Steven Levy. Macworld. Volume 11, Issue 5. May 1994.
In 1990, Jack McHenry confronted the question that every manager of a computer-design team faces when a project is completed: What next? In the dizzying world of personal computers—particularly at Apple, where some Macintosh models last about as long as a QuickTime video—this is a perpetually vexing conundrum. Any engineer’s dream is to develop a machine that represents the company’s flagship, a computer that embodies not only marketplace realities, but also the company’s future. At Apple, many designers had that dream, leading to an implicit competition that demanded a combination of technical knowledge, soothsaying, corporate politicking, and samurai confidence.
Four years ago McHenry’s team was coming off a win—the Macintosh IIfx, considered at the time a paragon of blazing speed and breathtaking power. This costly beast was one of the first salvos in a fusillade of about a zillion new Macintoshes that came out over the next few years, an ascending spiral of more-powerful models based first on the Motorola 68030 microprocessor, then on the 68040. But some motherboard visionaries at Apple understood that this entire family of processors—beginning with the 68000 that was chosen for the original Mac—was essentially a dead end. In order to process computation-hungry stuff like multimedia, telephone, and voice-recognition applications, Apple would need a more powerful chip. A few people thought the answer was to switch to something that used an exotic technology called RISC.
RISC is an acronym for reduced instruction set computer. Regular microprocessors, like those on the 680X0 family, have rich instruction sets that execute many commands. The inventors of RISC figured out that by cutting the instructions to a very few, you could have a faster processor. Sure, in order to perform the instructions that are no longer hardwired onto the chip, you have to run many more instructions using the few you have. But when the smoke clears, things come out faster with RISC.
Still, in 1990 RISC was a radical departure, a promising technology used in very expensive workstations. In 1990 Apple already had a group working on RISC technology: the Jaguar group. Jaguar got its start from a project that put a RISC chip on a board that fit onto a IIfx (the GC board), and now its members were charged with designing a powerful RISC-based computer. The Jaguarians had decided that any Apple RISC computer would have to break cleanly from all previous Macintoshes. In their thinking, the first thing you would do upon switching from the Mac to a new RISC machine was to feed all your software to a neighborhood goat—who would munch on it happily, since it was garbage.
Because this machine would be powerful enough to do all sorts of heretofore impossible things, several technology groups at Apple allied themselves with Jaguar—groups working on projects like interfaces to telephones, audiovisual capabilities, adjustable keyboards, and the voice-recognition system called PlainTalk. By 1990, there were between 10 and 40 people working on Jaguar-related projects.
McHenry’s team, however, approached RISC from a different angle. On a group ski trip in March 1991, they made a decision to develop a RISC machine that would be a direct successor to the Macintosh family and would run the current software base. Considering McHenry’s attachment to the Mac, this wasn’t surprising. In 1984, McHenry had been an experienced 36-year-old Silicon Valley hardware gypsy who was blown away by the introduction of the Macintosh. “From then on, my goal was to go to Apple and design Macintoshes,” he says, and that year he began doing just that.
McHenry’s new project was code-named Cognac, in honor of a RISC pioneer with a surname identical to a particular after-dinner liqueur. The key people on his team included IIfx project holdovers Robert Hollyer and Jonathan Fitch. The latter was also an old Apple hand, having worked on the Lisa.
Cognac and Jaguar, of course, knew of each other’s existence. They worked in different buildings, but each kept up on the progress of the other. And each thought the other team was hopelessly misguided. “We thought Jaguar was a second step, not a first step [to RISC],” says McHenry, with some tact. Others are more frank: “We thought their approach was impossible,” recalls Jon Fitch. “We viewed Jaguar,” says McHenry, “as the evil empire.”
But the tension between the two RISC teams was small beer compared with the disquiet that these units sowed within Apple in general. It is easy to see why: there were literally thousands of people within Apple devoted to extending the current operating system—the world of the 680X0—into the next decade and beyond. If either of these two small groups succeeded in moving Apple into the world of RISC, the company would see an upheaval not experienced since the Macintosh overthrew the Apple II. Though little publicized, this drama was the real subplot of Apple’s development efforts in the early 1990s.
The main problem with switching to RISC, of course, lay in the deep investment Macintosh users had in their software. One would expect the vendors of high-volume applications like Microsoft Word or Aldus PageMaker to eventually port their wares over to any new Apple platform. But a lot of Mac software falls into the realm of folkware—quirky applications that may not have won a large audience but help to constitute the overall library that makes the Macintosh so valuable. At best, RISC versions of those applications would be slow in coming; at worst, they might never be attempted. If it couldn’t run existing software, a RTSC-based computer would be of limited value to Apple’s current customers.
