DEC CHIPS IN WITH ALPHA
Recently , at DEC world in Boston, Digital Equipment Corp.(DEC), Maynard, Mass., demonstrated its reduced instruction set computing (RISC) design for a chip embedded computer architecture. DEC was able to show that design, called Alpha, clearly outperforms the latest available workstations competition. But your first question might be, why has Digital gone into the silicon chip business?
The answer is, because DEC knows proprietary systems are a vanishing breed. Digital will sell Alpha at all levels of integration - chip, board, and system - to other companies and to OEMs. DEC's strategy is all about very powerful workstations networked into specialized servers. As Richard L. Sites said on the subject of Alpha in the August issue of Byte Magazine, "The Microprocessor has had as much impact on the economics of computing as it has had on the performance of computers. As a result, any new CPU [Central Processing Unit] is likely to be conceived as a single-chip design." Alpha both defines the workstation and is DEC's entree into open systems world.
Alpha architecture is compatible with existing complex instruction set computers (CISC) and with DEC's VAX architecture is getting long in the tooth. Continuing to support the VMS legacy, while looking ahead to RISC in Windows NT and UNIX, reassures us old guys that we're still loved and positions DEC in the workstation marketplace of the future.
Technically, the chip is a whiz. It demonstrated clock speed at DEC world was 150 megahertz (MHz). Digital says it's the world's fastest microprocessor. Before very long, speeds of 200 MHz, and eventually 400 MHz, will be available. Very few applications today approach the limits of 32-bit addressing, so Alpha's 64-bit processing capabilities back up DEC's statement that this is an architecture for the next 25 years.
On the other hand, Alpha may not be the ideal chip for imbedded and real-time applications. For example, blazing speed in floating-point manipulation is not prerequisites for real-time performance. Higher priority is the representation, in integer fashion, of the large numbers generated by process control or other applications. Although floating-point overflow is handled well by Alpha, a rational approach to integer overflow and recovery is needed.
The chip's caches are adequate. But because larger caches are needed for today's control systems, 8Kbytes ain't enough for the on-chip data and instruction caches. My guys would like to see 64k for the caches. Because we "don't need" floating point, we'd prefer to have larger caches and treat floating point as an off-chip problem.
Alpha is important even if it isn't ideal for real-time and embedded applications. After all, real-time use constitutes only 10 to 15 percent of the market. Overall, technically, the chip meets the important benchmarks. But simply meeting benchmarks is not the name of game here.
DEC's change in focus from computer systems to a chip-embedded architecture is serious business. Even for a multibillion dollar company, competing in the processor chip market can tie up a lot of resources. Moving away from what has historically been the company's focus can be risky. But if the market accepts the chip, it will be a big win for the guys in Maynard.
Digital faces both technology and market challenge. Chip technology has so far gained by about a factor of ten every five years. If that trend holds, in 1997 Alpha would run at 1,000 to 2,000 MHz. In 25 years, Alpha's power would have to be 100,000 times today's performance.
In retail sales, the three most important factors are location, location, location. But for the semiconductor business it's timing, timing, timing. Even if you introduce a product twice as good as anyone else's you only have a little time to exploit your advantage. The competition will immediately design to technologically "leapfrog," based on what you've done, until they archive parity. And if there's a delay between announcement of a prototype and manufacture of production quantities, the "exploitation period" is commensurably diminished.
DEC has entered a tough race against experienced horses. Alpha is a good way of getting out of the gate. But is it too little, too late? I agree that it's an architecture with a 25-year life span. But 10 years have already gone by.
Perhaps what alpha does for DEC is similar to what sports cars like the Viper or the Corvette do for an automotive company. It demonstrates technical prowess, helps sell product, and gives engineers something to sink their teeth into. DEC's effort are to be supported. What's good for the computer industry is good for all of us. Hurray for Digital Equipment Corp. and best of luck.
As appeared in Manufacturing Systems Magazine October 1992 Page 40
References - Table of Contents
|R. Morley Incorporated
614 Nashua Street, Suite 56
Milford, NH 03055-4992 USA
Tel: 603-878-4365 FAX: 603-878-4385