IBM PC

A year to build a computer

IBM had been selling computers to businesses for decades and had watched the small machine market grow without it. The closest thing to a personal computer in its catalogue was a portable unit that IBM describes as a $9,000, 50-pound machine. William Lowe, director of the General Systems Division laboratory in Boca Raton, Florida, argued that a computer for small businesses and homes could be sold for around $1,500. Frank Cary, then chief executive, gave him a month to produce a prototype and a year to get a product out of the door.

The project was code named Chess, and once Lowe was promoted the work fell to Philip Don Estridge. He ran the group outside IBM's standard procedures, the only reason the schedule held: IBM's normal cycle for a new machine ran four to five years. Hardware was Bill Sydnes, software was Jack Sams, the interface code was written by Dave Bradley, and Mark Dean and Dennis Moeller worked on the bus that let disk drives and printers talk to the processor.

The motherboard was designed in 40 days and a working prototype existed four months later. In April 1981 the design went to manufacturing.

Off-the-shelf by design

The team broke with IBM practice in choosing parts it did not make. The processor was Intel's 8088, a cost-reduced version of the 8086 that kept 16-bit registers inside and exposed only an 8-bit bus to the outside world. The narrower bus made the rest of the machine cheaper, and Intel could supply the volume. The 8088 addressed 20 bits, one megabyte, against the 64 kilobytes that most eight-bit chips of the period could reach. IBM ran it at 4.77 MHz, derived by dividing a 14.31818 MHz system clock by three.

The rest of the shopping list was just as external. The monitor came from IBM Japan, the printer from Epson, and the operating system from Microsoft, which had acquired 86-DOS from Seattle Computer Products. Roughly half the system, including the expansion bus, display, floppy interfaces and keyboard, was lifted from the System/23 Datamaster, announced weeks earlier at $10,000. Because so little of the machine was IBM's own design, the company held only a handful of patents on it, and never enforced them.

What $1,565 bought

The base configuration at announcement was $1,565: 16 kilobytes of memory on the board, a keyboard, and no display or storage at all, though the machine could drive a television. A system with 64 kilobytes, a single 5.25-inch diskette drive and its own display came to roughly $3,000, and a business configuration with colour graphics, two drives and a printer about $4,500.

Memory was expandable to 640 kilobytes in the conventional address space, the ceiling the architecture carried for the rest of the decade. The first 400,000 boards held 16 kilobytes soldered in place; a later revision started at 64 kilobytes and reached 256 kilobytes on the board. Displays were a choice between two cards. The Monochrome Display Adapter produced 80 columns of sharp text and nothing else, and carried a printer port. The Color Graphics Adapter drove graphics and text and could feed a composite television, though IBM's own 5153 colour monitor did not arrive until March 1983. The keyboard was the 83-key Model F, inherited from the Datamaster, and reviewers treated it as a reason to buy the machine.

The specification IBM published

The decisive act was documentation. The 5150 carried five 62-pin card slots on the processor's input and output lines, and IBM published the bus and memory map in full at launch, intending that other firms would build cards and write software. Those slots became known as the PC bus and later as the 8-bit ISA bus. Within a year thirty to forty companies were competing to sell memory expansion cards alone.

Reviewers noticed. Byte concluded two years on that the PC won on an 80-column screen, an open architecture and a good keyboard, which no rival had combined first.

Three operating systems, one winner

IBM announced support for three operating systems, suggesting an open field. PC DOS, licensed from Microsoft, shipped with the machine. Digital Research's CP/M-86, the system most buyers expected, was not available for six months and drew almost no orders; UCSD p-System also missed the launch. PC DOS became the standard by default, and Microsoft sold the same code to other manufacturers as MS-DOS.

Clones, and a standard that escaped

IBM kept copyright on the BIOS, the 8-kilobyte firmware chip that every program depended on, and assumed that would be enough. It was not. The Columbia Data Products MPC 1600, released in June 1982, less than a year after the PC, was the first legal compatible. Compaq, Phoenix Software Associates, American Megatrends and Award then produced clean-room rewrites of the boot code, either building machines around them or selling the firmware to others.

The effect ran both ways. IBM's own expectations were exceeded by as much as 800 percent, with the company shipping up to 40,000 machines a month at one point and more than 750,000 in 1983. Time named the PC its Machine of the Year. But the standard it created belonged to everyone: IBM's share of the personal computer market fell from roughly 80 percent in 1982 and 1983 to about 20 percent a decade later, and in 2005 the company sold the division to Lenovo for $1.75 billion. The 5150 itself was discontinued in April 1987. Nearly every desktop machine since has inherited its slots, its memory map and its firmware conventions, a strange fate for a computer designed in a year from a parts catalogue.