Facebook, ARM, x86, as well as the future of the information center

Last week, Facebook announced a unique motherboard/daughtercard shape it dubbed “Group Hug.” Thanks to innovative function within the Open Compute Platform, the modern daughtercard enables CPUs from ARM, Intel, or AMD to be plugged into a single motherboard. At smallest, that’s the program — for today, the hardware interconnects continue to be inside the shape stage, because is the framework needed to manage disparate CPUs from several providers.

There are, easily put, big concerns left to settle. Adding different daughtercards to the same host to optimize CPU use is a remarkable idea, nevertheless writing software which knows how to manage the different assets continues to be a big task. In the rush to herald the advent of the bold modern era for host commoditization as well as the advent of ARM servers, certain substantial factors are being ignored.

Here’s the biggest: Contrary to what others have mentioned, mobile SoCs can not “remake the host world.” Wired tries to draw a parallel between Facebook’s experiments with turning off chip cache as well as the idea of substituting “brawny” CPU cores with “wimpy” ones. However it’s not which simple.

Take a consider ARM’s next-generation IP block for connecting about 16 processors, their caches, along with a range of extra program equipment, than the CCI-400 that’s currently shipping for the Cortex-A15.

The CoreLink 400 runs at half CPU speed

Here’s the host variant.

The CCN-504 doesn’t simply connect more chips; it ties inside as much as 16MB of L3 cache, dual-channel memory (with help for ECC), help for DDR4, 10gigE to 40gige, along with a host of different equipment. It’s an purchase of magnitude more complex than anything ARM has shipped earlier plus there’s a cause it’ll help both the Cortex-A15 plus future chips about ARM’s 64-bit extension, ARMv8.

IP blocks like this are because important to any ARM-based host because the underlying processor plus you’ll not see them inside a smartphone. Smartphones don’t need 16 processors with 8-16MB of L3 plus help for 10gigE, DPI, plus SATA. It turns out which when you begin adding server-class qualities to low-end processors, the finish products’ energy expenditure increasingly resemble every additional irrespective of the CPU architecture we commence from.

Intel plus (plus to several extent, AMD) can fight the trend towards commoditized hardware by trying to compete straight inside performance per watt plus by providing extra attributes which ARM chips don’t yet have. It’s true which ARM poses a risk to Intel’s host company, however, attracting a line from mobile phones to servers through an impressive box by Facebook ignores the real challenges ARM providers face inside struggling to break into the host marketplace.

What’s the future of the information center resemble? Complex plus evolving. ARM CPUs will have a element to play, nevertheless creating a full host ecosystem about these goods plus achieving mass-market penetration will take years. Facebook’s Group Hug platform might kneecap conventional host providers, however, it just threatens Intel when it can’t build inexpensive processors that provide greater performance per watt than its competition. At the Open Compute Summit last week, the providers about query were confident which their own solutions might confirm to become the ideal choice for powering next-generation servers.

AMD has the fruits of its SeaMicro acquisition, hot 64-bit ARMv8 processors inside the works, plus next-generation 28nm chips based about its Jaguar core launching this year, though there’s no info about whether Kabini plus Temash may show up inside servers. Intel has its own host Atom items plus can refresh those chips with 22nm processors based found on the initial quad-core, out-of-order Atom which debuts later inside 2013. ARM, naturally, has host vendors like Calxeda plus businesses like X-Gene, that plans to ship its own 64-bit ARMv8 design by the next half of the year.

The winner is decided by production, shape, plus scalability because much because CPU architecture. Historically, Intel has had a better handle about those issues than any additional provider found on the world. (See: Deliberate excellence: Why Intel leads the planet inside semiconductor manufacturing.) ARM could force Intel to innovate, yet the possibilities of the wholesale takeover are exceedingly little.

Now read: A tour of Google’s top-secret information centers