Intel Launches Centerton to Take on ARM
[caption id="attachment_6464" align="aligncenter" width="500"] Intel executive Diane Bryant helped unveil the Atom S1200 SoC.[/caption] As expected, Intel launched its Atom S1200 System-on-a-Chip (SoC) Dec. 11. Intel hopes that the 6-watt architecture—developed under the codename “Centerton”—will allow it to push back against the emerging ARM microserver market. The Atom S1200 is a dual-core 64-bit chip. It’s been optimized for servers and the data center, with features including ECC, virtualization, and hyperthreading (the lattermost is capable of delivering four threads per chip). The eight lanes of PCI Express 2.0, along with a memory controller supporting up to 8 GBytes of DDR3 memory, help make the S1200 a true System-on-a-Chip. The Atom S1200 will run at speeds between 1.6 GHz and 2.0 GHz; prices reach as low as $54 apiece in quantities of 100, according to Jason Waxman, Intel’s cloud chief. The architecture supports up to 1,000 cores per rack, with Intel executives suggesting that some OEMs have surpassed benchmark that quite handily. It also supports all the major hypervisors. Intel also said that it had twenty partners signed up to use the new processor, including big names such as Hewlett-Packard and Dell. The latter likely will announce its support onstage at the Dell World conference tomorrow at Austin, Tex. Executives said that the new Atom chips would be used for generic Web serving, basic switching, and low-end storage. “I really love it. I think it’s exciting in a lot of ways,” said Paul Santeler, vice president of the hyperscale business unit at Hewlett-Packard. Next quarter, HP will launch its Gemini server, the first incarnation of the “Project Moonshot” low-power initiative, which will use the S1200 chip. Much has been made of the microserver market and its use of low-power “wimpy” cores. Most servers use so-called “brawny” cores such as the Intel Xeon, which are traditionally used to deliver the most horsepower (and damn the power consumption!). But with increasing priority placed on total cost of ownership, power has become a greater factor over the past few years. At Facebook, for example, the argument for products like that Atom is a simple one: they do far more useful work per watt per dollar. For Intel, the question is a moot point. Diane Bryant, the vice president and general manager in charge of the Datacenter and Connected System Group at Intel, showed the audience a slide that compared a 560-node Atom S1200 server with a Xeon server using a fifth of those processors, or about 100 or so. The slide attempted to show the difference between maximizing the Web servers used (Atom) versus the maximum Web transactions per minute per rack (Xeon). Intel won either way: the CPU cost for the Xeon server was about $32,900, while the Atom cost was even higher, at $35,900. Bryant said that the company would continue to invest equally in both Xeon and the new Atom parts. An equally vexing question has been how much of the market will actually use these new low-power servers, and how many will prefer the traditional brawny cores like the Xeon. After saying he’d prefer to defer to industry analysts, who have varied their estimates of the microserver market between single digits and almost 25 percent, Bryant gave in and settled in a 10 percent figure: “As governments around the world continue to invest in IT buildout it becomes all about scale… It drives an incredible focus on total cost of ownership and a lower cost of delivery.” Back in 2009, Intel launched the category of microservers, Bryant claimed, and quickly followed that up with a 45-watt Xeon processor in 2010, succeeded two years later by a low-power, 17-watt Xeon processor. Meanwhile, rival AMD has made performance per-watt per-dollar its watchword, with the company launching plans for ARM-based Opterons as well as SeaMicro servers that take the company’s Freedom Fabric and marry it with architectures that include the Atom, Xeon, and AMD’s own chips. Fabric will quickly become the focus in 2013, as Intel readies its 22-nm Avoton parts. Bryant said that those chips would include an integrated fabric controller, part of a top-to-bottom commitment Intel has made to fabric during that year. The next-generation of the Atom S Series will be a 14-nm part due in 2014.