BusinessWeek Logo
In Depth March 20, 2008, 5:00PM EST

It's Too Darn Hot

(page 3 of 3)

Jana Leon

null

Illustration and chart by Ray Vella

Sun's Tremblay pioneered multicore chips that run slower and cooler Richard Morgenstein

Solutions to tech's energy crisis won't come easily. The largest data centers can cost more than $200 million to construct and are expensive to upgrade to the latest energy-saving technology.

There are a number of relatively quick fixes, however. Most server computers use only about 10% of their capacity at any given time, so data center operators are using software that shifts computing jobs between servers to make the most of their capabilities. New power management systems shut off servers and other equipment automatically when not in use. And most computer companies have caught the green religion. They claim their newest equipment is ultra-energy-efficient.

The problem is that widely accepted yardsticks don't exist for comparing one server computer with another to show which is the more green. For servers or data centers, there's no equivalent of the miles-per-gallon ratings for cars. The Environmental Protection Agency and Energy Dept. are working with industry groups to set up benchmarks.

Although more progress is needed, there have been some major advances in data center energy conservation. In each case, the breakthroughs were the result of scientists and others questioning the conventional wisdom in their industries.

Take the design of the chips themselves. A decade ago the chip industry had a single focus: making chips process bits and bytes ever faster. But Marc Tremblay, a top scientist at Sun Microsystems (JAVA), saw a fatal flaw in that strategy. The faster chips ran, the hotter they got, and eventually they'd be too hot to work properly. So he designed a so-called multicore chip, which has several processors on a single sliver of silicon. Each core runs slower and cooler than a processor that does the same amount of work all by itself.

ABOUT-FACE

Because of the complexities of semiconductor technology, it took nearly a decade for Tremblay's innovation to come to market. Since Sun's server computers based on Tremblay's designs went on sale two years ago, they have helped turn the company around and have dramatically slashed energy consumption. Each chip in those servers consumes just 70 watts of power, about one-third that of a conventional microprocessor.

Even the traditionally slow-moving utility industry is showing signs of surprising change. Take PG&E. Two years ago, after Bramfitt was handed the job of designing conservation incentives for Northern California's tech companies, he came up with a creative solution. He decided to provide financial incentives to companies that decide to use what's known as virtualization software. Without the software, each server computer typically handles only one program at a time, and only a fraction of the capacity is used. With virtualization software, many programs can be run on a single computer.

The result is that managers can pack more programs onto each server, thus burning less electricity and reducing the number of computers needed. Under Bramfitt's incentive program, PG&E pays customers for every kilowatt-hour of energy they save using the software.

Bramfitt's program is a business model innovation, but most of the data center energy savings are likely to be found in technology advances. IBM researcher Michel is focusing on a seldom-trod territory—where biology and physics meet. Michel, who has a PhD in biochemistry from the University of Zurich, is designing devices that he expects will one day cool chips with a system modeled after the human body. While the processors in server computers are typically cooled with air, these chips are chilled with a liquid delivered through a system similar to the body's capillaries. One of Michel's inventions is a metal cap that fits over a processor and sprays jets of water out of some 50,000 nozzles into microscopic channels etched in the metal. The channels circulate the liquid efficiently and cut the amount of energy required to pump the water.

While there have been substantial improvements, more innovations will be necessary in fields from technology to utilities. Putting data centers in Iceland or Siberia may help, but that on its own won't be enough to solve tech's energy crisis.

With Kerry Capell in Grindavik, Iceland

Reader Discussion

 

BW Mall - Sponsored Links