ARMs and Microservers: Shrinking Data Center Footprints and Power Consumption

Data center infrastructure is undergoing significant changes across a range of hardware platforms, but server technology appears poised for a particularly dramatic shake-up as designs become both smaller and more powerful.

The fuel in this fire are the new classes of low-power processors, led by the architecture developed by ARM Holdings, that are rapidly making the jump from mobile devices to enterprise-class microservers. This trend saw several key developments recently as top designers joined forces to make the technology more amenable to the high-density configurations of leading enterprise architectures.

The latest is Marvell and SanDisk, which are uniting the quad-core ARMADA XP System-on-Chip architecture with new SATA SSD modules in pursuit of an easily deployable, highly scalable system suitable for the demands of cloud computing. The design encompasses up to four 1.6 GHz ARM V7 cores that support both symmetrical and asymmetrical multi-processing, as well as 64-bit DDR2 and DDR3 memory, coupled with up to 256 GB of storage capacity and the 500 MBps read/ 430 MBps write speeds of the SSD’s 6 Gbps SATA interface.

At the same time, ARM Holdings has teamed up with chip-maker Taiwan Semiconductor Manufacturing to put its 64-bit v8 architecture on a 20 nm process, with plans to shrink that to 15 nm before long. The processors will feature TSMC’s finFet technology, along with new 3D transistor designs, to drive performance and lower power consumption. The move is seen as a direct challenge to Intel’s Ivy Bridge processor, which already occupies a 22 nm process and features the company’s Tri-Gate transistor.

Indeed, with ARM already dominant in the mobile sphere, enterprise servers represent the company’s next big push. ARM already claims a 15-fold improvement in leading power/performance benchmarks by putting its mobile technology in enterprise settings. With the new v8 architecture geared specifically for challenging data center environments, the company represents the strongest chance in decades to shake up an industry that has largely settled on the x86 platform.

A key component of this strategy is the fact that ARM does not produce any chips of its own. Rather, it licenses the ARM reference architecture to various chip developers and lets them add features that appeal to users. A case in point is Samsung, which is said to be preparing a 2014 release of an ARM v8 SoC aimed at hyperscale datacenters and other cloud-facing operations. The goal is to provide massive scalability at extremely low cost so providers can more easily tailor their hardware footprints to the rapidly changing workloads of the cloud.

ARM may have a good bit of momentum heading into the server arena, but the company isn’t going up against lightweights. Both Intel and AMD are fully aware of the threat that ARM poses and are knee-deep in their own low-power/microserver developments. These two have been slugging it out in the data center for almost two decades and have built up strong relationships with users, OEMs and channel suppliers.

ARM has already brought a lot to the enterprise table, but it will need to make a compelling argument if it hopes to convince CIOs to place their trust in an entirely new processor architecture.