ELBRUS

RUSSIA'S SEYMOUR CRAY
Russian supercomputer developer Boris Babaian is running a private tech company with global aspirations.
By Nikki Goth Itoi
The Red Herring magazine
November 1998
In the United States, Seymour Cray is widely recognized for building the multimillion-dollar supercomputers that formed the technological foundation of the Cold War. He has a counterpart in the former Soviet Union: Boris Babaian.
After 40 years with the Russian government's Institute of Precision Mechanics and Computer Technology, where he led the development of the Elbrus supercomputers underpinning the Russian missile defense system and space mission control center, Mr. Babaian has turned his focus to private industry. As chairman and chief technologist of Elbrus International, he and the same team that developed three generations of supercomputers are providing contract services like software and microprocessor design to Western companies.
INDEPENDENCE DAY
Mr. Babaian is justifiably proud of his engineering team's work for the Russian government. "Beginning in the mid-'50s, our team designed the most powerful Soviet computers, many of which are still in use in our space program and defense applications," he says. What the Elbrus supercomputers lacked in chip density, he explains, was compensated for by superior logic design. That design is what he hopes to market in the United States and Europe within the next two years.
The economic changes that swept through Russia early this decade made it difficult for Mr. Babaian's team to continue working as a government entity, and in 1992 the team began the long process of privatization. That year it also made significant inroads into the U.S. IT market when it started providing technical support for outsourced projects and joint software development services *****n Microsystems, Compass Design Automation (now owned by Avanti), and several smaller companies. Officially launched in 1997, Elbrus International is Russia's largest computer development firm, with offices in Moscow, St. Petersberg, and Novosibirsk and more than 400 employees.
Mr. Babaian is hoping to bring two major innovations to market, based on U.S. patents he filed starting in 1985. The first is a binary compiler (software that translates a high-level programming language into machine language) that accommodates greater cross-platform compatibility without the typical compromise in speed. And the second is a new microprocessor architecture that supposedly improves on the logic design of the Explicitly Parallel Instruction Computing model being developed by Intel and other U.S. companies.
ELBRUS ON THE TABLE
In late 1996, Mr. Babaian connected with San Francisco-based Plantagenet Capital Management to raise private placement funding of $3 million. Plantagenet has enlisted PricewaterhouseCoopers to help Elbrus find strategic partners interested in licensing its technologies. "We're talking to everyone but Intel," says John Zappettini, a managing partner with Plantagenet. "And we've had audiences at the highest level with each company we've approached." A major element of the partnerships would be a second round of funding for Elbrus pegged at about $60 million.
The partnering process has been slow going because many microprocessor experts doubt the viability of Mr. Babaian's Java-like proposition: "Rewrite never, run efficiently anywhere." As of August one major European firm was investigating the company and several U.S. firms had expressed interest, but no deals had been closed.
But according to Mr. Zappettini, in ad***ion to the strength of Elbrus's team, the main selling point of this investment opportunity is Mr. Babaian's reputation as a computer scientist. He was awarded the two highest honors in the former Soviet Union: the USSR State Prize for Achievement in 1974 in the field of computer-aided design and the Lenin Prize in 1987 for the Elbrus-2 supercomputer.
Mr. Babaian's legacy for Russia will closely resemble Mr. Cray's for the United States: machines that were essential for solving their countries' greatest national challenges. But Mr. Babaian's transition into the private sector may earn him something further--the distinction of being one of post-Cold War Russia's first internationally minded technology entrepreneurs.


Microprocessor E2K
Unfortunately, because of our crisis, Russian science is in a collapse, most of researches are closed or being closed cause of lack of money. Some time ago we were at the first edge of science, now it is not true in the most cases. It's nice to hear, that in such a situation some groups are still working at high-technology projects and gain some significant results. Good example is Elbrus group.
The core of this company consists of such a titans of Soviet computer industry, as Boris Artashesovich Babaian, one of leading researchers of IPMCE (the Institute of Precise Mechanics and Computational Equipment). The most of soviet researchers, who was constructing the most powerful soviet computers, are still working at such projects. That's why their new project, Elbrus-2000 microprocessor (abbreviated as E2K) has made such a sensation. It should be the fastest microprocessor of its class at this time, and even Merced and Mc-Kinley, new powerful microprocessors, announced by Intel, seem to be three times slower!
But let's return to our history. In the middle of December 1998 there was a great date to our science - 50 years of Russian computers. In the end of 1950th researchers from IPMCE have built a first Russian computer M-40, then , in 1960th, 2 processor computer which was used in first Moscow missile defence system. It was the beginning. Our scientific progress was fast and Russian achievements were very profound. In the end of 70th our group of researchers, headed by S.A.Lebedev, B.A.Babaian, V.S.Burtsev, have built a supercomputer 'Elbrus-1', multiprocessor computer, which was the first Russian SMP (Symmetric Multiprocessor) computer. Then, in the beginning of 80th there was built 'Elbrus-2', further development of 'Elbrus-1'.
