Processor Technology: What's In It For You?

The steady increase in the power of server processors shows no sign of abating. One indicator was the recent announcement that Intel is now in volume production with desktop processors made on 300-mm wafers and with 90-nm design rules. The technology, when in development, was called Prescott. The technology will soon appear in Xeon processors.

So what's the big deal with all this new stuff -- 90 nm, 300 mm -- and what does it mean? It means better performance at pretty much the same price in your servers.

This is a familiar phenomenon for people in the computer business. As time has gone by, computer power has increased almost beyond comprehension. All this has been possible because of something called Moore's Law. Originally stated, the law said that the number of devices on a given area of silicon would double every 18 months. Today, that is generally restated as processing power doubles every 18 months. And, says Hemant Dhulla, director of marketing for volume platforms at Intel Corp., there does not seem to be any roadblock in the foreseeable future that would stop that progression.

Right now, the current processor that Intel is selling has 130-nm design rules. The company's roadmap projects that it will begin volume manufacturing of a new Xeon processor, code-named Nocona, with 90-nm design rules and on 300-mm wafers. Such shrinkage in feature size, "plays a continued role in increasing performance," says Dhulla, "for example, in increasing the clock frequency of the processor."

That's not all. With reduced feature size comes generally reduced power consumption for a given device. That translates, roughly, to more processing power for about the same electrical power consumption. So although more processing power comes into the data center with new servers, there may not be a requirement for more electrical power. And, probably equally important, air conditioning infrastructure might not need upgrading. But you'll have a server that can handle more data and run faster than the one being replaced, and can probably fit into a smaller space.So the bottom line is cost, that is, more power for little increase in price. Another move in that direction comes from the 300-mm wafers. That's a silicon wafer of about 12 inches in diameter. Compared with the older, 8-in. wafers, there's room for a lot more devices on the larger wafer. So the increased number of die per wafer, and a reduction of overhead (for test devices and like) make up for the cost of the larger wafer. Dhulla says it takes multiple 8-in. wafers to get the same number of die you can get from one 300-mm wafer. The cost savings are clear.

Next quarter Nocona will come out with a 3.6-GHz clock rate and 1 MB of onboard cache. It will also have PCI Express, the next generation of PCI, on chip, will support DDR2 memory technology and will have "strong reliability, availability and serviceability capabilities built in," Dhulla says.

Moore's Law seems to be alive and well, and it's giving you more power every year.

David Gabel, an electrical engineer, has been testing and writing about computers for more than 25 years.