Visit Your Local PBS Station PBS Home PBS Home Programs A-Z TV Schedules Watch Video Donate Shop PBS Search PBS
I, Cringely - The Survival of the Nerdiest with Robert X. Cringely
Search I,Cringely:

The Pulpit
The Pulpit

<< [ Curtain Call ]   |  Go to the Back of the Bus  |   [ When Is 64 Not Two Times 32? ] >>

Weekly Column

Go to the Back of the Bus: Intel's Plan to Crush Competitors by Making More and More of Your Next PC

Status: [CLOSED]
By Robert X. Cringely
bob@cringely.com

Some of my better columns come from reader suggestions, and this one is no exception. A reader suggested I look into AMD's HyperTransport data bus technology, thinking that it could be the technical advantage that finally cuts Intel down to size. I doubt that this will be the case, for reasons you'll shortly read. But that doesn't mean that Intel isn't facing some real problems in the marketplace. Intel has troubles, and most of them are of its own making.

But first the data bus story. Back in 1975 when Ed Roberts and his MITS engineers were laying-out the circuit boards for the Altair 8800, they decided to put most components on circuit cards that could be plugged into a motherboard just like we do today. The specification for those card connectors, the order in which the connectors were lined-up and the signal standards used for communication defined what was the first personal computer data bus. They called it the "Altair Bus," but when it was copied by other manufacturers, they changed the name to "S-100 bus." They also changed a few of the wires, but then so did MITS in later versions of the Altair.

The S-100 bus ran at one megahertz, and eventually reached a top speed of eight megahertz with the Z-80H processor. Compare this to today's top PC bus speed of 266 megahertz. The S-100 bus was eight bits wide, while that 266 MHz wonder is 32-bits wide, which means today's bus is more than a thousand times faster than Ed Roberts'.

The S-100 bus competed with the Apple II bus devised by Steve Wozniak. In both cases, the traces were lined up the way they were just because that's the way the designer preferred. There were no deep thoughts given to speed, manufacturability, or to a migration path to faster speeds. Those things required a new bus.

The IBM PC brought its own data bus, which came to be known as the Industry Standard Architecture (ISA). The ISA bus was so successful — generally in the hands of manufacturers other than IBM — that Big Blue decided to take back control by moving the world to the OS/2 operating system and the MicroChannel bus, both of which IBM controlled much more strictly than they had MS-DOS and the old ISA bus.

You may have noticed that nobody, including IBM, still makes MicroChannel computers, so that strategy failed. When IBM adopted the MicroChannel bus, the rest of the industry came up with an alternate standard called the Extended Industry Standard Architecture (EISA), which was faster and wider than ISA and almost as good as the MicroChannel, and didn't require royalty payments to IBM. EISA lacked the MicroChannel's ability to support so-called "bus master" cards — cards that had the right to demand access to the data path rather like the way authorities close the highways when President Bush's motorcade comes to town. But bus mastering was a marketing gimmick more than anything, since it was a couple years before IBM or anyone else actually built a bus mastering card.

EISA had real technical limitations and MicroChannel was probably better, but with only IBM making MicroChannel computers, it was more expensive, and so doomed to fail.

Eventually, computers came to have two buses — an ISA bus for backward compatibility and a new, faster bus. There were, for a time, two competitors for this faster bus — the VESA LocalBus and the Peripheral Component Interconnect (PCI) bus, with PCI becoming the eventual winner and surviving until today. That 32-bit wide 266 MHz bus in your new gaming PC is the PCI bus all grown up.

For Macintosh fanatics, I need to point out that Apple went with another standard called NuBus beginning with the Mac II. NuBus was already an industry standard used in other types of computers, but Apple's version was actually incompatible with the full standard. Typical. Eventually, Apple also migrated to the PCI bus, which is why there are some — but not many — cards that will operate in either a PC or a Mac.

This sordid bus history brings us up to the present day, when everyone acknowledges that the PCI bus has reached the end of its development cycle and something faster is required. The current PCI-X standard is expected to last for five more years, after which we'll need something all new. AMD and a bunch of other companies think the next standard ought to be HyperTransport, which is VASTLY faster at more than 12 gigaBYTES per second, and is backward compatible with PCI. Others, including Intel, think other solutions are just as good or better. Intel is pushing its technology called Arapahoe, which probably isn't quite as good as HyperTransport, but (and here's the important part) WHO CARES? Both Arapahoe and HyperTransport will serve us well for a decade or two, and the most likely winner is a combination of the two. That's why AMD voted to adopt Intel's Arapahoe as part of the PCI standard.

The simple fact is choosing one bus standard or another has never been the major reason for a computer or computer company to be successful. Choosing wrong can lead to failure, as with the MicroChannel, but choosing right doesn't mean success.HyperTransport is interesting, but it is really no threat to Intel. What is the biggest threat to Intel?

Intel.

Having spent years telling us that the faster the clock rate, the more powerful the processor, Intel is about to change its tune for the forthcoming Banias processor. This chip will offer Pentium 4-class performance at lower clock frequencies. How will Intel explain to its customers that clock speed is NOT the best indicator of performance after so many years of pretending it is? This should be interesting.

Intel says the Banias "platform" includes wireless networking features, but this is both misleading, and an attack on competitors in the chipset market (the market for support chips that work with the main processor).Banias itself doesn't support wireless networks. That requires the Banias processor, theBanias-specific Odem and Montara-GM chip sets, and the Calexico WLAN chipset, all of which come from Intel. Motherboard makers often prefer to use their own support chips or those from other companies that offer better deals than Intel. By marketing Banias in this manner, Intel is effectively shutting-out those other manufacturers. But the impact of this move goes beyond support chips. If Intel can mandate that only Intel chipsets be used in motherboards, it will have an unassailable price advantage in the motherboard business, forcing some competitors out of the business entirely.

The question that remains unanswered is if Dell wants to use a different networking solution for its Banias-based computers, will Intel let them?Right now, it looks like Intel will not. I guess having survived an anti-trust investigation by the Feds, Intel now feels it can afford to crush a few opponents — opponents whose past efforts led to the sale of a lot of Intel processors.

But wait, there's more! Intel's new Pentium 4 offers Hyper-Threading technology, and there's a new Pentium logo with "HT" to indicate this feature. The logo, however, is not available to OEMs unless they use an Intel chip set, even though some third-party chip sets do, or will, fully support Hyper-Threading. This restriction is likely to artificially diminish the value of systems based on these third-party chipsets.

It all comes down to earnings growth. In order to keep its earnings growing in a market that is no longer enjoying 20 to 30 percent annual revenue growth, Intel has to make sure it makes more and more of the next computer you buy. The ultimate goal is to make sure that computer has an Intel processor with Intel support chips on an Intel motherboard. Then it is only a short step to having Intel make the whole computer, which they will gladly do. Maybe it will work, but this strategy is going to lead to interesting responses from Intel competitors and customers alike. Fortunately, such responses can only be good for the consumers, for you and me.



Comments from the Tribe

Status: [CLOSED] read all comments (0)