On Tuesday, Dell announced a prototype low-power server built with ARM processors, code-named “Copper,” which the company has been developing since 2010. Dell will not be selling the ARM server outright, but will provide these “seed units” to select customers and partners for application development and benchmarking.

There is a growing demand from Internet-based companies for custom-built servers that can scale performance while reducing datacenter operating expenses. The increasing popularity of cloud computing is also pushing the microprocessor market in the direction of power efficient servers for cloud providers who are relying on economies of scale to achieve and maintain profitability.

The ARM chips’s main focus is on power-efficiency and because of that they have become the processor-of-choice for battery-dependent mobile computing platforms. However, the chips that live inside today’s smart phones and tablet devices are becoming increasingly performance-optimized as well, making them attractive to server makers looking for that sweet spot between performance and power consumption.

In its official announcement, Dell noted that it is enabling ARM server development in multiple ways. While the company is delivering its ARM-based servers to select hyperscale customers and partners, it will also support the ARM development space by providing the servers to key ecosystem partners such as Canonical and Cloudera. Dell’s ARM-based servers will even be available to developers via remote access through a partnership with Texas Advanced Computing Center (TACC).

Dell “Copper” ARM server sled – click for full-size image.

Each ARM microserver runs at 15 watts, about a third the power draw of Intel’s Xeon E3 cloud-friendly chips. For the architecture, Dell selected Marvell’s quad-core Armada XP 78460 chip, which runs at 1.6GHz and handles up to 8GB of ECC memory. Copper’s 3U rackmount chassis contains 48 independent servers and 192 processor cores. There are four ARM server nodes per sled, and 12 sleds total (hence the 48 servers). The total power draw for a full chassis is less than 750 watts.

As it stands today, most servers, including Dell’s, are outfitted with x86 architecture chips, the vast majority from industry stalwarts (and noted adversaries), Intel and AMD. But in an effort to tip the power-profile in their favor, companies whose life-blood depends on enormous server farms, Web-scale outfits like Google and Facebook, are looking to the microserver, a low-power, and slower, platform, to fulfill their data crunching needs at a lower TCO.

Says Steve Cummings, executive director for marketing at Dell’s Data Center Solutions division:

We believe ARM-processor-based infrastructures demonstrate promise for Web front-end and Hadoop environments, where advantages in performance per dollar and performance per watt are critical. And we designed the server specifically for where the market is today, for developers and customers to create code and test performance.

A 2011 global census from DatacenterDynamics cited cost and availability of energy as top concerns of execs planning future datacenter expansions. The same report predicted that datacenter energy use would rise almost 20 percent over the following year. Cloud computing, by some accounts, is expected to offset some or even all of that increase. And no doubt, developments in hardware and fabric technologies, customized for the hyperscale computing era, will be part of the effort to keep energy costs down.

Dell joins the ranks of other server makers who are testing the microserver waters. Last November, HP announced its hyperscale intentions when it debuted Project Moonshot, seeking to redesign servers in preparation for a Web-scale era, and the HP Redstone Server platform, based on Calxeda ARM Cortex processors. The project’s goal was to fit 2,800 servers on a single rack. SeaMicro, a popular microserver startup that alternately packaged both Atom and Xeon chips into an ultra-efficient server design, was acquired by AMD earlier this year. Now AMD is in a position to offer its own low-power server building blocks, making them one to watch in the race to accommodate the ultra-scale datacenter market.

So is Intel worried? Even Intel’s Justin Rattner claimed that so-called weak processors can be “dramatically more efficient” on certain types of cloud workloads versus traditional x86 servers. Intel’s microserver candidates includes Intel Xeon E3 processors that range from 45W down to 20W, and at the Intel Developer Forum (IDF) in Beijing, the chipmaker announced a low-cost, sub-10-watt microserver platform known as Centerton. The 64-bit chip features two Atom processor cores and consumes only six watts of electricity. Intel Labs has also been working on a highly-specialized multicore chip, called the Single-Chip Cloud Computer, which debuted in December 2009. The company says the 48-core chip mimics cloud computing at the chip level and supports highly parallel, scale-out programming models. With all 48 cores running at once, the SCC is said to consume between 25W to 125W.

Of course none of these specialized chip architectures mean much without applications that can make use of them. And that pretty much sums up where the market is at right now, the testing and development phase. Prototypes like Copper give application developers a canvas for creating the next-generation of software, applications and tools that can take advantage of massively parallel computing platforms. The money saved on the hardware side from moving to stripped-down, bare-bones systems will more than pay for the software redesign, resulting in a net gain. That’s the idea behind this whole strategy: the sheer size of the deployment provides the economies of scale to make it work. Or that’s the theory anyway.

Dell says that its initial focus is on evolving the ecosystem, and that it will make its ARM servers generally available “at the appropriate time.” This is a work-in-progress, as Steve Cummings, explains in more detail:

The ARM server ecosystem is still immature, with a limited software ecosystem and (until now) no ARM-based servers from a tier one OEM. Plus, ARM is currently 32-bit technology, which means current 64-bit code would have to be modified to run on 32-bit, and likely be modified again when 64-bit comes out in the next year or two. So customers have told us they don’t plan to put ARM servers into a production environment, but instead want servers to test and validate in their labs.

The manufacturer behind the Dell ARM server chip, Marvell, believes the partnership will lead the way to bullet-proof cloud solutions. This message was underscored by Paul Valentine, vice president of marketing for the Cloud Services and Infrastructure (CSI) Business Unit of Marvell Semiconductor, Inc.:

“Today’s data centers run the distinct risk of over-extension due to the rising popularity of connected lifestyles and the resulting explosion in unstructured data. A key component of Marvell’s all-encompassing cloud-services platform, the Marvell ARMADA XP series of multi-core processors, represents a benchmark in security, scalability, performance and power conservation – ultimately offering a vast amount of headroom to cloud service providers looking to reinforce their capacities for the long haul.”