Calxeda might have failed in its quest to sell servers built with ARM-based processors, but there are plenty of people who are confident the ARM architecture will find a home in the data center. In fact, AppliedMicro CEO Paramesh Gopi recently argued as companies like his begin selling brawnier 64-bit ARM processors (as opposed to the 32-bit processors often referred to as “cell phone chips”), the server industry is going to see demand for them blow up.
He made a convincing, albeit self-serving argument in a recent interview (before news broke of Calxeda’s demise), claiming that lots of private cloud deployments built on 32-but ARM are already coming online. The next generation of ARM design — the 64-bit one dubbed ARMv8 (which powers the AppliedMicro X-Gene platform) — has support from software kingpins such as Oracle (around Java development) and Red Hat (around Linux support).
Once Java support is complete, it will be pretty easy to port applications from legacy x86-based servers to ARM-based ones, said Michael Major, AppliedMicro’s vice president of corporate marketing. And, he noted, there’s already work being done to recompile certain applications that require access to the Linux operating system.
Large vendors wouldn’t be putting so many resources into making their compatible with the ARM architecture if it wasn’t “moving the market,” Gopi said. “We’re at the tip of the spear that’s making the market kind of fall over,” he said.
On the hardware front, AppliedMicro itself has server deals with HP and Dell, and, Gopi noted, their webscale businesses selling to companies such as Apple, Yahoo and Microsoft are growing like crazy. Even Facebook, with its fast-growing server footprint, is stoked about 64-bit ARM processors and has worked with AppliedMicro on a motherboard that fits Facebook’s custom server designs. Rumors have Google building its own ARM processors, as well.
The 64-bit architecture is critical to adoption because it includes more than the 4 gigabytes of addressable memory that many enterprise applications require. However, while many applications require more memory, many — especially web and data-processing apps — don’t often require the computing horsepower the market-dominating Intel Xeon chips provide. The idea of ARM in the data center is so popular in part (arguably entirely) because of how much a lower power footprint could reduce energy consumption.
This all sounds great for anyone building processors based on the ARM architecture but, of course, nothing is that simple — especially when billions of dollars are involved. x86 kingpin Intel has been working to make its Atom processors suitable for server workloads and, although they’re not quite as energy-efficient as ARM chips, it appears to be selling a lot of them inside Dell’s and other vendors’ cloudscale servers. Furthermore, AMD is pushing its SeaMicro fabric-in-a-box servers that contain up to 256 custom-designed Atom processors (although AMD is also working on an ARM-based server).
VMware, another force to be reckoned with in data center software, also isn’t on board with ARM processors right now.
And then there’s market education aspect of selling a new architecture, which includes overcoming the misnomer that ARM-based server processors are “cell phone chips.” Although they’re more energy-efficient than most of their x86 counterparts, they’re not exactly the “wimpy” cores that some have labeled them, Major said. “You can’t do phone chips for servers,” he said. “You need enough [power], but not too much.”