These machines were at the first place in the world by their calculation speed. 'Elbrus-1' was used by Soviet Ministry of Defence. 'Elbrus-2' consisted of 10 processors and was widely used in the same sphere - in the Space Flight Control Center, in various nuclear research institutes, in air defence systems. Actually, it's still used in some places, that shows it's high reliability. Of course, at this time it's not a supercomputer - at the modern range it's speed rather moderate. As to architecture of this machines, it was very original - 'Elbrus' computers, as a some 'Burroughs' models used a null-address command system - they used stack. This means that all data are considered to be at the top of stack structure and all operations were performed at the data at the top of stack. Because of that it was called 'El-Burroughs' amongst their researching group. For example, an ad***ion command was just '+' - with no operands, cause both arguments were taken from the top of stack, and the result was placed there instead of them. This is very convenient for translation of arithmetic expressions, because the translator usually transforms any expression to it's so-called Lukashevich form - bracketless postfix notation.
Another innovation, which also supported the high-level language translation was the tag mechanism. This means, that all memory units (as for their size, it was 64 bit) had special tag, which displayed the type of data stored there. This allowed to check the correctness of a program at the run-time. For example, it could be found out that program attempts to add floating number with a string and of course, error would be reported. So, the principally new architecture decisions highly supported the efficient high-level programming. Unfortunately, their new project, 'Elbrus-3', which was considered at that time even more powerful than fastest supercomputers in the world, was closed in 1994 because of lack of financial support. So, our researching group from IPMCE was very strong - their achievements on some directions had no equivalents in the world. A number of new technologies were developed: high speed machine arithmetics (1955), failure-proof systems of continuous functioning with total hardware control (1964), symmetrical multiprocessors with shared memory (1979), protected programming, based on hardware-supported type control (1979), technology of binary compilation (1986). It can be easily seen how strong is the potential of this group, how deep is their experience. Their first results were got so long ago, that now there are the best specialists in the world, for example, in binary compilation and fast ALUs.
I hope now all reasons of the great success of their new processor became clear. The Babaian's researching group applied most of their developments in this microprocessor, so it is a result, in fact, of several decade experience. In 'Microprocessor Report' magazine appeared an article with terrified heading ' THE RUSSIANS ARE COMING ', that nearly blasted western scientific world, and here are some materials out of it.
Well-known supercomputer architect Boris Babaian often called the Seymour Cray of Russia - disclosed a new processor that his company, Elbrus International, is developing.
The E2K processor uses EPIC (Explicitly Parallel Instruction Computing) techniques similar to those Intel and HP will use in their upcoming Merced. According to Babaian, however, the E2K will deliver higher performance with lower power and less silicon.
The Elbrus team has the great experience because the core of the team, as I said, has been together for over than 40 years, developing supercomputers for the Soviet Union's defense establishments. The company has developed computers based on superscalar, shared memory multiprocessing, and EPIC techniques, long before papers on those subjects appeared in the West. Lacking semiconductor technology and access to the funds available in capitalist nations, however, Elbrus supercomputers have never quite matched the performance of their Western counterparts. So it shall be for the E2K, unless the company can find a partner or other source of capital.
In a 0.18-micron six-layer-metal process, Babaian says, the E2K will run at 1.2 GHz and will require only 35 W of power and 126 square mm of silicon (with 256K of on-chip L2 cache). We protect that in a similar process Merced will operate at 800 MHz in 300 square mm of silicon at 60 W. Merced, however, is ahead of the E2K in development by at least a year. Even more amazing, Babaian claims the E2K processor will be x86 binary compatible and, after a few tweaks, IA-64 (Merced) compatible as well. To achieve this feat, Elbrus will rely on binary compilation assisted by emulation hooks in the processor. Babaian believes, however, that his company's binary-compilation technology is more advanced than any other on the planet.
Elbrus team members are understandably proud of their baby. It is likely that their enthusiasm, as for most fathers, has led to some inflated claims for the E2K. But the capability and credibility of the Elbrus team does not permit the E2K to be dismissed out of hand. There is little reason *****spect that the team is overstating its processor's capabilities by any more than other competent design teams tend to when still a year from tapeout. Even if Elbrus missed all its targets, by 20% the E2K would still be I GHz, 150 sq. mm and 45 W - an awesome chip by any standard.
Evidently, RISC, superscalar, multiprocessing, fast ALUs, high-speed CMOS circuits, capabilities, optimizing compilers, binary compilers, and perhaps even EPIC were not such new ideas after all. The E2K also demonstrates that not all good ideas in computer architecture have yet been absorbed into microprocessors.
Of course, Babaian seems very hopeful about his new creature, he claims E2K the fastest microprocessor in the world and probably it will be so. This processor, with his high speed, efficiency, compatibility with Intel platform, low price (as a rule our electronics is about ten times cheaper) claims to flood the world and be the most popular in the world. That's why western processor monopolies are scared.
But in fact, the situation is rather sad: Intel says, it will produce Merced, and is working at it - no doubt, that in certain time it will be done in hardware (if not already done). With Merced Intel hopes to became the only monopoly at processor market - all its details already patented. And for E2K, we claim it's much better, but we can't produce it - in fact, this processor doesn't exist yet. Even worse: we have not only a lack of financing, we have no techniques to make microprocessors at all. It looks as Intel shows new processor and claims it the best and Elbrus Group says 'Our processor is much better, but it doesn't exist'. And it's a problem. Babaian is searching for western partners, he needs $40 mln to build a plant and make processors there, in our country nobody needs this.
Some days ago Moscow mayor, Luzhkov, promised to help Babaian with money, and I hope our processor will be produced and will be the best un the world. It will take two or three years to make this processor, including it's design in details, an then Russia will be at the first places at microprocessor market. Who knows...
Kiên