Skip to content

Intel Doubles Down on Microserver Biz

June 28, 2013

At a media event last week in San Francisco, chipmaker Intel Corp. outlined its datacenter strategy for the next few years. A chunk of the day’s talks centered on the evolution of the Atom microserver roadmap and a brand-new low-power server chip, based on the Broadwell microarchitecture. Many view the unnamed SoC as Intel’s response to growing ARM encroachment on its x86 dominance.

“We know that power continues to be a constraint for the datacenter, both power delivery and cooling,” said Diane Bryant, senior vice president and general manager of Intel’s Datacenter & Connected Systems Group during her keynote speech. “With that as a fundamental datacenter constraint, as we move to these large datacenter and scaled solutions, we’ve doubled our investment in low-power high-density processor product line.”

The Atom-based server roadmap, which Intel debuted last year, takes the Atom processor core and integrates it with server, storage or network specific functionality to deliver high-density, lower-power solutions for low-end networking and storage and the microserver space.

Intel’s initial foray into the microserver space with the dual-core, 64-bit Atom S1200 SoC, known as Centerton, was a bit underwhelming, with one analyst from the Linley Group calling it a placeholder for the upcoming Avoton 22-nm CPU.

Now the focus of Intel’s microserver roadmap has shifted to the C2000 product family, which includes the second generation 64-bit Atom SoCs: “Avoton” (for servers and storage) and “Rangeley” (for networking applications). Both versions are based on the “Silvermont” microarchitecture core. As the first Atom refresh since 2008, Silvermont promises a 3X improvement in power at the same performance level or 5X improvement in performance at the same power level. Intel has been shipping samples of these products to customers since March and is planning on a full roll-out in the second half of this year.

Intel’s first 14nm server SoCs, codenamed Broadwell (Xeon) and Denverton (Atom) are due out in 2014. The pairing highlights the work that Intel has done on the Atom side to align the the process technology for the two families.

Intel product roadmap, announced July 22, 2013

Intel has been offering its low-power Atom processors as an alternative to ARM designs. In his keynote, Jason Waxman, VP and GM of Intel’s Cloud Infrastructure Group, claimed that besides offering the familiarity of the standard x86 instruction set, Avoton beats the ARM architecture on power efficiency.

One of the big surprises for the day came during Bryant’s talk. She disclosed that Intel is launching a completely new SoC based on the Broadwell microarchitecture. The as-yet unnamed chip will sport integrated I/O, networking fabric and accelerators, and will target server, storage and network applications.

Intel is playing up the best-of-both worlds of this SoC, high performance and high density. “Some customers have been telling us they still want the performance of Xeon, but the extra features in the SoC products, so this will be the first Xeon-based SoC,” said Waxman.

The company has not yet decided on target workloads for the new SoC. “We’re still learning and we’re letting customers experiment and give us feedback on it,” said Waxman. “I expect that it will probably be an extension of where Avoton plays. It’s a way to work things up, rather than a replacement.”

This leaves some question as to whether Intel will go with the Xeon or Atom product branding. Xeon seems to be the more obvious answer, at first blush anyway. Not only is it based on the upcoming Xeon “Broadwell” architecture, but one would think Intel would want to play off its more powerful Xeon name recognition. There’s also the very real possibility that a stronger Atom microserver family will cut into Xeon sales. But whether product positioning by way of branding/marketing can ultimately affect this potentiality is doubtful over the long-term.

The main message for the day was an application-centric one:

“The goal here is to always have the infrastructure solution that is optimized for that given workload, whatever that workload is,” said Bryant. “So we have moved to looking at this space from an application-optimized infrastructure solution all focused on reducing the total cost of ownership of the datacenter, whether it’s enterprise IT, whether it’s the cloud service providers or whether it’s telco, the objective is the same: reduce the total cost of running that datacenter.”