When one talks of Russian processors, usually Elbrus and Baikal spring to mind. But Russia’s processor industry is far bigger. One of the developers whose solutions are rarely attracting the attention of the general public is NIISI, a division of the National Research Center “Kurchatov Institute.
The complete name of the organization is so long that it’s really hard to pronounce it without preparation. Federal State Budgetary Institution “Federal Research Center—Research Institute for System Studies of the National Research Center ‘Kurchatov Institute.'” It is easier to use the initials NIISI of NRC “Kurchatov Institute.”
Here is where the Russian KOMDIV family of microprocessors is developed—superscalar RISC processors with their own instruction set and original 32-bit and 64-bit computing cores.
The big difference between these and the CPUs you know is what they’re meant for. They weren’t created as an attempt to create a mass-market CPU for home computers or smart phones. Much of the effort is geared toward safe and trusted computing systems, including systems meant to process restricted-access information.
KOMDIV – Three Generations
Over the course of its work, NIISI has created three generations of KOMDIV microprocessors and BAGET computing systems based on them.
The KOMDIV architecture itself is an in-house development. At the same time, the designers also implemented hardware-level translation of instructions from MIPS-based CPUs. This permits the usage of some current applications and makes the migrations between different computing systems easier.
Another key element is that the developers say their complicated functional blocks are their own designs too. NIISI says this makes it very easy to move projects across manufacturing locations with similar process technology. This strategy helps to keep the hardware platform under control during the whole life cycle of the microprocessor at a time.
Thus, this is not only a CPU core but also a full-fledged domestic development, which includes computational, peripheral, and system-level components.
65 Nanometers, but a Serious Architecture
The 1890VM8 processor family is a nice example of this. The chips have been in volume production since 2016, based on a 65-nanometer technology.
The family has the 1890VM8Ya, 1890VM8AYa, and 1890VM8BYa variants. The CPUs are built using a 65 nm CMOS process, are supplied in a flip-chip BGA package, and have 1,294 pins.
65 nm is a somewhat old process node by today’s standards. But it would be wrong to compare such a processor directly to the CPUs of today that are built for PCs or cellphones. It’s not simply about transistor density and raw performance for specialized computing systems. Having predictable operation, the capacity to provide the essential interfaces, the long-term manufacturing capabilities, and the control over the architecture are all equally vital.
The 1890VM8 is a dual-core 64-bit superscalar microprocessor built on the KOMDIV64 architecture. The single die includes CPU cores, system and peripheral controllers, Level 2 cache, and specialized digital signal-processing blocks.
It can also be run as a 32-bit CPU, with associated addressing modes.
Seven-Stage Pipeline, Two Instructions per Cycle
Each core consists of a seven-stage, superscalar pipeline. The CPU features instruction prefetch, dynamic branch prediction, and speculative execution.
The core can fetch up to four instructions every clock cycle and execute up to two instructions per clock cycle.
Each core has its own Level 1 cache. The instruction cache has a size of 32 KB, and the data cache is 16 KB. The shared Level 2 cache is 512 KB in size, divided into 4 pieces. The internal bus of the processor is 128-bit wide.
The memory management system provides for conversion of 32-bit and 64-bit virtual addresses into 36-bit physical addresses. For this reason, the CPU uses an associative translation buffer (ATB) or jTLB and separate micro-TLBs for instructions and data.
Digital Signal Processing and Vector Computing
The processor includes two coprocessors to accelerate computations.
First is the floating-point operations. It enables calculations in single and double precision, as well as operations with pairs of single-precision values.
The second is a specialized vector coprocessor mostly targeted at linear algebra and digital signal-processing activities.
It has its own register file, including 64 128-bit registers. It enables operations for real and complex numbers in single and double precision.
Thus, the CPU was made in an old process technology that is now recognized as such, but the processor architecture was developed not only for regular program execution but also for specific computational demands.
A Large Number of Built-In Interfaces
The 1890VM8’s strengths include a wide array of integrated controllers.
It has two memory controllers for system memory, one for DDR2 and the other for DDR3, running at 400 and 800 MHz, respectively. Two RapidIO controllers and an integrated RapidIO switch are offered for creating multiprocessor systems.
The chip also has PCI controller with 33/66 MHz, two RS232 serial port controllers, two Ethernet controllers with 10, 100, and 1,000 Mbps, SATA 3.0 with two channels, and a two-channel USB 2.0 host controller.
Also in the package are SPI and I2C interfaces, an interrupt controller, five timers, and EJTAG in-circuit debug capabilities.
At the same time the indicated power consumption is only about 2 watts.
These Processors Even Power Tablets and All-in-One PCs
The fascinating thing is that KOMDIV was not limited to laboratory development boards or very specialized computing platforms.
The 1890VM8Ya CPU was used to create devices of the BAGET family, including a tablet and an all-in-one PC.
The all-in-one uses a mini-ITX motherboard. Four USB ports were used to link a video camera, wireless keyboard and mouse, and other auxiliary equipment. The motherboard also supported two SATA storage devices and a monitor, and documentation stated it could handle resolutions up to 2K × 4K.
In other specialized equipment, like routers and server systems, the same processor—the 1890VM8Ya—was also applied. Publicly available information on the specific configurations of these systems is somewhat scarce, but references and images of such items indicate that the processor was employed in more than experimental applications.
Why Does a 65 nm Processor Still Matter?
Today, a 65 nm processor cannot compete with mass-market modern chips in performance, energy efficiency, or transistor density. But the KOMDIV project was designed for a different set of needs.
Here, the crucial features are a proprietary architecture, control over the underlying hardware platform, capability of adapting the processor to unique systems, availability of necessary interfaces, and potential of maintenance of the designed solution over a long period of time.
That is why the tale of the 1890VM8 is mainly interesting as an illustration that the processor industry of Russia is far wider than just a few well-known manufacturers.
Behind Elbrus and Baikal there’s a whole layer of specialized Russian processor development that rarely gets much airplay in the general media. KOMDIV is an example of this hidden layer. These processors were not marketed as general-purpose consumer devices to compete with existing desktop CPUs, but they ended up in home computers, routers, servers, all-in-one computers, and even tablets.
That is why the 65nm 1890VM8 is noteworthy not as a technological record-holder but as a demonstration of the presence of a school of Russian processor research that developed in parallel with the more well-known processor initiatives of the country.