The Economics of Servers Could Soon Change

Marvell said today that it has built a chip designed for servers that uses the same architecture as chips inside cell phones. The chip’s four 1.6 GHz processors aren’t notable for their performance compared to today’s server chips, which use the x86 architecture used by Intel and AMD. However, the power savings will likely be significant compared with an x86-based chip.

Such power savings lowers the overall compute cost because it lowers the cost of running both servers and data centers (chips that don’t get as hot don’t require as much air to cool). We covered this phenomenon in a story on Smooth-Stone, which, like Marvell is using the ARM instruction set for server chips, but also with Sea Micro, which is using low-powered x86 Atom chips from Intel to deliver similar savings. Beyond the power savings leading to actual dollar savings, a bigger economic shift is at work — the threat posed by the ARM licensing model.

Intel and AMD share an uneasy duopoly when it comes to x86 chips (yes there’s Via, but Via’s chips haven’t gained ground in servers), which means that the folks who build servers have two choices. Even when AMD makes huge technological leaps, it generally finds itself still playing second fiddle to Intel because of a variety of competitive and less-than-competitive reasons. Add to this the overall horizontally integrated nature of the computing market today where chips come from one vendor, the servers from another and the OS platform from yet another, and the result is commodity boxes with the base processors having similar features and similar price points.

Contrast this with the ARM licensing model in which the British company that owns the ARM instruction set offers a license to hundreds of companies that want to build chips using the ARM core. A select few licensees take out an even more flexible license that allows them to tweak the core. Qualcomm, Samsung, Microsoft and Apple all have such licenses. Many, many companies can build processors around the ARM instruction set whereas Intel holds the x86 instruction set fairly close. So what happens when server manufacturers and companies building out large compute platforms realize that instead of Intel or AMD they can now select from a raft of chipmakers that are building chips for servers or appliances for cloud computing using ARM-based chips?

Already the huge problems of webscale and cloud computing are demanding specialized hardware and more vertically integrated products such as Cisco’s servers or IBM’s CloudBurst gear. For those pushing the computing envelope, could ARM and a new wave of chip providers tackling the server market lead to radical changes in performance, cost and designs that the tight confines of the x86 market and horizontal integration have so far limited?

Could ARM be an alternative to vertical integration that could lock companies into one hardware vendor or could it be a pain in the butt since it means a rewriting of software that had been designed for x86 machines? ARM may in fact be used to create better vertical integration. Apple’s ARM-based A4 processor in the iPhone and iPad is a wonderful example of that model. At some level the adoption of ARM inside large-scale data centers helps break the idea that server operating systems and individual applications are the norm, given that much of the individual applications and traditional server OSes aren’t designed to play well on the architecture. It brings us closer to the idea of the data center as a computer with the real operating system being at the management layer. If you can make the management layer play nice with a variety of hardware (x86 or not), then the determining factor for hardware and chips isn’t how nicely it plays with a bunch of horizontally integrated pieces but how well it performs from a price, power and efficiency standpoint.