IBM's Power6 Spotlights Big Blue's Processor ProwessIBM's Power6 Spotlights Big Blue's Processor Prowess

Did you know IBM was the first company to ship a dual-core processor? Most people don't; they incorrectly assume it was Intel or AMD. Wrong. IBM rolled out its Power 4 module, which contained two 64-bit, 1-GHz processor cores, in 2001. Yesterday, Big Blue's deep expertise in chip technology was on display when it launched the latest update to its multicore Power family.The new Power6--the Power5 was released in 2004--takes the clock speed up to 4.7 GHz and increases the size of the cache to 8 Mbytes. But my point is not to bore you with specs (which I find boring, too), nor to repeat the idiotic attempts of the IBM press release to make the chip "accessible" to everyday folk ("It can download the entire iTunes catalog in about 60 seconds!").

As well, I don't really care that the Power line--a relative of the PowerPC family, which was the Apple Macintosh processor for many years--is a multichip module rather than a unified dual-core processor packed onto a single slice of silicon. The latter is the so-called "native" dual-core approach, which AMD will implement with its upcoming Phenom architecture.

Intel CEO Paul Otellini deflected that issue--he's been queried on it, because Intel's quad-core processors place two dual cores side by side--by saying: "The initial ones are multichip, but so what?' You guys are misreading the market if you think people care what's in the package."

No, what I'm really trying to emphasize is that IBM's announcement of the Power6 is significant because there are so few chip companies around anymore which can afford to push the state of the art forward.

It's a little-discussed fact, but as someone who covered the chip industry at EE Times for a number of years, I can tell you that, even within the IT community, most people aren't aware of the daunting economics of chip manufacturing.

In years past, I recall Andy Bryant, Intel's CFO, taking the podium at Intel's Spring Analyst Meeting and saying that fabs (semiconductor manufacturing plants) cost upwards of $1 billion to build. This year, he said that launching a new processor product line costs $3 billion. (Of course there are lots of qualifications to these numbers, yada, yada, yada, but I'm trying to make a coarse-grain point.)

That's why companies that used to manufacture their own chips have been falling away like flies over the past decade. The few that still actually design their own processors--like Sun Microsystems--use the so-called "fabless" model, where they get another company (in Sun's case, Texas Instruments) to manufacture their chips.

Consider that Intel is currently building or upgrading eight (8!) fabs just to crank out its upcoming new generation of 45-nm chips. The only company which can match Intel's fab capacity is IBM. (In case you were wondering, AMD has far fewer fabs then Intel. AMD has a deal with IBM through which Big Blue can provide extra fab capacity to AMD.)

True, IBM's Power6 family will play only in the high-end server market, where it will compete with servers equipped with Sun's UltraSparc IV and T1, and Intel's Montecito (a dual-core Itanium 2). So IBM's not a player in the desktop processor market.

No matter. In some way that's very hard to explain, it seems to me that, thanks to IBM's Power6 prowess and its sheer manufacturing heft, Big Blue is all that stands between Intel and total dominance of the processor landscape.