PORTLAND, Ore.—Intel has been touting its server-class Xeon processors as an embedded solution platform for advanced pattern recognition tasks, for instance inside the receiver for radar, video, audio or other sensor signals. Today those signal processing tasks are assigned to ground-based computers necessitating a high-bandwidth connection between it and the sensor. Putting a Xeon processor in a ruddegized embedded platform enables avionics, space, surveillance and military pattern recognition tasks to be performed in the field rather than back at the data center, powering applications like detecting IEDs (improvised explosive devices) with a Xeon inside a Hummer.
"We are now able to bring two very large server-class Jasper Forest Xeon processors into an embedded form factor, which has never been done before," said Shaun McQuaid, senior product manager at Chelmsford, Mass.-based Mercury Computer Systems Inc.
Mercury's Xeon embedded computer houses two quad-core Jasper Forest Xeon processors for eight cores in total, squeezed onto an OpenVPX module that is already being put to use in high-end radar, multi-sensor electro-optical/infrared (EO/IR) and other embedded applications courting airborne-caliber SWaP (space, wattage and performance). Smart avionics applications that offload target recognition to the sensor include Predator drones, ship radars, satellite telemetry and mobile bomb detectors mounted on Humvees.
OpenVPX module squeezes two Intel Xeon processors (brown) each with four cores and six gigabyte (white) into an inch-high 6U rack.
The real story, however, is that it was not as easy as just plugging an Intel Xeon processor designed for a server farm into a mil-spec board—since avionics boards need to have their chips soldered down. In fact, shoehorning two 48 watt quad-core Xeon processors with 12 Gbytes of memory into a one-inch high, 10.5 inch wide standard avionics 6U rack without melting it down required Mercury to invent three new technologies: a way to solder down the embedded processor that Intel designed to be plugged into a socket; a way to get the heat out; and a way to translate its PC-oriented interconnection architecture into the RapidIO architecture used by embedded systems for signal processing, networking and communications.
To secure the Xeon processor to its board, Mercury invented a way to solder down land-grid array (LGA) as if it were a ball-grid array (BGA). To dissipate the heat, the company invented a thermal technique that enables heat to be sinked both above and below the Xeon processors. And to translate the built-in PCIx I/O on the Xeon into the embedded RapidIO standard, Mercury invented its protocol offload engine technology—POET—a field-programmable gate array (FPGA) that bridges between the Intel processor's native switch fabric (PCIx) and serial RapidIO.
Embedded pattern recognition can be performed by two quad-core Xeon processors (pink) each with 6Gbyte of RAM (blue) and and FPGA (brown) to offload PCIx to RapidIO and Ethernet.
There are quite a few embedded boards utilizing Intel Xeon Processors in multiple form factors ranging from COMExpress to ATCA. Check out the Intel Embedded Alliance directory for a list: http://www.intelcommsalliance.com/kshowcase/view/
Mercury claimed this was the world's first embedded computer to utilize Intel's Xeon processors. The biggest step that had to be made was figuring a way to solder down the Xeon, which is intended for a socket. There may be some less ruggedized applications that can use the Xeon in its socket, but still be classified as "embedded," but I couldn't find any. Do any of you guys (or gals) know of embedded applications of Intel's Xeon? (I asked Intel for a list, but got no response by press time.)
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.