| On March 22, Intel unveiled the Pentium — the next
generation of the microprocessor architecture that has
dominated the PC industry for more than a decade. With
this new processor, Intel has begun steering a difficult
course between tradition and revolution. On one hand,
Intel must prolong the life of the 15-year-old 80x86
architecture, a tried-and-true design that encompasses an
installed base of tens of millions of computers and
thousands of programs. The corporations and individuals
who have spent billions of dollars to buy that hardware
and software are as anxious to protect their investment as
Intel is. At the same time, however, Intel must also push the envelope of microprocessor performance to new levels that will satisfy the increasing demands of computing in the 1990s. Multimedia applications with full-motion video and robust new operating systems like Windows NT are pushing the limits of today's fastest desktop computers. To meet these needs, engineers are cramming more and more transistors into their CPU designs. Exotic new architectures that were undreamed of a few years ago are now beginning to appear. To keep pace, Intel must adapt those advances to its 80x86 chips without deviating too radically from the basic architecture that it first committed to silicon in 1978. The Pentium is not just another new chip but a new 80x86 chip — an event that happens only once every four years (see figure 1). As such, it's a tantalizing combination of inherited limitations and future possibilities. It needs to be judged by two standards: how well it carries on the 80x86 standard and how well it performs. It's imperative for the chip to be compatible with previous 80x86 chips, and it is. The Pentium is fast, too, though perhaps not as fast as you would like. It's also expensive. With sources estimating an OEM price of about $900 per chip in production quantities (Intel had not released pricing on the Pentium as this went to press), the Pentium is far too costly for today's mass-market, single-user desktop computers. But the same was true of the 486, 386, 286, and 8086 when they hit the market. Over the course of the next 12 to 24 months, the Pentium will primarily be harnessed to network servers and workstation-class machines priced from $5000 to $10,000, according to system vendors and industry analysts. Systems based on the Pentium will not be announced until May 17. Key vendors expected to make Pentium-related announcements at that time include AST Research, Compaq Computer, Hewlett-Packard, IBM, NCR, Unisys, and Zenith Data Systems. As prices inevitably fall and performance improves, the Pentium could become the dominant microprocessor of the mid-1990s. However, Intel faces a challenge it has never before experienced with a new PC-based CPU: competition right out of the starting block. Microsoft's Windows NT is the anticipated operating system of choice for Pentium systems, but Windows NT-ready workstations based on RISC CPUs will beat the first Pentium systems to market. For the first time ever, you don't have to choose Intel for your next-generation PC system architecture. Coping with RISC Intel has announced two versions of the Pentium: one clocked at 66 MHz and the other at 60 MHz. The BYTE benchmarks confirm that the faster Pentium barely exceeds Intel's stated design goal of delivering at least twice the performance of today's fastest 486, the clock-doubled 486DX2-66 (see the text box "Testing the Pentium" on page 96). The 66-MHz Pentium is rated at 112 MIPS according to the Dhrystone 1.1 benchmark, and the 60-MHz Pentium delivers about 100 MIPS. By comparison, a 66-MHz 486DX2-66 yields about 54 MIPS. A twofold increase in speed is enough to command a premium price in today's performance-hungry PC market, especially in the highly specialized niche of network servers. Because the performance of a server affects all the users on a network, any improvement is easier to justify when weighed against the higher costs. In addition, the Pentium marks the first time an 80x86 chip has broken the magical 100-MIPS barrier, which until now has been the exclusive domain of RISC-based microprocessors. "I think [the Pentium] sounds real strong," says Roger Rapp, an information technology manager at John Brown, a worldwide engineering and construction firm. Rapp, who is based in Mount Vernon, Indiana, is responsible for all the hardware and software, including 31 AutoCAD licenses, used at John Brown's offices in Indiana, Kentucky, and Missouri. Although Rapp is glad that Intel continues to develop the 80x86 architecture, he has a wait-and-see attitude. "I'm agnostic on the whole shooting match," he says. "I have to do whatever's best for the end user." Thanks to power-hungry users like Rapp, RISC