Demonstrating that x86 still has significant shortcomings when compared to other approaches to processor architecture design when manufacturing and market segment effects are normalized is one thing. It is quite another to explain how a different instruction set architecture could ever overcome an economic and process technology playing field that is so strongly tilted in favor of x86 as to approach vertical. Not only does a candidate to replace x86 have to overcome a huge financial and credibility barrier to entry but it needs to go well beyond mere parity to demonstrate a deliverable benefit substantial and sustained enough to justify change.
Let’s start by examining three failed attempts to replace x86 on the desktop and look at the lessons they hold:
In April 1991, Compaq, Microsoft, DEC, MIPS, and a collection of smaller computer companies banded together to form the Advanced Computing Environment (ACE) group. Their intention was to replace the x86/IBM PC with a new desktop standard using RISC processors based on the MIPS architecture. Due to a combination of MPU development delays, internal rivalries, and intense psychological warfare and economic pressure by Intel, the ACE group eventually faltered  . The consequences for many of the principals involved in ACE were as quick and final as the reckoning at the end of a Martin Scorcese film. Rod Canion was ousted as CEO of Compaq and MIPS Computer Systems went into a tailspin that didn’t end until it was acquired by SGI.
Apple/IBM/Motorola (AIM) Alliance
In late 1991 Apple Computer, IBM and Motorola announced an alliance to bring RISC to the desktop . Motorola and IBM agreed to jointly develop a family of processors based on a modified version of IBM’s POWER architecture to replace the fading 680×0 CISC processor line in the Macintosh. At its peak, the PowerPC era boasted a thriving market for Mac clones and ever faster speed grades of 60x processors that rivaled the best x86 MPUs in speed and cost. However Apple was unable to withstand the competitive pressure exerted by a myriad of low overhead Mac cloners  and with Motorola itself preparing to join the party it abruptly pulled the plug. The combination of the loss of price/performance competitive Macintoshes, the inability of IBM and Motorola to keep pace with P6 based x86 MPUs, and the corrosive effect of hard core, outspoken Macintosh supporters increasingly hostile to mainstream IT decision makers ended any chance that AIM could challenge the Wintel hegemony . As a result the Mac family went from 8.3% of the world PC market in 1994 to less than 3% today.
In late 1996 Digital Semiconductor, the chip manufacturing subsidiary of DEC, announced the Alpha 21164PC . This was a low cost, stripped down version of the performance leading 21164A (EV56) intended for use in PCs running a native Alpha version of Microsoft’s Windows NT operating system. To help fill in for a monstrous lack of native Alpha NT applications DEC even produced a revolutionary software product, the FX!32 combination emulator/binary compiler, that would allow x86 NT applications to be run seamlessly on Alpha hardware at a substantial fraction of native code performance . To help fill the anticipated huge demand for Alpha processors DEC even licensed its designs to Korean chip giant Samsung. However motherboards for the 21164PC remained expensive, Windows NT didn’t take the world by storm, and x86 processors soon transitioned to 0.25 µm. Demand for Alpha MPUs fell far short of what DEC Semi needed to survive, let alone keep Samsung happy .
In all three cases we had MPU vendors (and in the case of MIPS, foundry partners) that couldn’t keep up with Intel in both design and manufacturing to deliver faster and cheaper MPUs for a sustained period. In addition, MIPS Computer Systems and Digital were not economically robust enough to engage in a sustained campaign to establish a new ISA against an entrenched and lavishly funded adversary. The AIM alliance was the best equipped contender in terms of financial resources and process technology but it couldn’t keep up with Intel’s pace in MPU design. In addition, the AIM alliance not only took on Intel but Microsoft too, greatly compounding the challenge.
Discuss (83 comments)