Getting the IBM PC -- or PC clone -- with the right 'bus' may be critical to what you can accomplish in the future
When you are managing a company, you barely have time to catch up with what you did yesterday, let alone look to the future. But if you are planning for growth, and if some of your work is based on IBM PCs or clones, there are some hardware developments that you should know about.
For the past six years, hardware (as well as software) for most PC models was compatible, so your computer system could grow right along with your business without too much ado. If you needed more memory or wanted to connect to a local area network, for example, you or your dealer simply plugged in accessory circuit boards inside the computer, and that was that. But the comfortable period of hardware compatibility is now ending. PCs have split into several incompatible and semicompatible variants, and planning will be more difficult. The changes will first affect people who need high-performance microcomputers, but eventually nearly everyone will feel the pressure.
The driving force behind these changes is the push from speedier and more sophisticated applications that are under development now. These new applications will require a high-performance bus -- the internal set of wires linking key parts of a computer. Along with the central processing unit (CPU), the bus defines the limits of performance in a microcomputer -- how much information can be moved and how fast. Whereas earlier PC buses accommodated a single central microprocessing chip, high-performance buses will support multiple processors, enabling a PC to handle complex tasks more quickly. So if you are looking ahead to sophisticated applications for your company, you'll want a computer with a high-performance bus.
At the moment, the only available high-performance bus has been launched by IBM as part of a new hardware design called MCA (microchannel architecture). In addition to making technical improvements, IBM developed MCA to stem the tide of competing PC clones; it is asking clone makers for licensing agreements or royalty payments on MCA, something it did not do with its earlier buses. To outflank IBM, nine clone makers have in turn banded together to develop a royalty-free, high-performance bus called EISA (extended industry standard architecture). The trouble is, models with EISA will not be available until late 1989. (See "The Bus Stops Here," page 2, for a rundown of PC buses.)
MCA and EISA are completely incompatible; an accessory circuit board designed to be plugged into one bus will not fit into the other. MCA is also completely incompatible with earlier IBM PC buses; EISA will be partially compatible. Fortunately, IBM PC software is not tied to a particular bus, so MS-DOS, OS/2, and Xenix/UNIX programs will run on all PC buses and bus variants.
So what should you do if you are in the market for a PC and you want to use it into the 1990s?
Although MCA is already available, high-performance accessories that can take advantage of it are not. For today's tasks, MCA thus offers little advantage over the older AT bus. MCA is also expensive, in part because it is available from only a few vendors. An 80386 microcomputer with a modified AT bus performs as well as an MCA model -- at least for the time being -- and most 80386 micros are much cheaper. You can install many of the older PC accessory boards on an AT bus, but this bus will not support future additions, such as multiple processors.
EISA could offer the best of both worlds -- high performance and backward compatibility with older accessory boards. But EISA computers do not yet exist even in prototype. And backward compatibility will be an advantage for only a short time; most boards sold today will be supplanted within a few years by more advanced boards. Although the first EISA microcomputers are likely to be nearly as expensive as MCA models, competition among the clone makers will probably drive down EISA prices more quickly than MCA prices.
If you want a low-cost 80386 machine right away, you face a tough choice because current designs will soon be eclipsed by EISA. The most reasonable strategy is to buy only the lowest-priced 80386 models now and plan to reassign them within the next two years to less demanding tasks. At that point, you can invest in EISA computers for your most important work.
You can avoid some of the headaches by buying accessories as external products whenever possible, rather than as internal boards. High-speed modems and fax modems, for example, can be purchased as independent units that will connect to any microcomputer, whatever the bus.
Your buying strategy will also depend on how your company expects to use PCs over the next few years. Here's a timetable for the more sophisticated applications under development, which will require a high-performance bus:
* Local area networks (LANs). LANs will be one of the first applications to benefit from a high-performance bus. One important component of most networks, the file server -- the microcomputer that supplies information to all the other computers in the network -- can perform more effectively with a network processor in addition to the CPU. The first LAN processors should be available in 1990, so if you are buying a computer for network file serving, be sure it has a high-performance bus. A LAN processor is less important for the other computers on the network.