processors pose the greatest threat to the future of the Pentium. Unbound by any requirements to maintain compatibility with old hardware and software, RISC chips are free to embrace radical new architectures and advanced design theories. As a result, they can deliver impressive levels of raw performance that leave most CISC-based chips in the dust (see the box "Pentium vs. RISC"). But they also tend to be expensive, partly because none has achieved the volume that leads to inexpensive mass production — at least not on the scale of Intel's most popular chips. All that may change, however, with the dawning of Windows NT. This operating system threatens both the Pentium processor and Intel's long reign over the PC market in two different ways. First, Windows NT offers symmetrical multiprocessing, which means Pentium-based computers will be required to compete on price/performance grounds against systems that include two or more 486 chips working together. Second, Windows NT is strategically designed to be independent of the underlying hardware; it's supposed to run equally well on such non-Intel RISC chips as DEC's Alpha series and the Silicon Graphics/Mips R4000. In the Pentium's favor is the fact that DOS and Windows 3.x applications will not run natively on an Alpha- or R4000-based NT system. This is because these RISC chips are not binary compatible with the 80x86 instruction set. To run DOS and Windows applications on these systems, NT will transparently emulate the 80x86, using technology developed by Insignia Solutions, a company best known for its SoftPC emulator. As with all software emulation, there's a performance penalty, so apples-to-apples comparisons between 80x86- and RISC-based systems will be difficult, if not impossible, when running existing applications under Windows NT. But comparisons will be much easier — and tempting — when running new software written for NT. Until now, the only common ground for comparing the performance of 80x86 and RISC processors was Unix. Although Unix also offers symmetrical multiprocessing and platform independence, the vast majority of PC users have avoided it (see "Is Unix Dead?," September 1992 BYTE). Those users may be more receptive to the more familiar Windows NT, and NT will pit the Pentium directly against its RISC rivals for the first time. Intel insists it will not shrink from such comparisons. While the RISC-versus-CISC debate has been raging, Intel has been quietly adapting key elements of the RISC philosophy to its decidedly CISC-based 80x86 architecture (see the photo on page 100). A few hints of this were evident in the 486's FPU and cache, but the Pentium goes much further in this respect. The prime example is the Pentium's superscalar pipeline, a common feature of RISC chips. Under certain circumstances the Pentium is capable of executing two machine instructions simultaneously, because it has not one, but two integer pipelines arranged in parallel. It also has intelligent branch prediction and other advanced RISC features (see the text box "Inside the Pentium" on page 102). Even though the Pentium's FPU pipeline does not execute in parallel, other improvements in the chip significantly boost its floating-point performance. Preliminary benchmarks indicate the Pentium's FPU is about twice as fast as the 486's. This, together with the superscalar ALUs, positions the Pentium closely with competing RISC chips. "For the average 'schmo,' the average businessperson doing word processing and spreadsheets, [the Pentium] is not going to be a big deal, and I see no reason for a lot of them to go past a 386," says Nancy Thalblum, a New York consultant and member of the NYPC users group. "But for other kinds of people, you have to have the raw horsepower." For those people, Intel says that the Pentium's hybrid architecture is just a starting point. Faster versions of the Pentium are on the way. The initial Pentiums are based on 0.8-micron, 5-volt BiCMOS technology, with about 3.1 million transistors. Future versions, which may appear by late this year or in early 1994, are expected to use 0.65-micron, 3.3-V technology and will run at 100 MHz or faster, according to industry analysts. Intel won't comment, but if it keeps with its marketing strategy of spinning off numerous variations of a basic microprocessor, you can anticipate a wide variety of Pentiums for many purposes: servers, desktops, and portables. Meanwhile, if history repeats itself, the Pentium will benefit from the same performance gains and manufacturing efficiencies that have made other 80x86 microprocessors faster, denser, and less expensive. Sales are projected to ramp up quickly: from 400,000 chips this year to 2.1 million next year, according to Ken Lowe, senior industry