RTC Magazine

On Thursday, January 7 of this year, according to one editor in the embedded computing space, it “rained” Intel Core i7 press releases. According to another editor, the military/embedded market seemed to be “turning backflips” on that day, forecasting what he described as a “tectonic shift” in the industry. January 7 was the date Intel announced no fewer than 27 new Core i3, Core i5 and Core i7 processors—originally codenamed Arrandale—and what followed was a veritable deluge of announcements of products based on those processors. According to one commentator, by the time of Intel’s CES press conference, no fewer than 400 PC designs and, more importantly, at least 200 embedded devices based on the Core i7 were ready for launch. GE Intelligent Platforms was one of the companies responsible for that deluge with the announcement of five new single board computers based on Intel’s Core i7 technology (Figure 1). Why the excitement?

For many years, embedded systems designers have been caught between two attractions in terms of choosing a processor architecture. On the one hand, Intel’s raw performance has always been compelling—but this often came at the cost of a hard-to-manage power envelope. Then there was the growing PCI/PCI Express infrastructure, and the opportunities it offered to create more cost-effective solutions. Add to that the fact that the Intel architecture supported not only Windows, but also, for example, Linux and VxWorks, and its attractions were undeniable.  

On the other hand, Freescale offered what seemed to be a better understanding of the need to offer solutions that made a trade-off between processor performance and power consumption/heat dissipation. The PowerPC architecture also offered access to AltiVec, the floating-point processing capability that is invaluable in sophisticated applications such as digital signal processing. And there was no questioning Frescale’s understanding of the need to ensure extended product lifecycles in line with the needs of the embedded industry.

The world, however, has changed. The single-minded zeal with which every processor manufacturer relentlessly pursued increases in clock speed and throughput has been tempered by a realization that those are no longer the be-all and end-all of processor design.

Why? Simply because the unarguable trend—whether in the PC market with laptops, followed by netbooks, followed by tablets and slates, or in the embedded market with military applications such as unmanned vehicles and man-wearable computing—is to put more and more processing capability into smaller and smaller environments where power usage needs to be conserved and heat dissipation minimized. Now, the discussion is routinely not about clock speeds, but about performance/watt, about the size, weight and power (SWaP) characteristics of a solution.  

The kinds of products Intel was designing for the mobile PC market were the kinds of products that the embedded market needed. When the company announced 12 new products specifically for the embedded market at CES—products that would have the seven-year lifecycle necessary for the embedded market—pieces of the jigsaw started to fall into place.

But extended availability wasn’t the only good news Intel brought to the embedded computing table at CES. Also announced was the fact that these new processors would include, among many other desirable features, ECC memory and a floating-point processing capability.

Floating-point processing capability is a fundamental requirement for many of the most sophisticated embedded computing applications, such as radar, sonar and sensor processing or in any commercial environment where digital signal processing is a substantial element, such as the growing interest in facial recognition for ATMs. In fact, the Core i7’s floating-point processing capability was not new, as such. GE Intelligent Platforms had benchmarked Intel’s earlier Penryn family of processors, which also featured floating-point capability, and found that it compared favorably with the embedded market’s incumbent supplier, Freescale, and its AltiVec processor. A significant attraction of the Core i7’s inclusion of an even more capable floating-point processor is that Freescale no longer supports the AltiVec processor.  

Beyond this, the Core i7’s Hyper-Threading Technology promises to substantially improve the performance of single precision FFT operations. Hyper-Threading Technology allows each execution core to function as two logical processors. Coupled with the ability to execute multiple instructions across the floating Arithmetic Logical Unit (ALU), each of the logical processors can execute multiple single-precision floating-point instructions per clock cycle—which has the potential to deliver a quite compelling boost in performance.

The provision of ECC memory is equally significant. Embedded computing applications typically rely on the very highest levels of data integrity in what are often mission-critical and even life-or-death environments. The majority of military/aerospace requests for tender, for example, specify it as a requirement—a requirement that boards based on the Core i7 can satisfy.

