Choosing a processor a balancing act

Story

November 26, 2015

Charlotte Adams

Abaco Systems

Designers consider performance vs. flexibility, as well as thermal management for processors.

How does a designer choose a processor for a single-board computer? The answer depends on many factors besides a chip’s number-crunching prowess, including “macro” issues that arise from the outside environment, “micro” issues that are created by the chip itself, and intermediate issues relating to the board, box, and subsystem in which the device will function. Designers have to balance these factors and the interplay between them to select the best fit.

At the highest level the designer considers the likely physical and security challenges to the circuit. For example, can the chip be soldered to the host board to resist vibration and acceleration forces? Can it reject tampering and malware attacks? At the next level the designer considers factors such as the host board’s function, power budget, and size constraints. Will the board require graphics capability and can that capability be integrated into the processing chip? In some cases a chip’s versatility may trump its raw throughput. At the micro level the designer will consider factors such as throughput, power consumption, heat dissipation, and size.

Measuring performance

Perhaps the most important considerations in chip selection are comparisons of size, weight, power, and cooling (SWaP-C) versus performance tradeoffs at chip and board levels.

Compute performance is the primary requirement for a central processing unit (CPU), but performance is not an end in itself. More important is performance per watt: Is the CPU’s performance affordable in terms of electrical power consumed and heat dissipated? Is it electrically and thermally viable for a given application? Despite its sparkling benchmarks, would the chip have to be run at a much lower-than-advertised clock speed to maximize reliability?

Thermal-management issues are a major cause of electronics failures, and heat removal is the highest bar to performance growth in embedded applications. Board manufacturers are deploying cooling solutions involving miniaturized heat pipes, piezoelectrics, and specialized materials. Integrated circuits must also cooperate, however. Since most of the power they consume turns into waste heat, chips must provide the best possible processing rate per watt, as measured in GigaFLOPS or MIPS.

Embedded boards typically employ chips designed for the mobile or workstation market, depending on SWaP-C tradeoffs. Embedded computing often avoids server-class CPUs because they are big and hot and are usually socketed to the host boards. These traits, though acceptable in a data center, are anathema in a rugged environment, where size, heat dissipation, and vibration resistance are critical concerns.

Then there is the question of performance vs. flexibility. Do you go for conventional instruction throughput or parallel processing power? Or, within the constraints of embedded applications, do you want a combination of the two; that is, a chip with both conventional and graphics capabilities? Versatility is becoming increasingly important in applications such as electronic warfare, radar and sonar, command and control, intelligence/surveillance/reconnaissance (ISR), and signal processing.

Opportunity calls

Chipmakers, smelling an opportunity, have devised variations of high-end CPUs that maximize throughput but minimize die footprint, power consumption, and heat dissipation. New variants are solderable, thermally viable in embedded applications, and available in a range of configurations, allowing additional functions such as integrated graphics processing.

This convergence of supply and demand allows board designers to architect widely different products using a single family of chips, a fact which could translate to quicker time to market, faster upgrades, and longer lifecycles with lower costs to customers over time.

A product that considers performance per watt, thermal viability, and versatility is the GE Intelligent Platforms SBC328, a rugged 3U VPX single-board computer employing Intel’s Xeon E3-1505M v5 server-class CPU with integrated graphics processing, antitamper and information assurance features, and a vapor chamber-style cooling system. The board can process 358 GigaFLOPS at less than 50 W (Figure 1).

Choosing the right processor for an SBC involves juggling multiple factors such as compute performance, throughput per watt, and thermal viability, all in the context of larger concerns. Thorough analysis of these factors results in the optimal processor choice, most effective board design, best fit with the intended application, and lowest cost to the customer over time.

Figure 1: GE’s SBC328 rugged 3U VPX single-board computer is based on sixth-generation Intel core technology (“Skylake”).

(Click graphic to zoom by 1.9x)