This Chip Battle Royal Ain't Over Till It's OverRobert D. Hof
The evidence looks overwhelming: RISC has fizzled. This chip-design technique, which makes computers run at blinding speed, was supposed to challenge Intel Corp.'s bread-and-butter microprocessors. Yet in 1991, a half-decade after reduced instruction-set computing arrived, only 308,000 RISC chips went into desktop computers, says market researcher InfoCorp, vs. the 20.4 million mighty Intel sold. "We have won the RISC battle," boasts Intel President Andrew S. Grove.
The war isn't over yet, however. Software giant Microsoft Corp. soon will release a version of its Windows program that gives RISC chips a new crack at Intel. At present, some 20,000 programs run on Intel-based PCs, five times the number for the best-selling RISC chip, the Sparc from Sun Microsystems Inc. So even a twofold jump in computer speed, the edge RISC typically provides, isn't enough to get most computer buyers to junk their software investments.
Microsoft's new Windows NT software might start to change this. It will work not only on Intel chips but also on RISC chips from MIPS Computer Systems Inc. and Digital Equipment Corp. Any existing PC program can then be run under Windows on machines that use these chips. Here's the message for Intel: It can't count on its software advantage to fend off the RISC intruders forever. And this has turned up the heat under its campaign to speed chips to market.
MAKING HASTE. Intel's chips are based on a technology called CISC, or complex instruction-set computing. The CISC approach dates from the dawn of digital computers: It feeds instructions to a computer's brain in clusters of related operations. Years ago, when computer memory was expensive and in chronically short supply, this made the best use of a scarce resource. Now that memory is plentiful and cheap, better techniques such as RISC can take hold.
Despite its name, RISC doesn't always mean reducing the number of instructions, though that helps. Mainly, it imposes limits on the number of tasks contained in each instruction. With RISC instructions, everything gets done in one split-second tick of the chip's internal clock. A computer that executes each instruction in one clock cycle can barrel along without looking back. By contrast, a computer processing a long CISC instruction must periodically check to see that the clustered tasks are getting done in the proper sequence, wasting time. Simple instructions also mean simpler circuits that need fewer transistors. So there's extra room on a RISC chip for special, speed-enhancing circuits, such as so-called cache memory.
To speed up its microprocessors, Intel is stealing as many tricks as it can from the enemy. It has tweaked some of the most commonly used CISC instructions so they execute in a single clock cycle. Intel's manufacturing engineers are constantly hammering out ways to cram more transistors on a chip, so designers can boost speed and expand on-chip memory. The result: Intel's upcoming 586 chip will chew through 100 million instructions per second, close to the mark of established RISC rivals such as Sun's Sparc. Still, the latest RISC designs, such as DEC's Alpha and Texas Instruments Inc.'s SuperSparc, keep the edge in speed decisively in the RISC camp.
MARKET EDGE. Intel's most potent countermeasure is its huge market share. It can spend great sums on research and development and new plants, since it can spread the costs over millions more chips than RISC chipmakers can. So Intel can invest more than its rivals and still undercut their prices per measure of performance.
Intel doesn't have many other options. It has the skill to jump entirely to RISC if need be and, in fact, already turns out RISC co-processors for use mainly with graphics programs. But for now, its microprocessors must rely on some CISC to handle that massive library of existing software. To deal with this restraint, Grove depends on Intel's strength in design, production, and marketing. All that can derail Intel, he declares, "is if we fall asleep at the switch."