Earlier this week we discussed
new details on AMD's
two next-generation cores, Bobcat
. AMD built the two processors to fill two distinct markets, with Bobcat focusing on netbooks/notebooks, and Bulldozer in servers, workstations, and high-end desktops. It's now rumored that AMD is investigating whether or not it could make a decent bit of scratch by launching Bobcat-powered server processors. Don't laugh—it makes more sense than you think.
The server industry has embraced virtualization as a means of improving utilization and overall efficiency, but it's not the option. There are various reasons why—in some cases, high-end x86 processors chew through too much energy per virtualized machine to scale effectively, standard servers may not have the customer's preferred allocation of resources, or there may not be a well-suited x86 server chip that fits the desired workload and price point. Over the past two years, a type of server design called physicalization has sprung up to meet this need.
On paper, Bobcat looks to have what it would take to make a decent low-end server chip. Power consumption and core power gating will have to be evaluated in real-world scenarios.
Physicalization turns virtualization on its head. Instead of using big iron to simulate multiple smaller systems, physicalization connects a number of diminutive chips and boards within a single server rack. This approach still allows for virtualization, assuming each individual chip is at least a dual-core processor, avoids creating hot spots, and can, depending on the workload, offer similar computational throughput/scaling for much less power. As it turns out, more isn't always better-a set of Westmere's may provide best-in-class performance, while eating an entire server room's worth of power (or exceeding available air conditioning capacity) in just a few racks.
One reason neither AMD nor Intel are particularly thrilled about physicalization is the fact that the vendors pushing into this space have often used consumer hardware to build out their products. Aside from the reliability concerns that raises, neither company has nearly the margin on consumer parts than they do on server hardware. On the other hand, neither company wants to see the other pick up part of a market on the sly. There's no word on what Intel might do if Bobcat parts start gaining real traction in a low-power/energy efficient server, but we expect the company would answer with some form of ultra-low-power Xeon part, rather than tap Atom for a server project.
Bobcat doesn't need to drop into physicalized servers in order to potentially compete; there's always the possibility that companies will opt for small Bobcat blades or other microserver buildouts. The biggest question, of course, is whether or not the upcoming chip can deliver enough performance to entice OEMs--but there's reason to think it can. Even Atom, after all, caught the eye of certain companies interested in building low-end servers. Intel adamantly refused to build servers around Atom, but if that diminutive processor could handle any sort of server load, (and vendors like SeaMicro have proven it can) Bobcat will as well.
AMD may be exploring this option, but don't expect them to throw a lot of weight behind it initially. Bobcat and Bulldozer will still generally end up in their target markets. If Bulldozer looks like it'll hit later rather than earlier, or if AMD believes there's a definite demand for this sort of low-power processor in a server environment, it'll move quickly to put an 'official' Bobcat server platform in play, rather than allowing consumer parts to drift into that space after additional OEM validation.