Computing giant Hewlett-Packard announced some new additions to its Project Moonshot lineup of servers, and what sets them apart is not so much what they can do but what is in them.
Making good on a prediction we made at the start of the year, HP has added chips based on the ARM architecture to these servers — not Intel chips, the ones that currently run practically all of the main computing engines inside the servers that run in corporate data centers and power most of the Internet.
Most of HP’s servers will still be using Intel chips. But HP sells more servers than anyone in the world, and this is the first time in the last few years that companies buying servers will have any choice about the kind of processors inside them.
Previously, the only other choice in server chips was the Opteron from Advanced Micro Devices, a chip that has failed to keep competitive pace with Intel. ARM chips power most of the world’s smartphones and tablets, and have occasionally been used in notebook PCs; they are designed by the British firm ARM Holdings, then licensed to other vendors.
In broad brush-strokes, the main advantage ARM chips have is power efficiency: They can get a lot of computing work done without consuming a lot of power. That’s why they’re widely used in phones, where battery life is so important. Intel chips have a lot of computing oomph, but have generally tended to be a bit more power hungry. (Intel is usually quick to argue the performance point, though the fact is that its mobile business in struggling.)
Project Moonshot, you’ll remember, is HP’s attempt to redefine the mainstream servers for use in massive data centers. Moonshot machines are small, built to be highly customizable, designed to be sold in large numbers and engineered with minimal power consumption in mind.
Building a data center is essentially a question of packing as much computing power into as tightly a packaged space as possible while controlling the cost of power required to keep it running and to keep the machines cool. The point of HP’s Moonshot is to pack more computing engines into the same space as that of a traditional server rack while keeping the power bill essentially the same.
HP is using ARM chips from two vendors, AppliedMicro and Texas Instruments, in these machines. The two chips are aimed at different types of computing jobs. HP named the U.S. Department of Energy’s Sandia National Labs and the University of Utah as customers for the AppliedMicro-based machine, which is a more mainstream server. It named eBay’s PayPal as a customer of the one using the TI chip, which is aimed at some specialized uses.
So aside from simply adding choice in the marketplace, why would someone want a server with an ARM chip? Gina Longoria, analyst at Moor Insights and strategy, a research firm in Austin, did a deep dive on one of the new machines in a research paper out today.
Longoria focused on a crunchy, but important, metric known as “total cost of ownership,” which basically combined both the cost to buy the hardware, and then the cost of running it — supplying it with power and keeping it cool — over a defined period of time. The findings were interesting, though mileage will probably vary depending on a lot of factors: Over three years, the cost to buy and operate them could run as much as 35 percent lower than traditional servers, and requires only two-thirds as many physical racks. While that may not sound exciting, it’s the sort of thing that makes a CIO’s eyes light up: Multiply that several times within large data centers with thousands of racks of servers in it, and you’re talking real money.
HP isn’t the only one going down this path in its server business. Dell is expected to follow soon, and others will likely get in on the game eventually.
This article originally appeared on Recode.net.