Is the Market Ready for ARM Servers?

Is the Market Ready for ARM Servers?

December 1

Back in 2011, ARM Holdings and its partners began presenting a vision of building servers that were extremely power efficient and specialized. They would follow SoC design, and would reduce not only the cost of the physical servers but would also drive down operating costs due to their space saving design and power efficiency.

Intel and AMD have dominated the traditional server market for nearly two decades. In that time they have seen their fair share of challengers. Few companies have the deep pockets to build expensive fabrication plants and maintain a significant R&D budget to remain in a game that rewards blockbuster hits such as the Intel Pentium and Xeon processors.

Today’s servers are primarily powered by the Intel Xeon or AMD Opteron processors. They are powerful, scalable, and support a wide range of hardware and software. These are the Ferrari and Lamborghini of processors, optimized for performance over gas mileage.

But what happens when the Prius of processors comes along and decides that, instead of going head-to-head with Intel and AMD, they focus on delivering features that are not generally associated with the competition: A specialized chip that’s low in cost and sips power. That’s the hand ARM intends to play, and it just might be a winner.

ARM chips may not carry the same panache as the Intel Xeon, but they are actually the most widely used processors in the world, powering billions of smartphones, tablets, TVs, and many other consumer and industrial products. Every iPhone, iPod, iPads, Android and Windows Phone sold uses an ARM chip.

ARM licenses their chips’ designs to over a dozen semiconductor companies, unlike Intel and AMD who design and manufacture their own chips, but don’t license them. ARM’s licensing model results in a competitive ecosystem that helps keep prices low and rewards incremental chip improvements.

Specialized Workloads

Nearly every gadget produced today connects to the internet. We’re seeing a proliferation of cloud-enabled devices that allow users to interact with remote cloud-based software. Health trackers like Fitbit and the Nest Learning Thermostat are two examples. These and thousands of other devices are creating a lot of data that can be analyzed to provide better context than a local application. One example of this is how my Nest thermostat connects to the cloud to tell me the temperature people in my area set their Nests to, allowing me to plan accordingly.

ARM servers can be optimized to handle these highly specialized workloads. Instead of running in a highly virtualized runtime environment, ARM servers are able to run individual, specialized workloads that are optimized for specific applications these devices demand.

Low Power, Lower Cost, & Space Density

Power consumption is a major consideration, be it in-house or at a data center. ARM chips, with their roots in mobile, have optimized every component of their chip design to sip as little power as possible. That makes a lot of sense when you’re building chips for phones and tablets that are used untethered from a power cable. Intel and AMD have prioritized performance over power efficiency, which has made sense for many years.

But all that is changing as the number and size of data centers continues to grow along with our desire for more cloud services and connected devices. Think about how you managed your movie and music collection just a decade ago compared to today. Netflix, Spotify, and other steaming content services have taken over that task for millions. All that content resides in the cloud and requires highly specialized servers to deliver it to millions of customers at the right time and on the right device.

Not only are ARM servers more energy efficient, but they take up less space in a data center where real estate is immensely valuable. The first ARM servers to hit the market, such as the modular HP Moonshot 1500, have a much smaller physical footprint compared to traditional servers. To quote Patrick Moorhead of Forbes:

“Datacenters can put up to 1800 ARM servers in a single, 47U rack, which could take 10X as many racks using a standard architecture.  This extreme density reduces, per a given unit of work, the datacenter size, energy consumption, complexity, and cost.”

According to HP, the Moonshot uses 65% less power, takes up 90% less space and requires 98% less cabling. Even if those numbers are apply to a small number of users, the message is clear that ARM intends to offer a compelling option in the data center of the future.

Traditional servers will be with us for many years. But we will see a transition to servers that are optimized for specialized workloads, and that’s where ARM servers shine. The stakes are so high that you can bet that Intel and AMD won’t stand around while ARM devours their prized customers.

Photo courtesy of HP