4G Core i7 boards: Best of both worlds
For High-Performance Embedded Computing (HPEC) applications with very tight space constraints, OpenVPX is the logical technology choice and 3U is the logical form factor. High-speed serial links like PCI Express and Gigabit Ethernet also make sense. But what about the processor?
It’s usually a good idea to choose the latest and greatest. An obvious candidate is Intel’s quad-core, fourth-generation (4G) Core i7 processor introduced in June. The new “Haswell” technology promises greater throughput and higher speed than its predecessor. It has approximately twice the performance per watt – a plus for embedded tasks – 30 percent better graphics capability, and double the data width for integer operations, compared with Intel’s third-generation technology. But what’s really new is its deep and enhanced parallel processing capability. 4G is a CPU and a General Purpose Graphics Processing Unit (GPGPU) combined.
What does it buy?
But what does it buy you in a 3U board? The new technology is thought to run sequential operations 20 percent faster than Intel’s third-generation processor. So traditional command and control applications like weapon launch – where the data sets are small but latency requirements are strict – would run faster on the new boards.
More importantly, however, the 4G Core i7 features a graphics processing engine with some 40 execution units, or cores – more than twice the count of Intel’s third-generation chip. The most obvious advantage to be gained from the improved graphics unit is improved displays. The wider integer pipeline can also improve image processing, which is often performed in an integer format. And users might be able to forgo a separate graphics processor, making room for other assets.
The 4G technology can enable qualitative as well as quantitative advances for CPU boards. The execution units can be orchestrated to perform many identical operations simultaneously across a data set. Such a procedure is common in signal processing applications, where raw input from a camera needs to be converted to some engineering unit. The new chip also supports the latest release of OpenCL.
A second area of parallelization in the 4G processor is that each of the four cores has two execution threads. So each core could perform two parallel operations if the process lends itself to multithreading. Memory bandwidth for the 4G version is 25.6 GBps, up from 21.3 for the 3G processor. This means faster thread execution, while applications using the new Fuse Multiply Add (FMA) instructions – which provide two operations per clock cycle – will run faster.
A third area of parallelization involves the 4G Core i7’s vector engine, known as Advanced Vector eXtensions (AVX), which adds a 256-bit integer instruction set, twice the data width of the 3G chip. Like a GPGPU, AVX performs Single Instruction Multiple Data (SIMD) parallel operations on a data set, and the wider data width of the 4G chip can double performance in some operations. But unlike a GPGPU, the vector engine involves much less latency at the outset. (The GPGPU’s initial latency is balanced out over a large data set.) For smaller data sets, AVX would have the edge over dedicated parallel processors. By allowing multiple operations to be executed simultaneously, the vector pipeline also would be an asset for signal processing. While the 3G device was a highly competent signal processing engine, the move from 128-bit to 256-bit makes its 4G equivalent more compelling.
The 4G’s technology could open doors to applications previously beyond the scope of SBCs. It can execute not only traditional CPU work but also sensor applications like sonar and signals intelligence on the same board without the price, power/heat, and real estate penalties involved with specialized processors. An example of an SBC implementing the new Core i7 is the GE Intelligent Platforms SBC326 (Figure 1), a single-slot, 3U OpenVPX board with an XMC site for signal processing adjuncts like high-speed analog-to-digital converters or FPGAs.
More bang for the buck
There are limits to how much processing and graphics capability can be squeezed onto a small SBC, and how much data can be run on and off the board. But the rapid evolution of processors – such as Intel’s Haswell with integrated CPU and GPU – gives customers a lot of bang for the buck.