RISC vs. CISC.

The battle over which processor architecture is really the best has dragged on for nearly 30 years.

So far, Intel, with its CISC (complex instruction set computing) processors, has won in nearly every case in supercomputers, data centers, PCs and laptops because of their ability to hit performance benchmarks, while RISC (reduced instruction set computing) from vendors like ARM have won in phones and industrial devices like smart meters where overall efficiency and lower prices were required.

Calxeda, formerly known as Smooth-Stone, Marvell and others now say they can change the equation with ARM cores tweaked to run in servers. They key competitive advantage is energy consumption.

A server powered by Calxeda chips will cost half as much to make as standard Intel-based servers, according to Karl Freund, an IBM and Hewlett-Packard alum who has just become Calxeda’s vice president of marketing. At the same time, they will consume just one-tenth the power. The power figure includes savings from air conditioning, but even with air conditioning not counted, Calxeda’s servers will use far less energy.

“The main difference is that these cores were designed for battery performance where every milliwatt counts,” he said.

When idle, Intel chips fall back into a sleep state where they consume 80 percent of their ordinary power, he said. Calxeda’s chips fall into a near-comatose state. Since servers (and their processors) are really only active 15 percent to 20 percent of the time, much of the power that goes into data centers is wasted right now.

“You will get a server core that’s so efficient that you won’t mind that it is 20 percent efficient” in terms of utilization, he said.

And power is a huge concern. Yahoo recently built a data center near Buffalo, New York so that it could cool it with winds from Lake Erie. Calxeda investor Mike Dauber of Battery Ventures asserts that many large data center customers are worried about physically being able to obtain enough power to run their future data centers: internal power consumption will have to increase if corporate growth is to continue.

The company is already in discussions with large server manufacturers and websites. Some large websites like Google even make their own servers, potentially making it easier to gain a foothold. That websites tend to run Linux instead of Microsoft helps, too.

But it’s a big if. Several companies -- Montalvo Systems, Rise, Transmeta, Cyrix, IDT, National Semiconductor -- have tried to take on Intel with cheap and/or low-powered processors. Many raised millions in VC funds and lined up partnerships with high-profile companies like IBM.  Most died horrible deaths. Advanced Micro Devices, which operates under a design license from Intel, has survived, but barely at times. Since it was founded in 1969, AMD has actually lost more money than it has made.

The manufacturing prowess, aggressive product roadmap and even the sales deals Intel can put together are formidable. One executive at an Intel competitor told me once how he signed a small deal with a Taiwanese motherboard maker. Intel execs were on a plane soon afterwards, and the deal collapsed.

Even classic RISC chip vendors like IBM, Freescale (formerly Motorola) and Sun have been pushed to the margins. Back in 1993, Intel occupied only around 3 percent of the market for server chips. By 2003, Intel and AMD owned over 90 percent of the market.

“There is a niche there, but there is also a strong, entrenched competitor,” said Dean McCarron of Mercury Research, one of the premier research firms in processors. “If they can demonstrate significant power savings, they could find customers.”

Second big if: Calxeda is just starting out. It has raised $48 million and has alliances with ARM and Texas Instruments. But the company has yet to “tape out” or finalize the blueprint of its first chip. It does not have samples. Freund admits that Calxeda will have to beef up the number-crunching ability of the floating point unit inside its chips to meet the demands of data centers.

The chip, in fact, will only run 32-bit code. Many server apps have already graduated to a 64-bit world. (Editor’s note: suffice it to say that 64-bit code started getting big toward the beginning of the decade.)

Freund points out that the Calxeda, unlike the vast majority of Intel wannabes that have gone before, will not directly attack Intel. Instead, the company only aims to insert its chips into servers used to cache data, run Java programs or animate websites.

“You don’t want to run Oracle or SQL Server on it,” he said. “You do not want to take them and plop them into a blade server.”

In this market sliver, the stripped-down nature of Calxeda becomes an advantage. The chip won’t contain circuitry that might really only be needed by a desktop or standard server.

“If you are going to be all things to all people, there is real estate in there that people aren’t going to use,” he said.

(Calxeda means 'smooth stone' in Latin.)