Last week, Intel (s intc) made known its plans to produce an Atom processor designed for use within servers, but the company’s road map — which doesn’t have the server-ready Atom available until 2012 — raises the question of how serious Intel really is about pushing an alternative to its flagship Xeon architecture. With likely more than a year, and several Xeon iterations, between now and the server-ready Atom release, there’s a lot of time for Intel to keep pushing Xeon as the low-power chip of choice. And even after the proposed Atom chip hits the market, Intel’s sales strategy could show just how serious it is about changing its tune to address the needs of energy-conscious customers.
Today, Amazon.com infrastructure guru James Hamilton suggested in his Perspectives blog that the decision to refrain from offering the server-ready Atom processor until 2012 — a process that, technologically, shouldn’t take that long — might be part of a plan to see how well low-power Xeons sell before setting pricing for the Atom chip. As he noted, “Atom is much cheaper than Xeon so Intel will lose money on every workload it wins where that workload used to be Xeon hosted.” Hamilton seems to think that a server-ready Atom chip will only be a truly interesting development if Intel prices it like it prices Atom now — as a “low-cost, high volume device part” rather than as a “server part.”Hamilton’s theory makes sense, as low-power chips like Atom and the forthcoming lineup of ARM-based (s armh) server chips will only be seriously disruptive if they give buyers something to think about. An energy-efficient chip that requires moving to a different architecture and, possibly, instruction set but that costs about the same as a more fully featured and higher-powered, but still relatively efficient, chip on a tried-and-true architecture like Xeon might not present much of a choice at all.
And, as Stacey pointed out in covering Intel’s Atom-for-servers news, Intel already doesn’t seem convinced that Atom necessarily has to replace Xeon as the chip choice in in low-power servers. Citing comments from Intel’s Boyd Davis, she wrote that Intel predicts about 10 percent of the server market will be eyeing up low-power servers, but that Intel thinks Xeon can handle the majority of that demand. Certainly, Intel isn’t letting up on pushing new Xeon chips for servers in the meantime. On Friday, Dell cloud ambassador Barton George posted a video of Intel’s Raejeanne Skillern talking about upcoming Dell (s dell) cloudscale servers featuring low-power Xeon processors.
As of now, the future for micro servers seems primarily centered around Xeon; the only Atom-based server in production is from SeaMicro, which contains 256 specially built dual-core Atom chips. Reporting on that story, Stacey quoted Intel’s Jason Waxman, who pointed out that because SeaMicro’s server includes so many Atom chips, Intel isn’t cannibalizing its Xeon business by partnering with the server startup. For the time being, at least, SeaMicro looks like a way for Intel to hedge its bet on Xeon as the best choice for low-power server chips while it waits to see what happens with lower-power Xeon chips and whether ARM chips gain any traction among major server vendors. If the market demands it, Intel might make Hamilton’s day and sell server-ready Atom chips as the device components they are; otherwise, they might end up being little more than a token offering whose price ultimately ends up making Xeon look even more appealing.