Intel (NASDAQ:INTC) issued a press release on March 9 announcing its next generation low-power data-center chip family known as Xeon D. Many of the details of this chip had already been leaked by CPU World, among others, but it's always interesting when a previously unannounced product finally gets official.
Xeon D, according to Intel, offers up to 1.7 times the performance per watt of its prior generation, Atom C2750 for low-power micro-servers, and up to 3.4 times the performance. The company says that initial four- and eight-core products for micro-server workloads are available today, with additional products -- optimized for Internet of things, network, storage, and micro-server workloads -- coming in the second half of 2015.
These parts integrate between four and eight Broadwell CPU cores (Intel's high-performance cores), 2x10Gb Ethernet controllers on die, and 24 lanes of PCI Express 3.0 connectivity all on the main die built in Intel's 14-nanometer technology. On a second chip found on the same package as the main die, Intel includes Serial ATA 3.0, USB 3.0, and USB.
Intel says that these initial products will be suitable for dedicated Web hosting, memory caching, dynamic Web serving, and warm storage. The company also claims that future Xeon D products -- presumably with dedicated accelerators for more specialized workloads -- will be able to go after things like wireless base stations, entry level storage area networks and network attached storage appliances, and edge routers.
It even managed to get the "Internet of things" buzz-phrase into the press release, claiming that these processors should also do well in "industrial IoT devices."
Over 50 systems in design
According to Intel, there are "more than 50 systems currently in design" based on these new Xeon D processors. The company cites major industry players, such as Hewlett-Packard, Supermicro, and Cisco -- among others -- as currently designing systems around the Xeon D chip family.
Of course, while it seems as though there are plenty of systems being designed around these chips, it'll be interesting to learn what portion of Intel's server chip sales ultimately end up Xeon D-based.
So are we not even talking about Atom?
One thing that I've found very interesting is that Intel made direct performance comparisons of the new Xeon D to the prior-generation Atom C2750. Although Intel has a part codenamed Denverton on its roadmap that is expected to be built on the company's 14-nanometer process and based on a newer Atom core than what's in the C2750, not much has been said about this part.
I suspect that Intel will continue to develop Atom-based parts for the datacenter, but I wonder if these parts will ultimately be second-class citizens to the Xeon D family of parts. In fact, as Intel becomes more adept at developing its high performance core IP to be more modular (like the Atom cores are today), then it wouldn't be surprising to see Intel offer a wider portfolio of high-performance core based products for different workloads.
The Atom cores are smaller and likely more efficient at lower performance and power points on the same process, so my expectation is that Intel will still have chips to service those workloads (especially ones where having a huge amount of smaller cores is preferable to a lower number of faster cores). My bet, though, is that the majority of the workloads will ultimately benefit from fewer Xeon-class cores rather than a larger number of Atom-class ones.