What Ever Happened to... Intel's Dream Chip?

by John C. Dvorak, February 1997

Anyone who was around in the early 1980s will recall the Intel iAPX432 chip: a processor that was one of the biggest flops in Intel's history, but one of the more fascinating chips the company ever designed. Intel intended this chip, and its subsequent progeny, to be the desktop computer chip for the '80s and '90s.

The iAPX432 appeared as the X86 architecture was coming to an apparent end with the 80286, as far as desktop computing was concerned. Even the numbering schemes for the chips changed as Intel devised the iAPX (Intel's Advanced Performance Architecture) moniker for its entire line of processors. The iAPX 432 was to be the flagship, with all other chips renamed accordingly.

The Ada Fad

Everything changed, though, when the 432 hit the market and turned out to be a dog. The 432 was simply too ambitious an undertaking. Even more interesting is the fact that the 432 was optimized to run Ada programs. A government committee had recently knighted Ada as the programming language to replace all others. And nobody, of course, paid any attention to this edict. Nobody, that is, except Intel.

It's hard to say when the Ada aspect of the chip came into play, since the design work began in 1975, years before the Ada fad. The idea was to design a chip that was object-oriented where typical linear access methods were not executed. All sorts of advanced features were included, which only added to the overhead and complexity. The first announcement of the 432 came in 1980, with the first shipments arriving in 1981. The processor was sold in four parts, which were released at various stages. The processor itself was made of two chips: a decoder/sequencer and an execution unit. A third chip was an I/O controller, and Intel later added a bus-interface unit and memory-control unit to the mix.

There were a lot of pieces involved in this chip design, but even today's Pentium Pro consists of two chips plus other necessary support chips. Curiously, the lead engineer for the 432 was superstar designer Fred Pollack, who became the lead architect for the Pentium Pro. While there are supposed to be no similarities between the 432 and any X86 chip, the multiple-chip aspect looks to be a carry-over.

Pollack was not the only superstar involved with the 432 chip. John Doerr (a famous venture capitalist), Dave Best (another venture capitalist), Casey Powell (chairman of Sequent), and others were involved in the design as well. Once the chip was released, the designers gave up on it as a product and moved on. It's believed that the 432 was abandoned after 1984, although supplies of the chip set may still have been available as late as 1993. The concept behind the chip remained alive, however, and slowly evolved into what is today's Intel 960 embedded processor.

The 960 began as a joint venture between Siemens and Intel, using 432-inspired ideas to create a general-purpose machine. Again, Pollack was the lead architect. It was quickly determined that the early 960 designs made it perfect as an embedded controller. During the various iterations of the chip design, Intel actually invented a 33-bit processor that was never manufactured, a curiosity I find amusing. Another amusing aspect is that floating around someplace is a 960 Ada compiler. (The same Ada optimization exists within this chip, but Intel doesn't like to mention it.) One Intel engineer, who wanted to remain anonymous, told me, "There's not much of a market for embedded processors that run Ada!"

Unlike the 432, the 960 has become one of Intel's hottest-selling chips, and few people realize its heritage, since it's most often associated with the 860 RISC chip. Oddly enough, it has no architectural connection with the 860 chip, according to Intel. (I'm sure that if it had been named the 432/II, it would not have been as successful.) Intel notes that the 860 was designed in California and the 960 in Portland, where the 432 was originally done.

The CISC Circle

The 432 has since been discussed in many computing circles and is considered "the most CISC" of all CISC chips ever invented, with the possible exception of a little-known processor called the Rekursiv, which was actually a processor board created to replace a VAX in a factory-automation application.

The issue of CISC and RISC rubs a raw nerve at Intel. Discussion ends rapidly as the company tells everyone that RISC was a marketing gimmick and that the term is almost meaningless. Intel's argumentative Will Swope poses the standard Intel query, "If the PowerPC has more instructions than an X86 chip, then how can it be a reduced instruction set chip?" According to the company, the only difference between RISC and CISC after boiling it down is that RISC has a fixed-length instruction set that uses a simpler decoder mechanism. Nothing else matters.

Intel has avoided discussing the 432, as it reflects badly on its string of recent success. If you go to the Intel Web site and look at its own history of great processors, the 432 isn't even mentioned. This is curious, since at the time of its introduction, it was considered the greatest chip the company ever designed, the one that would lead us to the 21st century. Of course old-timers like to chide Intel about the 432 to prove a point of vulnerability, and Intel would like the story to go away, I'm sure. I suppose this column won't help them much, will it?