As with the floating-point processing capability, integration of ECC memory is not new for Intel—but it is new in a chipset designed with the performance/watt characteristics required by the mobile computing market. Previously, designers had to choose between higher power consumption/heat dissipation and ECC memory—or lower power consumption/heat dissipation without ECC memory.

Also integrated within the Core i7 chipset (Figure 2) is a powerful graphics capability which, according to Intel, will obviate the need for a separate graphics processor in an increased number of applications. Early GE Intelligent Platforms benchmarks, using the 3D Mark suite of tests, show a level of performance comparable with previous platforms that employed a discrete graphics processor. For example, ATI’s Radeon E2400 discrete GPU scores around 16,000 in the 3DMark tests—while the Core i7’s integrated graphics comes very close with a score of 12,000. That level of performance has made the provision of two DVI ports (rather than the previous single port) appropriate for GE Intelligent Platforms’ new family of Core i7-based products. Of course, graphics processing is also about number, variety and bandwidth of graphics I/O that can be supported, so the requirement for discrete graphics capability will still exist—but for many applications, the capability provided by a single board computer based on the Core i7 will suffice.  

In its announcement, Intel has loudly trumpeted the 32 nanometer manufacturing process used to fabricate the Core i7 chip set. This shrinking has contributed significantly to the higher degree of integration that means that three devices have been replaced with two. The front side bus (FSB), for example, is no longer a connection across the printed circuit board—it is integrated into the chipset. Not only does this integration result in higher speed performance and/or lower power consumption, but it also frees up valuable board real estate—a precious commodity to single board computer manufacturers looking to add the maximum possible amount of functionality. In the case of the VR12, for example, it has allowed the provision of two XMC sites, where otherwise only a single site would have been possible. This provides significantly more flexibility to users in configuring the capabilities of the board and minimizing overall slot occupancy. The additional real estate has also enabled the VR12 to feature onboard flash memory, as well as a plethora of connectivity options. 

Another important benefit of the Core i7 family of processors is just that—that it is a family. Intel has recognized the importance of being able to trade processing performance against power consumption/heat dissipation: the Core i7 is offered at power consumption (TDP) levels of 35 watts, 25 watts and 17 watts—with clock speeds of 2.53 GHz, 2.0 GHz and 1.06 GHz respectively - enabling customers to choose the performance/watt ratio most appropriate for the planned application. Early indications are that the Core i7 will offer either more processing performance per watt compared with earlier products (estimated at around 20%), or will offer lower power consumption per unit of processing performance than its predecessors.  

Much of the coverage of the Intel Core i7 announcement has focused on those features to which Intel has given interesting names, such as Turbo Boost and Hyper Threading. Intel Turbo Boost Technology is said to automatically accelerate performance, temporarily increasing clock speed depending on workload, while Intel Hyper-Threading Technology is claimed to enable smart multi-tasking by allowing each processing core to run multiple “threads.”These are, unquestionably, some pretty cool technologies—especially for students of microprocessor design—but, for the embedded market, they all merely contribute to the higher levels of performance/watt that the Core i7 is expected to deliver. They are all also illustrative of the fact that companies like Intel are looking far beyond simply ramping clock speeds and reducing die sizes in order to achieve worthwhile performance gains.  

Intel’s Core i7 announcement is therefore significant to the embedded market for a host of reasons.  It confirms Intel’s renewed commitment to a market that the company says is now worth a billion dollars a year to it. That commitment is illustrated by its continuing focus on performance/watt rather than pure performance, and the range of performance/watt options it offers; by its inclusion of the error correcting memory capability that means little to PC users but everything to many embedded users; by the inclusion of a highly capable floating-point capability at a time when the embedded market’s historically most significant supplier of processor technology no longer does so; by its inclusion of a superior graphics facility; and by the degree of integration, which has made board space available for manufacturers to increase the functional density of their offerings.  

Freescale is, and is likely to continue to be, a significant supplier to the embedded computing market. However, with its Core i7 announcement, Intel has clearly placed an important marker in the sand—and that can only be of enormous significance to vendors like GE Intelligent Platforms and to users of embedded computing worldwide. 

GE Intelligent Platforms.
Charlottesville, VA.
(780) 401-7700.
[www.ge-ip.com].