Would you believe, a teaser ad might say, in a microprocessor that will allegedly outperform Intels Itanium? That such a processor was being designed by Russian scientists held over from the Cold War? That the designers hadnt really been heard from since the late 1990s?
Well, Elbrus MCST (Moscow Center of Sparc Technology) could still turn out to be a mountain of hype. But the great white whale of microprocessors, the Elbrus E2K, resurfaced recently amid claims that the project is still alive.
Evgeny Babayan, the chief executive of Russias Elbrus MCST, also known as ZAO MCST, said the E2K project is still underway and the company hopes to have a prototype out by the end of the year.
“At the end of year will be first prototype silicon,” Babayan wrote in an email to ExtremeTech. “We dont want many noise (sic) before it.”
According to the companys own roadmap, that chip, dubbed “Elbrus”, will likely achieve just 400-MHz on a 0.18-micron process, but a successor, the “E2K”, could scale to 1.2-GHz by 2004 if everything goes as planned. In a 0.1-micron process it will have a clock frequency of 3 GHz and deliver scores of 500 and 1200 using the SPECint95and SPECfp95 benchmarks, respectively, the company claims.
While early cores were clones of Sun Microsystems Sparc architecture, the company is branching out with its own instruction set. Elbrus says it has developed a “binary compliation” technology, similar to Transmetas “code-morphing” technology, that will make it compatible with the X86 instruction set.
While that clock speed may seem slow, Elbrus claim to fame is a hyperparallel architecture that apparently influenced the design of the Transmeta Crusoe chip, among others. By performing more operations in parallel, the argument goes, more work can be done through the use of a slower clock speed, thus consuming less power.
“Today state-of-the-art American computers perform up to six operations each clock versus twenty-four operations performed on our computers,” said Boris Babayan, chief technological officer for Elbrus, in the translation of an interview given to Kommersant, a Russian publication, in April. “The results achieved by the Russian microprocessor school are not within the reaching distance of western computer architects yet.”
The company, then known solely as the Moscow Center of Sparc Technology, was formed on April 8, 1992, although the MCST center existed during the Cold War building supercomputers for the Soviet Union. According to the company, funding was stopped after the collapse of the U.S.S.R. Still, the company employs a number of engineers on various computer-related projects, including trying to develop a successor to the Elbrus supercomputer systems the company developed throughout the Cold War. Elbrus microprocessor design team employs 100 engineers, Evgeny Babayan said, although funding them is a continual problem.
“Its a tough question whether we will be able to turn around this situation to suit our own ends,” Boris Babayan said in the Kommersant interview. “Yes, we can offer technical and conceptual solutions the Americans have not hit upon yet. But we lack money. Even if the government allocates $100 million to design new architectures, it will be unable to find (an) extra billion dollars for the marketing purposes. We are seeking different ways of cooperation with western companies as probably the only chance for the Russian microprocessor science to gain world recognition.”
Elbrus counts five U.S. firms as its partners: design house Avant!, Cisco Systems, Infineon Technologies, Sun Microsystems, and Transmeta Corp. Officials at both Sun and Transmeta say the design team is real, that the Elbrus design is viable, but that the companys future rides on its ability to raise capital.
“I have a lot of background with those guys—we worked together when I was at Sun,” said Dave Ditzel, vice chairman and chief technical officer at Transmeta Corp., Santa Clara, Calif. “Theres a funny story—at some point the mayor of Moscow wanted to pay a fee so they could use Sparc in the (companys) name.”
“Id call a lot of them friends of mine,” Ditzel said.
Likewise, officials at Sun Microsystems Inc., Palo Alto, Calif., confirmed that Elbrus had worked with Sun on developing some undisclosed software.
The problem for others is that so far, the company hasnt produced any silicon to demonstrate to American analysts, who retain a healthy dose of skepticism. “Until they have anything to report, I dont think that theres a story here,” said Peter Glaskowsky, editor-in-chief of The Microprocessor Report, published by Cahners MicroDesign Resources, San Jose, Calif.
The Microprocessor Report has previously published an article on Elbrus, although the company has never attended MDRs Microprocessor Forum, a top showcase for new designs. Keith Diefendorff, a former MDR analyst, the author of the Reports article, “The Russians Are Coming,” and now vice-president of product strategy for embedded processor vendor MIPS Technologies Inc., Mountain View, Calif., said he has no new information about how Elbrus is faring.
“Ive talked to Boris a few times since then, mostly about business, funding, et cetera,” Diefendorff said. “I havent talked with him for a while (and) dont have any current insight into the status of their part.”
Two Elbrus engineers also surfaced during a Avant! Users Resource Organization for Real Applications (AURORA) conference in February, where they presented a paper on “cell-based pass transistor logic”, used in developing high-speed, low-power circuits on a 0.18-micron standard manufacturing process from Taiwan Semiconductor Manufacturing Co. (TSMC).
“I certainly wouldnt dismiss them out of hand,” said Dean McCarron, an analyst with Mercury Research Corp., Cave Creek, Ariz. “They do have a track record in this technology.”
McCarron added that funding was certain to be a critical factor in the companys success, but the fact that the company is relatively unknown shouldnt be seen as a drawback. “History proves that interesting results come from small design teams,” he said. “Like the Athlon.”