While the false millennium is less than 50 days away, the era I refer to is the one where Intel has controlled the desktop with a single architecture. From the 8080 to the 8086 to the 386 and beyond, Intel has dominated market and mindshare. What are the tools that Intel used to build this position, and how will they fare after 2000?
The story of Intel's rise to power as a cpu manufacturer has been told before. They put together a powerhouse by melding a number of tools and techniques, building on their early lead and taking advantage of opportunities. Clearly they heeded General Patton's dictum to "Get there firstest with the mostest" and applied it to semiconductor manufacturing. Intel has consistently invested large sums in advanced fab plants to stay at the leading edge of technology, using the performance edge of the advanced fab to price its products high enough to repeat the cycle a few years later.
Intel also played hardball in the intellectual property arena to keep its competitors at a disadvantage. The combination of advanced technology, massive software support of the x86 architecture, high prices and restrictive licensing has kept Intel in a dominant position since the early 1980's. This is not to belittle their technical prowess. Intel has pushed the x86 to performance levels that would have sounded impossible ten years ago, while competitors have trailed Intel until very recently.
The big questions is: Can they keep it up? The answer to this involves more than just Intel and its x86 competitors. Change is the only constant in the IT industry, and now even the ground under Intel's long reign is starting to move. I'll look at the changes in technology, competitive processors, software and industry alliances that suggest what the next ten years will bring.
The IT industry has quietly become pervasive in our lives. Everyone is aware of the home computer revolution, but few are knowledgeable about the infrastructure that has grown to support it. When the industry passed 100 billion dollars per year in revenue, it had matured to where no one player could cover all the bases of any segment of the business. Specialization changed from an option to a necessity, and companies entered the period of coopetition. Fab technology has not been a secret, but the fab costs have escalated to two billion dollars for a SOA 0.18 micron fab line, like the one just completed by AMD in Dresden, Germany. As a result, a lot of semiconductor companies are now 'fabless', they build chips on lines owned by others either in partnership or by contract.
The broadening of the market for all kinds of processors, especially the embedded processor market, which is ten times the size of the desktop market, has spread the skills and knowledge of how to design and build fast processors to many companies around the world. The overall size of the market has enabled large investments in all aspects of the industry. This has led to the current very competitive industry where the economics of a processor is far more important than its technical value. While the industry gets bigger, its products get smaller for both performance and economic reasons.
All of these factors combine to compete indirectly with Intel. Because there are many specialized embedded processors See "Embedded Processor Watch". Intel's earlier dominance of embedded processors has become a very competitive market. With production runs of millions, it has become economic to custom design processors for specific embedded tasks. Intel's earlier general purpose embedded designs had the advantage where volumes were small, but with large volumes and many skilled competitors, Intel is just another, albeit large, competitor.
If the processor is the engine of your computer system, then software is its fuel. Turn on a computer without software and what you have is an electric doorstop. Having written code since the early 60s, I can report firsthand that the software industry has not advanced very far, and in some respects, has regressed. The details of this will become another column [Why Software Is Hard], but there are some recent developments that will affect the future. Microsoft's monopoly in desktop software and their aggressive stance against any competitor who might threaten that monopoly has had one effect not anticipated: Their competitors realized they had better cooperate before they became extinct. Cross platform development, almost expiring in the early 90s, has had a comeback in the last few years.
The ultimate cross platform development tool is Java, invented by Sun Microsystems, and adopted by everyone with one notable exception - Microsoft. In typical form, Microsoft has attempted to dilute the value of Java by tying its version to Microsoft APIs in an non compatible way. While this issue is now being resolved in the courts, Microsoft has had some success by slowing the adoption of Java. In addition to Java, Linux has become a ten year overnight wonder and brought broad credibility to open software development and a powerful cross platform OS.
IBM has adopted cross platform development, joined the Linux supporters and taken Java to new performance levels including on Linux. More importantly, they have developed native code Java compilers for Win32, OS/2 and AIX and incorporated those compilers into The VisualAge for Java product. The Entry edition of VA for Java is available as a free download from IBM's Visual Age Developer's Domain (VADD). We are almost at the point where software developers can have their cake and compile it too.
