When ‘Intel Inside’ no longer matters

These days, the business press paints a rosy picture of Intel: Pentium machines are zipping out of dealer stock, P6 is looking good, and P7 (Merced) is enjoying a full-court press from HP and Intel PR flacks. However, just as the Internet caused some rethinking of strategy at Microsoft, it should be giving Intel pause as well. It’s time for Intel to explain how it’s going to build a chip that will drive the Internet client of our wildest dreams. Internet clients — PCs, Unix workstations, or cheap terminal devices — will be differentiated on the basis of user-visible, multimedia-rich features. How well does an Internet client display a video clip? What audio quality can it reproduce? Can it run through a complex animation, entertaining the kids cruising through the sixth level of toxic sludge and serving the corporate risk manager looking at a landscape of potential currency translations? Will it suffice as a low-cost video conferencing terminal, so I can reduce the number of appliances I have in my corporate and home offices?

Raw CPU power might help with some of those multimedia features; after all, everything in the list above can be reduced to number crunching when you take them down to the lowest common denominator. The problem is that integer or floating-point horsepower won’t suffice and may not even matter in two or three product cycles. Here’s why. First, the number of users constrained by slow desktops is trending toward zero. Jon Forrest published this observation in the February issue of Upside (http://upside.master.com/print/feb96/sb9602.html), and it’s not one to be taken lightly. Second, graphics hardware is getting faster, cheaper, and better at a far more rapid pace than are general-purpose processors. Take a look at Sun’s Creator3D frame buffer for an example of this order-of-magnitude leap.

The current goals of the chip designing process — cramming more transistors running at a higher speed into an ever-decreasing piece of silicon real estate — smacks of machismo. It’s not a trivial process, as IBM’s on-again, off-again PowerPC 620 effort has demonstrated. It’s even more difficult when the silicon foundry has multiple bakers at the controls — HP and Intel in the case of the P7. But there’s a more basic issue here: You don’t need the whole thing to go faster, you just want the interesting parts accelerated. That is something Intel did remarkably well not too many years ago. The i860 and i960 processors found their way into the hearts of many graphics boards. Why wasn’t any of that technology mainstreamed into Pentium, P6, or P7? (For another example of a company with incredible point technology that couldn’t bring it into the mainstream, look no further than Apple and properties such as QuickTime VR.)

There’s another part to the story, of course (we haven’t mentioned Java yet). Intel is the Trigger to the Roy Rogers of the Northwest. “Intel Inside” matters when you’re in a Windows environment. But what about when you’re in a Java environment? “Intel Inside” is replaced with “Java Compatible,” an acclamation of the chip, operating system, and programming environment working in concert. Java quite simply makes CPU branding irrelevant. The CPU could use 18-bit words and be lacking an integer add instruction (yes, processors like that once existed), but if it makes for a good Java platform, it will succeed.

What makes a Java platform “good”? All of the graphics goodies outlined above, plus low power consumption, native support for some language features like threads, and a simple compiler model. To equal the performance of native C or C++ code, Java bytecodes will have to be fed through a just-in-time compiler that produces native code after all of the security and language sanity checks have been passed. The easier it is to generate efficient, tight code for the chip, the better those just-in-time compilers will be. Sorry, but bizarre register usage and a CISC instruction set drain the brains of even the Russian compiler mavens. Simplicity wins; it makes for a programmer-friendly compiler and a faster run-time environment.

What else is out there besides the bevy of brains in Unix workstations? Oracle is using the DEC StrongARM chip in its Network Computer. Sun has announced a line of Java-centric processors, ranging from the low-power, low-cost picoJava to the marriage of graphics and Java in UltraJava. Today, Intel enjoys a tremendous volume advantage, allowing it to drive prices down on its high-end CPUs. However, StrongARM and the proposed picoJava chips don’t play in the 00+ high-end game — they are in the 5 range required for embedded systems. SGI is learning something about the embedded market through its sales of MIPS chips to the game makers, and its VRML work is driving the demand for high-performance, 3-D graphics capabilities on the Web browsing desktop. Microunity continues to exalt its media processor, fascinating even some of the doubters on Wall Street. Toss in UltraSPARC and Alpha on the high end, and there are a solid half dozen potential Intel competitors, all with some potential strong suits.

Does this mean Intel’s franchise is on the verge of evaporation in a cloud of Java steam? Absolutely not. Intel is still strong, still producing an amazing volume of chips, and still selling for X times earnings. The market for processors is still showing signs of vigorous growth. It just looks like the growth will be spread over several players, particularly in the home market.