PORTLAND, Ore. — Intel's first Xeon system-on-chip (SoC) is twisting ARM's microserver ambitions, Charles King, principle analyst at Pundit-IT in Hayward, California told EE Times.
"Its not just Intel's first SoC in the Xeon family," King told us. "Its the beginning of a new era at Intel — expect them to move very fast in SoCs. We are going to see many more SoCs specifically designed to combat ARM microservers plus serve many datacenter functions."
Today's Xeon SoCs are designed for hosting and cloud services such as web hosting, memory caching, dynamic web serving, and warm storage. But the future Xeon SoC's to which Hayward refers will be optimized for storage and network-optimized products such as storage-area networks (SANs) and network attached storage (NAS), mid-range routers, wireless base stations and embedded IoT devices. About 75 percent of current designs-wins for the Xeon D are for network, storage and IoT designs, whereas microservers are under development at Cisco, HP, NEC, Quanta Cloud Technology, Sugon and Supermicro.
Intel's previous strategy to best ARM in microservers was to beef-up its Atom line with its second-generation Atom processor C2750, but no more, according to King — "Atom will become a consumer only SoC," he told us. The Xeon D-1500 family delivers 3.4-times faster performance per node and up to 1.7-times better performance per watt, which will also make it useful in high-end IoT devices that ARM cannot match, according to Intel. So far the only ARM 14-nanometer 64-bit core was made by Intel for Altera. Samsung has also shown a 14nm ARM 64-bit prototype core, but release no details or delivery estimates. The closest ARM has come to the Xeon D 14nm 64-bit cores are those made with TSMC's 16nm process, which are not due out until later this year.
The Xeon D-1500 was actually announced last year, but Intel has taken its time to make sure its ARM microserver killer was optimized for cloud, telco service providers and web hosters. Cast in Intel's 14nm 3D FinFET process, the Xeon D family will bring down the price of microserver SoC's to under $200.
To minimize the price, and reduce their power consumption to a minimum (~20 watts), Intel includes a second chip inside the same package as the Broadwell Cores to handle the SATA, PCIe G2, USB and SPI interface functions. All the other functions, including high-speed I/O using 24 channels of PCIe 3.0 for non-transparent bridging (NRB), are handled on the 14 nanometer 3-D FinFET chip in the same package.
The Xeon D, which is available in both 4-core and 8-core models, is also designed to extend Intel to the edge of the datacenter networks by providing a high-density, yet low-power, solution, according to Raejeanne Skillern, general manager of Cloud Service Provider Business at Intel. Besides microservers, Skillern also recommended the third-generation Xeon D family for storage, network and Internet of Things (IoT) applications.
"Xeon performance in a low-cost and low-power SoC for information and communications applications that will enable exciting new services to be delivered to business and consumer users," Chris Gianos, lead architect, recent told us at a press conference.
Two 10 Gbit per second Ethernet ports are provided and the Xeon D can directly address 128GB of main memory with built-in error-correction, hardware virtualization, and Advanced Encryption Standard-New Instructions (AES-NI).
Both the 4-core $199 models (D-1520) and the 8-core (D-1540) $581 models are available today. The rest of the Xeon D-15XX SoC models for network, storage and IoT applications will be available by the end of 2015.