Cross platform development increases the value of competing processors. The non x86 processors have aggressively improved their chips. HP, MIPS, Sun and Compaq's Alpha are all close to releasing new high performance versions of their chips. See "Microprocessor Reports". All these chips are upwards compatible with the code for prior chips because of, again, economics. It simply is not affordable to force the users to recompile everything for a new processor.
Linux, while not currently considered a high end server product, will bring two critical benefits to these competing processors - volume and software. For almost two decades Intel has had the big advantage of high volume manufacturing and wide software support. The recent developments of Linux and Open Software will narrow that advantage significantly.
Intel, in a gutsy move, has copied what IBM did in the early 60s with the 360 system. They are betting the company's future on a new architecture known as IA-64, and pushing the envelope by moving instruction optimization into the software rather than doing it in the processor. Even though IA-64 will execute IA-32 [aka x86] instructions at a good speed, it appears to me that Intel is facing a real risk of leaving the x86 business to its competitors. Developing a new architecture is only the beginning.
Because the IA-64 processor pipeline is not internally forced to wait for data, that job now falls on the compilers and it is not a trivial one. Intel has been developing new compilers, and working with other compiler vendors. Next, Intel must design a real chip, code name Merced, that actually works. These hurdles have been passed. For details on Merced, see the Wolfe's Den columns on Byte.com.
Now comes the hard part. They must manufacture and deliver this chip soon enough, profitably enough and in volume to meet or surpass the performance of the other new chips coming in 2000. This has not yet been done, but given Intel's resources and commitment,they will probably succeed. Even if they do, Intel's continuing dominance is not assured. For they have potentially left AMD a big opening by de-emphasizing the x86 architecture. In addition, AMD has matched Intel's x86 line in both performance and clock speed with AMD's new Athalon chip. AMD's use of Alpha's slot A bus gives them a performance edge over Intel's slot one.
Finally, AMD's new fab in Germany expects to deliver Athalon chips in volume at up to 1GHz speeds, starting in Q2 2000. And if AMD's x86-64 catches on, the 64 bit x86 compatible extensions that are planned, then Intel no longer has an uncontested line without competitive pricing pressure. See AMD. Just when Intel is taking its biggest risk, AMD and other chip makers are bringing a new set of guns to bear on Intel's market.
Even more significant, the environment Intel is facing is much tougher than what IBM faced in the 60s. Back then, IBM was the biggest of the computer companies. The industry was known as IBM and the Seven Dwarves, since the combined size of the competitors was less than IBM. The tougher environment now stems from the larger and more competitive infrastructure, and the much faster rate of technology development. IBM did have teething problems with the 360 line, both hardware and software.
We have benefited from the insightful book "The Mythical Man Month" by the manager of the OS/360 project, Frederick Brooks, who tells of software development lessons learned the hard way during OS/360 development. But IBM had lots of time to correct the 360 problems, and went on to dominate the industry even more as 360 was followed by 370 and today, the 390 series. Intel will not have the luxury of that much time to correct problems. Even if Intel avoids major problems with IA-64, their competitors will benefit from Linux, Java and cross platform development.
Perhaps just as important, customers who have seen what the Microsoft monopoly did are less likely to make the mistake of giving one processor vendor all of their business. The end of the era is the end of the dominance of one processor architecture. It is not the end of Intel, nor does it even presage their dramatic shrinkage. Instead, Intel will become something like IBM is today, very large, dominant in some areas, but no longer the company around which all software development revolves. The strong worldwide infrastructure, intense competition from new and old companies alike, and broadening software development will gradually whittle away at Intel's dominance.
What will this mean for the customer? More
choice, lower prices and ultimately, better
systems. Competition, long restrained in
both software and hardware by the two behemoths,
Microsoft and Intel, will increase. As television
did when cable entered competition with the
three broadcast systems, choices will multiply.
Today we have five major
broadcast systems, independents, PBS and
hundreds of cable channels. Early in the
next decade we will see the new options on
the desktop. Linux is already here, Java
is reaching maturity and even new operating
systems like BeOS are attracting new users.
This is only the beginning.
[30]
All content on this site is Copyright 2001 by Bill Nicholls