While running Macintosh software was not a priority for the Jaguar team, the Cognac team had to figure out how to build a computer that could run both existing programs and the new generation of applications that would be created specifically for the RISC computer. This meant choosing between two alternatives: either using the RISC chip to emulate a previously existing Macintosh, or shipping a two-in-one computer—a RISC machine with a 680X0 Macintosh chip set alongside.
Emulation is no picnic. It often slows a machine down, and who wants to buy an expensive computer that makes your existing software look like it’s working underwater? The history of dual-processor machines is littered with failures—no matter how elegantly you package them, they wind up as costly compromises. “Dual processors are a nightmare,” says Jon Fitch. “So we were looking for a way not to include a 680X0 chip in the machine.”
The breakthrough came late in 1991 when the Cognac team discovered what became known as the 90/10 rule. As McHenry explains, “It turned out that in a typical Mac application, 90 percent of the computing time is spent on 10 percent of the code.” This meant that it was theoretically possible to do a very fast emulator—it might be possible to actually produce a machine with only the RISC processor. “We could leave out the 680X0!” says McHenry.
It was the Jaguar team that wound up shopping for the chip manufacturer to provide Apple’s RISC processor. After a whirlwind tour of every potential RISC provider, including MIPS and even IBM, the Jaguar team wound up making a decision: it would use Motorola chips. Apple execs decided the Cognac team would use the same chips. (The rejected suitors were disappointed, since an Apple adoption of a company’s RISC chip would mean a minimum tenfold increase in sales.) Both teams continued working independently. Cognac’s hardware design progressed, and in late 1990, the team reached what’s called gray screen (the point at which a prototype, its circuitry working, lights up a monitor). Still, the team knew that its efforts would be wasted if the Macintosh emulation was sluggish. “If there was a [speed] penalty in buying this over the 680X0 Mac, we couldn’t do it,” says Jon Fitch.
The man charged with producing the Cognac emulation scheme was an engineer named Gary Davidian. Before coming to Apple and working on various ROM toolboxes, he had been an experienced microcoder—a person who writes directly to the high-speed instruction sets on microprocessors. When Davidian joined the Cognac team, the Motorola decision had not been made, so every couple of weeks he was writing a new emulator, depending on the RISC chip du jour. It was not easy, making those chips pretend they were 68000s, but as he progressed he learned many tricks. By the time the group settled on Motorola in 1991, Davidian was ready to apply all his knowledge. By midyear, he had an emulator that ran inside a Macintosh LC box using a RISC chip—look, Ma, no 68000! The RISCified LC, which they called RLC, accepted off-the-shelf Macintosh software and ran it at speeds comparable to a Mac II.
“The RLC worked great—it blew away the company,” says McHenry. It emulated almost everything that ran on the Mac—not only the new things but some of the earlier software as well. One particularly triumphant moment was the Cognac team’s demonstration to the Jaguar team in late 1991. The Jaguar engineers attempted to break the emulator with offbeat software—they even dug up some ancient 400K floppies to stump the RISC machine. But the RLC maintained its charade that it was a 68000 computer. And when the RISC machine ran software written specifically for its processor, it churned out Mandelbrot fractals at dizzying speeds.
The success of the RLC prototype was a real milestone on the road to RISC. “Until then, people didn’t think it was possible,” says Wayne Meretsky, who was then a Jaguar engineer. “They didn’t understand that the limited instruction set could do all that.”
Deal with the Devil
As the Cognac team produced more prototypes—by 1992 they had an RLC running the Finder in a Macintosh IIsi box—Apple’s executives were busy with another matter entirely. John Sculley and his colleagues had begun negotiations regarding a joint venture with Apple’s former blood enemy: International Business Machines. Ironically, one of the things that first brought Apple and IBM together was the Jaguar team’s search for a RISC chip. Once that back channel was opened, it turned out that Apple and IBM officers had plenty to talk about (maybe they discussed how they hated Bill Gates). In the summer of 1992, after months of top-secret discussions, the companies made an announcement: they would work together on several projects, the most important of which would be the development, working together with Motorola, of a new microprocessor that would be the heart of new machines from Apple and IBM. The chip would be called PowerPC. And it would be a RISC chip.
Apple people called it, with a measure of mockery, The Deal of the Century. As far as Cognac was concerned, it meant at the least a retooling of its emulation scheme—using a chip that literally hadn’t been designed yet. Even more daunting, the team had to bring on software wizards to develop an entirely new operating system built around this vapor processor. The PowerPC itself would be a collaboration between companies with cultures so different that some doubted that anything at all could come from it. (Picture Roy Cohn and Alger Hiss sharing a soldering iron.) And all of this had to be completed to meet an incredibly tight deadline-January 24, 1994, the tenth anniversary of Macintosh!
What happened next? See next month’s column to learn how Apple met the challenge.