Intel isn’t letting ARM, VIA, or a bunch of startups run away with its server business, and is pushing the concept of low-power chips to address smaller workloads, such as those needed by Facebook or a web service. Today, it outlined its plans for the micro server category and said it would create server chips with power consumptions as low as 10 watts by 2012.
Intel’s Boyd Davis, general manager of the Data Center Group, said on a call today that the chipmaker would address this market for highly dense and low-power architectures for cloud-based web service companies, a market Intel estimates might become 10 percent of the total server market in the next four to five years. Today, Intel serves that market with Xeon chips running at 45 and 30 watts, but in the second half of next year Intel will offer a 20-watt Xeon chip and eventually a 10-watt low-power Atom chip for servers. Davis noted, however, that the market may never reach that 10 percent figure because computing needs may change or be fulfilled by even newer types of servers. He also guessed that the majority of that 10 percent will be taken care of by Intel’s Xeon processers. But Atom will still be important for the few customers who demand even more power efficiencies and don’t want to virtualize their applications and Intel won’t leave those customers behind.
That wasn’t just a dig at the efforts of ARM or startups using ARM or other instruction sets, such as Tilera or Calxeda. There truly are advantages today to using higher-level chips, such as the fact that they run 64-bit software common in the enterprise, they can address more memory and they can contain virtualization hooks built into the chip. But there are also advantages to micro servers, and Intel isn’t going to let anyone take on its markets without a fight. It already designed a special x86 Atom chip for SeaMicro, a startup building a micro server for web applications.
Intel must address a widening market for server chips that have to take into account resource-intensive database applications with highly redundant and scaled-out web servers, but it must do this profitably. As Davis said in the conference call, “How do we have a category that is optimized enough to address their needs but versatile enough to make it worth our while?”
The micro server can address several issues that plague webscale operations, such as space and power constraints. For example, SeaMicro’s server crams 256 dual-core Atom processors in a box that is a quarter of a rack and consumers a quarter of the energy. Another issue is communications. In the mindset that says more processes require more servers, one also has to deal with more ports creating a networking nightmare. Micro servers generally have a fabric inside the box that cuts down on ports and handles the networking complexity.
Regardless, the move toward a more compute-dependent society means that we’re refining our compute needs for specific workloads. As Gio Coglitore, director of Facebook Labs, said on the call, “We aren’t trying to cram a desktop processor into a server anymore … For Facebook there is no one-size-fits-all CPU.”
There may not be a one-size-fits-all CPU, but Intel wants to make sure there is a one-size fits all instruction set, and that the winning one is Intel’s own x86 architecture. For more on micro servers and greener data centers, come see me interview Google’s Green Energy Czar Bill Weihl and Christina Page, Director Climate and Energy Strategy at Yahoo, at our Green:Net event in San Francisco on April 21.