The Need for Speed?
by
Tom Barnum
VersaLogic Corp.

Determining the correct processor for use in an embedded application requires careful evaluation of a variety of factors. Some of these factors include:

Raw performance - How many million of instructions per second (MIPS) are required to complete the software cycle in the allowed time?
Operating environment - In what temperature range will the system be operating?
System considerations - Is there a power consumption limitation that must be met?
Production life - How long will the system be in production?
Cost - What processor most closely meets the cost/performance requirements of the application?

Beyond these considerations are a host of additional concerns, which must all be addressed to make the best choice for a successful product development.

Long-Term Availability
One of the fundamental precepts of the chip industry is Moore’s Law. Gordon Moore, founder of the semiconductor manufacturer Intel, predicted in 1968 that the number of transistors on a chip would double every 18 to 24 months. This increase in power provides the impetus for growth in the computer industry. As technology continuously changes, the need for new hardware and software to support the newer and faster chips has fueled economic expansion and created a turbulent computer industry.

Moore sees an end to processor speed growth at some point, ("We can’t exceed the speed of light" - Gordon Moore, 01/14/98) but doesn’t project when this might occur. Continued improvement in technologies to photolithographing circuit paths (moving from ordinary light waves to ultraviolet radiation and x-rays), the introduction by Intel of technology that assigns 2 bits of data per memory cell (instead of current 1-bit/cell limitations), and IBM’s use of copper (instead of the poorer electrical conductor, aluminum) suggests that processor speed will continue to follow Moore’s Law for the next 5 to 10 years.

Moreover, the long-term availability of a given processor is questionable since chip manufacturers continuously displace recent chips with even newer technology. 1997 saw the introduction and rapid deployment into not only mainstream PC but embedded markets of Intel’s MMX and the Pentium II processor supplanting earlier versions of Intel’s processors. The market structure of the industry requires CPU manufacturers to obsolete their own products in a relatively short period to maintain profits.

OEM users often make a tremendous investment in development and marketing of high-end machinery. This investment will pay off only if they can continue to sell their product for a number of years. In the past, the manufacturers of embedded systems have ensured that their board-level products would be available for a long period of time. However, the rapid development cycle of CPUs for the desktop market (and the discontinuation of all previous CPU types) have left OEMs and embedded system manufacturers with nowhere to turn.

Both Intel and AMD have recognized the need for long-term product support in the industrial computer industry, and have moved the stable 486 class processors to their "Embedded" Divisions to help fill this need. These products are currently available off-the-shelf from the standard distribution channels, and will be produced for years to come. The Pentium (non-MMX version) has technically been moved to Intel's Embedded Division, but it is not yet readily available. Some of the Intel Pentium MMX and none of the Pentium II CPUs have been moved to Intel's Embedded Division. Although most of these chips are available currently, this situation changes rapidly (every 6 to 9 months). As newer CPUs are introduced for the desktop market, production of the current chips are discontinued in short order. However, the availability of a particular Pentium CPU changes every 6 to 9 months as Intel pushes faster desktop CPUs into production, and eliminates slower ones. The Intel and AMD 486/586 CPUs are stable products that are currently, and will continue to be, available to the embedded marketplace for many years.

Typical chip set availability for industrial applications should be 5 to 10 years, compared to the less than 2 year life cycle of desktop components. With the rapid changes in processing power, maintaining an available supply of processors and chip sets has become problematic for board level manufacturers, which in turn has created doubts in the minds of many OEMs as to what products can be expected to be available to meet their long-term requirements. As engineers design their systems for "optimal speed" rather than "high speed", chip manufacturers will be counted upon to maintain a more stable supply of products.

Power Consumption
The faster CPU chips and their requisite cooling systems drive up the system’s power consumption. As desktop CPUs approach 300 MHz + clock rates, more and more system resources are being allocated to deal with the new higher power requirements. Special on-board power supplies are needed to create the exacting voltage / current requirements of these new processors. These leading edge designs are always pushing the power and cooling system requirements to the limits. Embedded designs frequently encounter power limitations for CPU usage, and the combination of a high-wattage processor and electrically-powered cooling devices can easily consume 4 amps or more.

Operating Temperature
These power hungry chips also drive up the temperature in the system, which is a particularly important consideration for embedded developers. When the heat output from the CPU combines with the high industrial range temperatures frequently found in embedded system environments, expensive processes can be required to keep the CPU cool enough to run. For instance, a 486 CPU might dissipate less than 5 Watts while a Pentium II might dissipate over 25 Watts. Only a simple heat sink is required to operate the 486 up to +70 degrees C, while an elaborate fan / heatsink is needed to operate the Pentium II at normal room temperatures. Some of the newest, highest speed CPUs actually have temperature monitors on them so they can shut down briefly when they get too hot. Not a good idea in a real time environment!

Make the Right Choice for Your System
As we approach the Millennium, engineers are faced with a dizzying and diverse array of embedded architectures - VME, Compact PCI, PC/104, PC/104-Plus (PC/104 with PCI connectivity), STD and its progeny, STD 32. Driving this diverse set of platform types are high-tech companies (both hardware and software) promulgating the need for speed. In the industrial / embedded computer market, blinding speed is usually not the most critical aspect of system performance. Speed is important for software developers and players of video games. However, virtual reality programs, immersive 3-D worlds, graphics blasters, and video-based visual computing capabilities are not considerations for a typical embedded system OEM.

Of course there are systems that must have the speed found only in the latest and greatest processors, but many systems would be better served with more mature, established processors. Considerations such as longevity, power consumption, operating temperature - not to mention price - are drawing many system designers away from the bleeding edge of technology back to proven performers that meet all of their needs.

Sources for charts:
AMD x86 data sheets and application notes
Intel x86 data sheets and application notes
Aavid Heat sink data sheets

MMX, Pentium and Pentium II are Trademarks of Intel Corporation, all rights reserved.