You can’t go a month without reading about how some web giant like Facebook, Google or Amazon is trying to design its own silicon as a way to continue lowering the cost of computing. Adding to the trend AMD, which has played second fiddle to Intel for decades as the only licensee to the x86 architecture, has taken an ARM license.
This year, we expect to see a production-ready ARM-based server with a 64-bit instruction set that can compete with Intel in certain segments of the market. And it’s not just ARM. IBM has open sourced its Power architecture and has Google joining the OpenPower Foundation and designing a server (although it’s a long way from joining and building a prototype to actually deploying production servers).
The big story is that when it comes to webscale, companies are recognizing that general-purpose compute is, well, just too general-purpose. That’s why it’s not just the lower power consumption touted by ARM chips that offer value to these compute providers — they also are using the modularity that ARM can offer to build purpose-built silicon for different workloads and environments. It’s not just the threat of so-called micro-servers, but storage and networking that is being displaced.
And as the provider at the top of the heap, all eyes are on Intel, which is currently trying to mitigate its failure to break into the mobile market and relying heavily on the data center business to keep its revenue and profits up. The woman in charge of this division, Diane Bryant, met with me last week in San Francisco to discuss Intel’s plans for more modularity to give its customers what they want.
Bryant will be speaking at our Structure conference June 18 and 19 in San Francisco about Intel’s focus on the data center, how it plans to stay on top and what she sees for the future in terms of her clients’ needs. But she gave me a taste of what’s to come by laying out Intel’s plans to take general-purpose processing and make it specialized enough to keep her top clients placated.
Since 2012, Intel has been designing specialized chips for its biggest clients with over 30 shipping currently. For example, Intel created a custom Xeon E5 CPU for eBay that dynamically throttled CPU performance up or down based on the workload helping to manage spikes in demand. It’s also working with Facebook to figure out the optimal number of cores and the clock speed for chips running in the Facebook environment, which a marks contrast to 2009 when, at our Structure conference, then-Facebook VP of Engineering Jonathan Heiliger demanded chip firms innovate for the needs of webscale customers.
Bryant also hyped up Intel’s work on its Rack Scale architecture that crams computing, networking and storage in a rack unit, making that the new unit of compute in massive enterprise and webscale data center environments. When she hits the stage at Structure she’ll not only address the increasing amount of potential competition, but tougher questions around ensuring that the CPU holds its value in and the challenges ahead around encryption, scale-out computing and new workloads that Intel is designing for.
It’s easy to point to Intel as a lumbering giant that missed mobile and the importance of lower power consumption even in the data center, but it would be idiotic to think that it’s not doing its best at the CPU, hardware and even software level to ensure that its business continues to prosper as computing spreads to more places and does more things. You’ll get more details on that in less than a month at Structure, so register now.