Intel intros Atom S1200 SoC for enterprise apps, data centers

SAN FRANCISCO -- In one of its final announcements of the year, Intel has unveiled its new Atom S1200 processor product family.

During a media presentation on Tuesday morning, Diane Bryant, vice president and general manager of Intel’s Datacenter and Connected Systems Group, described that the Atom S1200 System-on-a-Chip platform is the "world's first" 6-watt, 64-bit SoC for datacenters and enterprise applications.

See also: The low-power struggle: Intel 'Centerton' Atom vs ARM Cortex A9
Bryant added that the new Atom SoC extends Intel's "breadth of application-optimized datacenter products."

Other design notes include dual-core power with Intel's hyper-threading as well as virtualization technology, error-correcting code (ECC) memory, and more than 1,000 nodes per rack.

One of the primary objectives is to deliver right solutions stack to deliver memory and CPU-intensive workloads. That requires a great deal of scalability, and Bryant acknowledged that comes up against potential rises in total costs of ownership.
Bryant continued that it is consistency in the datacenter that drives down costs while variation drives up costs. For Intel, Bryant said, the focus is to drive down variation across the datacenter.

While Bryant noted that Intel wants to address all kinds of workloads across the datacenter with the S1200 family, engineers did design this processor family with some specific use cases in mind, such as micro servers, communications, and low-end storage.

Available and shipping now, Bryant noted that there are over 20 OEM partners already signed on to use S1200, including HP, Huawei and Dell.

Jeffery Snover, a lead architect for the Windows Server team at Microsoft, said that Microsoft is "very excited" that there is now a "very low-energy part that can run the demands of servers." He added that Windows Server and Microsoft's partners "will be able to run on a low-cost, energy-efficient Atom processor but also on high-end Intel-based servers."

Reiterating one of the foundations of the Wintel alliance, Snover also cited that Intel's growing range of servers are "safer and easier" for customers to adopt because they are backwards-compatible with previous versions of Windows ecosystems.

Paul Santeler, vice president of the Hyperscale Business Unit for Industry-Standard Servers at Hewlett-Packard, remarked during the presentation "that there's an interesting dynamic in our marketplace" right now.

HP will be using the S1200 SoC for its Project Moonshot production system for extreme low-energy computing. Santeler added that the next-generation Gemini product will be introduced in the next quarter.

He explained that "in today's environment" as everything moves towards large-scale applications, we're at the point where we have to be designing complete solutions and systems that focus on those different and various applications.

Quite simply, Santeler said, "one size no longer fits all."
That presents an interesting challenge for Intel as it is attempting to serve all possible datacenter workloads with S1200. From one angle, that looks like a "one size fits all" attempt, but it looks like Intel is trying to make the case that it can offer a solution that is optimized for any individual workload.

One of the driving forces for more efficient and highly-scalable datacenters is the growing amount of data stemming from social media.

No company probably knows more about that than Facebook, with more than one billion members worldwide.

To just get a snapshot of how much information is being posted to Facebook, Frank Frankovsky, vice president of hardware design and supply chain, cited that there are more than 4.5 billion likes, posts, comments and photos added per day.

Acknowledging the obvious that "that's a lot of compute power," Frankovsky said the challenge for Facebook's engineering team is to determine how to get the right CPU applied to the right workload.

In expressing support and hope for the S1200 product family, Frankovsky commented that when you take really efficient processing and integrate some of the connecting chipsets into a single package, you can drop the number of power required and the costs of those chips significantly.

Frankovsky concluded, "It's all about useful work per watt per dollar."