* Computer-aided design (CAD) and architecture. If you use CAD software today, you know that every time you modify a design you have to wait a long time while the computer redraws (that is, recomputes) the image, particularly if the image has many colors or shades of gray. Graphics processors greatly speed up this process; such processors are already available in some video boards, but future versions running on a high-performance bus should work much faster.
* Mainframe access. If you need your microcomputer to communicate with a mainframe, a communications processor can improve the link by translating microcomputer requests for information into a form readily understood by the mainframe and vice versa. Communications processors will replace the simple links now available, in which the micro serves as a "dumb" terminal to the mainframe.
* Voice messages. A new class of software and hardware will record and play back voice messages directly on your computer. Such recordings will become a form of voice mail, traveling over computer networks as electronic mail does today. A voice message could be self-contained ("We'll meet with the new CEO for lunch on Monday"), or it could travel with and annotate a computer file, such as a spreadsheet ("This analysis assumes sales will meet the projections Peter described at yesterday's meeting"). Some products offering voice annotation -- Wang Laboratories Inc.'s Freestyle, for instance -- are already available; they will blossom along with fast networks and sound processors.
* Simulation. Computer simulations of real-world affairs -- such as econometric models of the garment industry or a mock-up of an assembly line -- require prodigious computing power. Today such simulations can be done only on a mainframe computer, but auxiliary processors will make them possible on a micro. A fast processor designed to manipulate mathematical arrays, for example, could compute mathematical models much faster than even the math coprocessors that are common today.
* File and database processing. Software of the mid-1990s will construct automatic links among files. When you type someone's name to start a letter, for instance, your software -- behind the scenes -- will automatically look up all previous correspondence with that person as well as your correspondent's mailing address and any information related to the topic of your letter. Maintaining automatic links like these takes more computing power than the CPU can effectively supply. Instead, a specialized database processor will independently search through and match text, automatically updating its own dictionary and lists of synonyms.
Eventually such features will be so compelling that all microcomputers will support them, and the definition of a high-performance bus will shift upward once again.
THE BUS STOPS HERE
A guide to the various buses on the market
Buses have evolved to keep pace with the increasing speed and sophistication of central processing units and computer software -- starting with the first 8-bit designs (so called because the bus could transfer 8 bits of information at a time) to 16- and 32-bit designs.
* The XT bus. This is the original 8-bit bus introduced in the IBM PC in 1981 to support computers based on the Intel 8086 and 8088 CPUs. The bus quickly became a standard, and hundreds of companies produced hardware circuit boards that could be plugged into it. Most low-end PCs and compatibles, including the IBM PS/2 Model 25 and some Model 30s, still use this bus.
* The AT bus. This is a 16-bit modification of the XT bus that was created simply by adding an extra connector to the original XT bus. IBM first introduced it in the PC/AT in 1984. Nearly all desktop machines based on Intel's 80286 chip and produced today incorporate the AT bus. IBM still uses the AT bus in the PS/2 Model 30/286. The AT bus is also used with a hybrid CPU, the 80386SX, which is found in machines such as the Compaq Deskpro 386s. Most circuit boards for the PC/XT will work on an AT bus, except for memory boards, disk-drive controllers, and other boards that require fast information exchange with the CPU.
* Modified AT buses. Virtually every 80386 microcomputer available today, except for IBM's, uses a proprietary 32-bit modification of the AT bus. These modifications are all mutually incompatible, although many accessory boards for XTs and ATs work in these computers. Because there are so many of these buses, no advanced accessories will be developed for any of them.
* MCA (microchannel architecture) bus. For all of its 80386 models and some of its 80286 models, IBM broke away from the XT and AT buses and developed a completely new design. MCA is better engineered than the earlier buses, and accessory MCA circuit boards are "self-configuring," which makes them easier to install; you can plug them in without having to set a complicated series of switches. IBM is the dominant supplier of MCA computers, although a few other companies, such as Tandy Corp., also make them.
* EISA (extended industry standard architecture) bus. This is in effect a standardized 32-bit modification of the AT bus, which should eventually supersede other modified 32-bit AT buses. Like MCA circuit boards, new EISA boards are supposed to be self-configuring. The EISA bus should also accept many old XT or AT boards, although these would not be self-configuring. At present, the EISA specification is unfinished, and no hardware exists. EISA backers expect the first computers by the end of 1989.