Over the last two years, data has seen explosive growth and now 2.5 quintillion bytes are created daily. This number is so large that it represents 2,220 petabytes, where each petabyte represents 1,024 terabytes. It's essentially an incomprehensible amount of data that's created each day. An easier way to understand it is that it's so much data that 90% of the world's data was created in the last two years alone.
Running an ever-increasing army of processors to power and ultimately analyze that data has its fair share of associated costs. Data centers that house this information have become challenged keep a lid on energy costs. According to The New York Times, data centers consumed 76 billion kilowatt-hours of electricity in 2010, or about 2% of all U.S. electricity consumption. Since data hasn't stopped growing, it's likely data center energy needs will grow in lockstep. The evidence to support this claim lies within the rise of extreme low-energy servers, which Intel (INTC +4.96%) estimates to represent 6% to 10% of the server market. ARM Holdings (ARMH +0.00%) and its proven power-efficient architecture is currently being courted for the job, but many challenges lie ahead before this low-energy dream can become a reality. However, if ARM can succeed, there's a sizable opportunity for it to encroach on Intel's stronghold in the server market.
The rise of the cluster
No, I'm not talking about your favorite cereal, I'm talking about server farms that utilize tens of thousands low-power processors to collectively take on larger tasks by breaking them into byte-sized, manageable pieces. This style of data processing would be like using an ant farm instead of the Incredible Hulk to get the job done. It's compelling because it works really well for cloud-based services like Facebook, Amazon, and Google. Given ARM's current undisputed energy advantage in mobile computing, efforts have been ramped up to develop the same pedigree for server applications.
Viva Barcelona!
The Barcelona Supercomputer Center is building Mont-Blanc, the world's most energy efficient supercomputer by using NVIDIA's (NVDA +2.71%) Tegra 3 mobile processor over the typical Intel Xeon found in supercomputers. For comparison, the typical Xeon processor consumes 50 to 100 watts of energy, where each Tegra will be tuned to consume just 4 watts. Although the project is using off-the-shelf materials, it's not an off-the-shelf initiative. Practically every piece of code written for Intel x86-based architecture needs to be repurposed in order to take advantage of Tegra 3's ARM-based architecture. For data centers, this is great news, because whatever is learned and developed during the process will be openly shared with the community.
First movers
NVIDIA, Dell (DELL +0.00%), and Hewlett-Packard (HPQ 2.65%) all have their sights set on developing products for the ARM server revolution. NVIDIA is developing Project Denver, a custom 64-bit ARM chip design for servers, which is important because servers rely on 64-bit code, and this will be one of the first 64-bit ARM designs available sometime in 2014.
Dell has begun shipping its Copper servers to select customers so they can begin testing and validating ARM architecture for 32-bit standards since 64-bit designs won't be shipping until 2014.
Like a good competitor, HP has announced similar efforts under its Project Moonshot, except that it decided to ditch ARM for Intel's newest Atom chip before shipping the goods. That's because the Atom chip delivers 64-bit support, the same power consumption, a broader ecosystem (and user base), and generally less operational challenges. If HP's decision is any indication, there's far greater demand for Intel's out-of-the-box compatibility over the academic promise ARM hopes to deliver.
Challenges and uncertainty
Between exploiting a massive number of low-power processors, sustaining performance, and programming every aspect of the initiative, it's safe to say the challenges are plentiful for ARM to transition into the server space by 2014. But the greatest challenge of them all is Intel. By the time ARM is ready for prime time, Intel's Atom chip will be fabricated on 14-nanometer transistors, essentially closing the power gap between itself and ARM's latest and greatest. At that time, ARM chips are expected to be fabricated on a larger, less power efficient 20-nanometer node.
If low-power server clusters become the de facto standard for cloud-based data centers, Intel has a huge lead on ARM, as nearly everything in the data center has been developed in favor of Intel's architecture – which commands a 90% market share in the server market. When you also consider Intel's fabrication process will be a solid generation ahead of ARM's fabrication partner, Taiwan Semiconductor, there won't be much reason left to convince businesses to deploy ARM server farms. Despite the efforts being made, the practicality of implementing ARM-based architecture in the data center doesn't appear to be all that practical.
