Ahead of competition from ARM servers, Intel is putting more weight in the server space with a new Xeon D family of chips, which will be in systems next year.
Xeon D chips will be the first server chips based on the Broadwell architecture, and will go into dense servers starting next year, the chipmaker said at the Intel Developer Forum.
Intel has started shipping the Xeon D chips to customers for testing, said Diane Bryant, senior vice president and general manager of Intel’s Data Center Group, during a keynote on Wednesday.
The chips are targeted at hyperscale workloads, in which servers are consistently added to scale performance. Companies like Facebook and Google buy servers in the thousands to deal with growing workloads.
Xeon D is effectively a system-on-chip, and will combine many components to deliver power-efficient performance for Web-based, storage or networking workloads. The chip family will have I/O and controllers for networking and storage appliances.
Intel already offers the Xeon E3 and also Atom chips code-named Avoton to process online workloads. But the Xeon D chip will be faster and more power hungry than Avoton, which is based on an architecture called Silvermont used in mobile chips. The Xeon D chips will draw 15 watts of power and more, Bryant said.
The goal with Xeon D is to help customers pack more performance-per-watt in servers, Bryant said.
Intel previously projected Xeon D, code-named Broadwell DE, to ship as early as late 2014. But PC, tablet and server chips based on Broadwell have been delayed due to manufacturing glitches.
Bryant also said Intel wanted to provide chips that could be customized for specific workloads, which it already does for specific customers like eBay. During the keynote, Bryant said that Intel will allow FPGAs (field programmable gate arrays) to be built into the same package as chips, though didn’t point specifically to Xeon D. Intel has packaged FPGAs with server chips for years now.
Xeon D will gear up Intel for a battle with ARM, whose processor designs will be in servers next year. AppliedMicro and Advanced Micro Devices are making 64-bit ARM chips for dense servers, but the products have been delayed. Nvidia and Samsung have cancelled plans to make server chips based on ARM.
Intel’s chips already go into more than 90 percent of servers, and server makers like Dell have said that the chances for success of ARM servers are diminishing due to product delays. The ARM server chips are based on the ARMv8 architecture, and have integrated networking, storage and I/O controllers. Intel also has a head-start on software development over ARM.
Xeon D will also be Intel’s first server chip made using the 14-nanometer process, which will provide power-efficiency and performance benefits. The first chips made using the 14-nm process called Core M are already due to become available in tablet-laptop hybrids starting later this month.
Intel earlier this week announced the Xeon E5-2600 v3 chips, which are based on the Haswell architecture. The Xeon E5 chips have accounted for more than 80 percent of Intel’s server chip shipments. Intel’s Xeon E5 line will transition over to Broadwell sometime next year.