device includes the implementation of other types (levels) of parallelism inherent in the computing process:

• parallelism of tasks in multi-machine systems;
• parallelism of control threads on shared memory;
• vector parallelism.

Microchip production in Russia

In Russia and Belarus there are five large microprocessor production facilities - Zelenograd Mikron and Angstrem (bankrupt in 2019 [28]), a secret factory in Kurchatov/NIISI, auxiliary production in Voronezh and the Integral factory in Belarus.

Also, in addition to large production facilities in Russia, there are several small ones, with technologies at the 1.5-10 micron level (for Roscosmos and co.), but they do not fulfill commercial orders, and there is very little information on them. So, it is difficult to calculate the total number of factories.

Mikron and Angstrom use equipment purchased from ST, AMD and IBM. Micron already actually produces microcircuits at 90 nanometer standards on 200 mm wafers (SRAM and Elbrus). The 65 nm process technology is slowly being tinkered with, the first prototype was released back in 2014, and in 2020 the pure CMOS process finally started working. On Angstrem - 600 nm on the old line, 130 nm from AMD and 90 nm from IBM on 200 mm wafers were launched by the beginning of 2020.

At this point, Russian ill-informed pessimists shout “horror, horror, but Intel has 14 nanometers, and will soon launch 12.” This is due to the common misconception that advanced devices can supposedly be made exclusively on the latest “nanometer”. This, of course, is not the case - an advanced process may be too expensive or not suitable for temperature characteristics, for example. The simplest example is the STM32 (French-Italian company) microcontroller, which is very popular in Russia and is leading in its class, based on the British ARM Cortex M4, which has been produced since 2011 to this day. It is made using 90 nanometer technology.

Russian factories Mikron and Angstrem can be used for the production of certain products, such as microcontrollers. In addition, they are of strategic importance - specialists study around them, whose experience will also be useful in contract manufacturing at the Taiwanese TSMC.

More complicated is the situation with fraternal Belarus, which lives on contracts for the production of cheap microcircuits for Russia. Modernizing this plant would require a lot of money, which Belarus is in no hurry to invest yet. Nevertheless, there are a large number of microelectronics specialists working around Integral, who can be used to design processors.

The old 800 nm line at Integral works fine, the 350 nm line was launched for quite a long time, but in the end it was debugged and launched. It is noteworthy that Integral has a relatively high percentage of domestic consumables (starting from plates).

It is important to understand that the United States imposes restrictions on the transfer of technology to build a factory in Russia according to the latest standards. But even the construction of a factory that lags behind the cutting-edge (“minus three generations”) would require an investment of 5-6 billion dollars, and a lot of additional resources would have to be spent on training specialists. In this sense, the current Mikron and Angstrom (the equipment from which was purchased at prices an order of magnitude lower) represent a good compromise for the current moment. For now, Russian designers can use Mikron for some projects, and TSMC for more complex ones (like Baikal).

It is also worth mentioning Crocus Technology, which takes finished CMOS wafers to Russia, deposits MRAM layers here, and then sends them back abroad for the final layers.

The Dutch company Mapper has a production site for MEMS components in Russia. This area is already working - it is photolithography with micron standards, which Mapper opened, probably to fulfill the formal requirements of Rusnano. These areas do not provide Russia with the opportunity to obtain advanced photolithographic equipment bypassing US export restrictions.

There are also a number of production facilities for microwave microcircuits on non-silicon substrates (for APAA and co., microwave microstrip filters), with electronic lithography, etc. (ISHFPE RAS and co.).

When design in Russia develops, the issue of more expensive factories can be considered again. Unfortunately, Russian investors with an “oil and gas” mindset on market conditions are not particularly willing to invest in the development of commercial microelectronics, since initial share estimates are unusually high by Russian standards.

x86 architecture compatible

The developers initially considered ensuring effective binary compatibility with the dominant Intel x86 microprocessor architecture as a fundamental architectural requirement. It is implemented on the basis of hidden dynamic translation and its support in the Elbrus microprocessor hardware. Also, the defining properties of the new domestic architecture include developed hardware support for secure computing (modular programming), which significantly facilitates the work of programmers when creating large software complexes with limited execution times.