In terms of sheer processing power, Intel is behind the RISC curve. Its next-generation Pentium, the P6, may only keep the gap from widening even further. But for Intel, that may be enough.
Dave Andrews
On April 1, Intel will reduce the price of its 66-MHz Pentium chip from $871 to $750 each in quantities of 1000, company officials say. Given that chip's optimized SPECint92 rating of 67.4, that means a system vendor will pay about $11.12 per SPECint92 for a 66-MHz Pentium at the new price. By comparison, Sun Microsystems' MicroSparc II processor running at 70 MHz delivers a 54-SPECint92 rating at a volume price of $400 each, or about $7.40 per SPECint92. And the PowerPC 601 chip, running at 66 MHz, delivers a rating of 60 SPECint92, for about $6.45 per SPECint92--roughly half th
e cost of the 66-MHz Pentium.
Industry experts and analysts agree: In the microprocessor war, Intel has lost the price/performance battle to RISC. Intel's response? It's getting more aggressive on pricing and is accelerating the introduction of new processors. It was expected to introduce in March 90- and 100-MHz versions of a new 0.6-micron, 3.3-V Pentium family, code-named the P54C, that analysts expect will offer a 30 percent to 50 percent performance boost over today's Pentium.
Intel's other recent processor introductions include the DX4/100, a 486 processor that operates at 100 MHz internally and is about 50 percent faster than a 486DX2/66, neatly filling a gap between the fastest 486 and the slowest Pentium. Says Lew Paceley, marketing director for the P6 line at Intel, "We just keep driving our engine harder and harder."
Intel says its next-generation 80x86 processor, code-named P6 and slated to start appearing in late 1995, will deliver roughly 300 MIPS, compared to the approximat
ely 100 MIPS that you get from a 60-MHz Pentium today. Such an accomplishment--if Intel can meet that goal within that time frame--means the company will have reduced the amount of time between the introduction of its successive next-generation microprocessors by about a year. A late-1995 rollout of the P6 will represent an elapsed time of only 33 months after the first Pentium was announced in March 1993; 44 months passed between the introduction of the first 486 and the debut of the Pentium.
But by late 1995, Motorola and IBM expect to be deep into volume production on the PowerPC 620, which Motorola says will deliver four times the performance of today's PowerPC 601. Sun expects to deliver by the end of 1995 its next-generation MicroSparc III processor, which will achieve a rating of 150 SPECint92. Other companies, including Hewlett-Packard, DEC, Mips, and their partners, are pushing their RISC architectures up the performance curve as well (see the chart, below left).
"Even though the P6 loo
ks impressive, it's still going to be well behind the performance of the PowerPC and other RISC chips," notes Linley Gwennap, editor in chief of the Microprocessor Report (Sebastopol, CA). "The rule of thumb is that RISC is either twice the performance at the same [chip] price as Intel, or the same performance at half the price." Gwennap also says that Intel is investigating multichip-module packages for the P6 that can increase performance and reduce footprint but would also be much more expensive than stand-alone CPUs.
Pure processing power is not the only way to measure the strength of a particular computing platform, however. Platforms are also measured by the breadth of available applications that are written by ISVs (independent software vendors). The 80x86 architecture, unlike RISC, runs thousands of DOS and Windows applications at native speeds without having to resort to slower software-emulation solutions such as Wabi or Insignia Solutions' SoftWindows. "The PowerPC and other RISC guys need t
o work on the software problem," Gwennap says.
Until more ISVs introduce 32-bit applications that run native on a given RISC platform, Intel will have a strong argument to stay with its 80x86 architecture. If you are looking for full compatibility and the fastest possible execution of today's Windows applications, the 80x86 architecture is for you. Indeed, the Framingham, Massachusetts-based International Data estimates that over 36 million PCs were sold worldwide in 1993, versus about 1 million workstations.
But fast execution of 16-bit Windows applications is not the chief objective of these new RISC PCs. Companies such as DEC, HP, and IBM, together with their system partners, are developing low-cost RISC-based PCs for a new emerging market of low-cost enterprise workstations. The enterprise workstation arena should see the most intense competition between RISC and CISC.
Enterprise workstations integrate high-performance CPUs with 32-bit operating systems such as Unix, Windows NT, and O
S/2 and run applications that manage processes that directly affect a company's bottom line. These applications, according to Frost & Sullivan, a Mountain View, California-based market-research firm, include customer information systems, client/server databases, production, and R&D programs.
"The enterprise desktop is a market that is primarily business-operations-oriented," says Andrew Allison, consultant and editor of Inside the New Computer Industry (Carmel, CA), a newsletter that focuses on this new breed of PC. "It's far from clear that Intel has any software advantage in that market."
By their very nature, enterprise-critical applications have to run on operating systems that offer the stability and fault tolerance that are not available in today's fragile DOS-based environment. Intel's overall system-price advantage is eroding as well: This year, you can expect to see a wave of high-performance desktop RISC workstations selling in the $3000-to-$4000 range (e.g., HP's Series 700 systems, w
hich start at $3995).
Start-up companies such as Deskstation Technology (Lenexa, KS) are banking their future on developing low-cost, high-performance, RISC-based Windows NT PCs for companies that are downsizing their business operations. For example, Deskstation is now shipping its v4600, a low-cost workstation based on IDT's R4400-compatible R4600 processor running at 100 MHz internally. It comes with 16 MB of RAM and a 240-MB hard drive and costs $2995.
Don Peterson, Deskstation president, says that over the next two years the company will deliver a range of systems designed to satisfy the needs of customers using everything from desktops to data centers; by the end of the year, the company plans to offer multiprocessor systems that will run demanding database applications. "We believe we'll be able to deploy low-end servers for under $10,000 that will offer performance that's better than [that of] a two-processor Pentium," Peterson says.
The degree to which Intel can convince customer
s to purchase Pentium-based enterprise workstations that offer RISC-like performance depends in large part on Microsoft and how well Chicago, the next generation of Windows, succeeds in integrating 16-bit DOS and Windows in a 32-bit multitasking operating system that runs well on 8-MB-RAM PCs. If businesses decide that Chicago doesn't offer a compelling enough platform to run their applications, they will closely examine the new breed of RISC PCs, which will have the horsepower to run the next generation of sophisticated, layered 32-bit operating systems, such as Windows NT and CPU-hungry videoconferencing, database, and 3-D graphical applications.
Illustration: Projections are based on interviews with company officials and industry analysts. In addition to the cost of the microprocessor, other factors influencing overall cost of a system include level of processor integration, operating-system RAM requirements, chip die size, availability of low-cost peripherals, and application prices. Because
many enterprise applications will rely more on integer than floating-point performance, processor SPECfp92 ratings are not shown.
Illustration: Databases will be a key application for enterprise workstations as companies turn to these programs to manage their high-volume, real-time data-entry and update needs, according to market-research firm Frost & Sullivan. Database applications do not generally require excellent floating-point performance, unlike complex financial-modeling and 3-D graphics applications.