Intel is one of the biggest beneficiaries of Google's rise to power: its chips power hundreds of thousands of the Internet giant's servers. But late Monday night, Google showed it's hedging its bets with IBM's Power processor.
Gordon MacKean, a senior director at Google and the chairman of the OpenPower Foundation that's dedicated to shared designs around the processor design, published a photo on Google+ of a Google server motherboard using IBM's new Power8 chip.
"Today I'm excited to show off a Google Power8 server motherboard in the OpenPower booth at the [IBM] Impact 2014 conference in Las Vegas," MacKean said in his post. "We're always looking to deliver the highest quality of service for our users, and so we built this server to port our software stack to Power."
The average person has no need to know what hardware Google uses. But if Google can find a way to make its services faster, cheaper, more reliable, or more power-efficient, that can benefit anyone using Google's services -- and of course Google's own finances. The company reported a profit of $3.45 billion for the first quarter of 2014.
It's not yet clear how widely Google plans to use Power-based servers, if indeed it plans to use them at all for anything other than testing. "A real server platform is also critical for detailed performance measurements and continuous optimizations," and to test the progress of the OpenPower alliance, MacKean said.
Google doesn't talk much about its infrastructure, but it's vast, expensive, spread around dozens of data centers globally, and sophisticated enough to do things like tell you within 0.58 seconds whether ketchup includes high-fructose corn syrup. With search, YouTube, Gmail, Maps, and many other successful services, winning a place inside Google's computing infrastructure is a major coup.
That's because Google spends a huge amount on capital equipment like servers and the data centers that house them. Google began a hardware spending spree in 2013 and in the first quarter of 2014 reported capital expenditures of $2.35 billion, its highest ever. No wonder server equipment makers want a piece of the action.
IBM has its own investments to make, too: its Power8 chip took three years and $2.4 billion to design. IBM introduced its Power8 processor and its own servers using the processor on April 23 amid a new push to answer the spread of servers based on Intel's Xeon chips.
Intel tapped into the "scale-out" movement that saw its chips used in data centers packed with countless lower-end servers rather than a smaller number of more expensive, centralized machines. Xeon chips, the highest-end members of the x86 chip family, got their start in lowly PCs but have accumulated better performance and reliability to handle the more demanding work of servers that typically run around the clock. Although chips using ARM Holdings' designs have given Intel a drubbing in the mobile processor market, powering almost all smartphones and tablets, Intel has maintained its stronghold in the server market.
Google is willing to branch out, though. In August 2013, IBM announced the OpenPower Foundation, an effort to bring some of ARM's relative technological openness to the Power server market. Google was among the founding members of the foundation; others include ARM chip licensee Samsung, graphics chip designer and ARM licensee Nvidia, and data center networking company Mellanox.
OpenPower gives Google some flexibility, but it's not trivial moving from one chip design to another. New versions of software must be recreated from the original source code, something that can be challenging especially for code that's hand-tuned or set up to take advantage of a particular chip's hardware acceleration features.
Although OpenPower makes it possible to license, modify, and manufacture Power chips, and to build their own servers using them, but IBM is still in the driver's seat when it comes to the core processor designs.
Google already designs its own x86 servers. IBM clearly hopes the OpenPower approach will help its processor designs fare better in mammoth data centers.