analyst at Dataquest (San Jose, CA). Upgrade Paths One variation of the Pentium is already known. Code-named the P24T, it's an upgrade Pentium for 486-based computers, sort of a Pentium OverDrive. Some 486 machines that are now available, such as AST's new Premmia 4/66d, have a special socket with an extra row of pins for the P24T. When the P24T is introduced early in 1994, it will pop into this 238-pin socket, preempting a machine's existing CPU. However, some of these computers won't be quite as fast as systems that are designed especially for the Pentium, because the P24T will access its secondary cache memory over a 32-bit bus running at only 33 MHz. True Pentium systems will boast a 64-bit bus that accesses the secondary cache at full CPU speed (see the box "Three Ways to Upgrade"). Another type of upgrade will eliminate the P24T's bus bottleneck by isolating the Pentium on a special daughtercard. Systems with this upgrade will unleash the full power of the Pentium by linking the chip to its secondary cache at CPU speed over a 64-bit bus. The daughtercard will plug into a special connector on the motherboard. AST's Premmia 4/66d offers this option in addition to the P24T socket. More of these kinds of systems will be announced on May 17, and their designers claim they'll match the performance of computers with Pentium-based motherboards. Some users are wary of upgrades, however. "The vendors have caught on that if they want to make users' check-off lists, they've got to include that upgradability," says Jonathan Braunhut, director of the information center for the New York City controller's office. "In reality, we see that when you want to take the processor up, the video standard has been advanced. Typically, the local hardware storage requirement has advanced. Ultimately, if you've replaced everything but the $50 metal case, what have you accomplished?" But other users will take all the MIPS they can get. "At this point, I'd pay about $1200 to upgrade a system [to the Pentium]," says William Zeph Ginsberg, a New York City architect. "I use CorelDraw, and I'm getting into Windows applications, and I'm sure tired of watching that hourglass. Life is too short to spend waiting for your hard disk to access. The big CAD drawings and the complicated Windows files just need more power." Fine-Tuning the Software So much attention gets focused on the hardware aspects of microprocessor design that another factor — software engineering — is easy to overlook. For nearly two years, Intel has been working with tool developers to ensure that specially optimized compilers will be available when the Pentium hits the market. More than any previous Intel microprocessor, the Pentium needs optimized code to reach its full potential. Although it's fully compatible with code written for earlier 80x86 chips (the instruction set is almost identical to the 486's), significant performance gains can be realized by writing code that exploits its superscalar architecture. This is not surprising, because optimized code is also important in maximizing the performance of the RISC chips that inspired crucial elements of the Pentium. "As processor architectures become more exotic, with parallelism and so forth, it becomes more important for the software to know what the hardware is trying to do," says Dan Palka, manager of Intel's Compiler Partners Program. "Now there's much more awareness that the software architects and the hardware architects have to work together." Few applications are carefully hand-coded in assembly language anymore, so optimized high-level compilers are critical to extracting maximum performance from the Pentium. Even small improvements to these compilers have a ripple effect among millions of users, because virtually all of today's commercial applications software (and much of the system software) is written and compiled in C or C++. According to Intel and the tool developers interviewed by BYTE, programs optimized for the Pentium will run about 30 percent faster than unoptimized programs. In addition, users can expect gains of about 10 percent when these programs are run on the 486 and perhaps 5 percent when they're run on the 386 (see figure 2). The optimized code won't run at all on earlier chips, but most 286, 8086, and 8088 systems lack the speed and memory to run today's major applications. If a computer can't comfortably run Windows 3.1, it's not current enough to run the software that is likely to be recompiled for the Pentium. Don't expect to buy these optimized programs anytime soon, though. For one thing, the first optimized compilers weren't scheduled to be available until this spring. Borland introduced a new version of C++ for OS/2 with Pentium optimizations in February, and Microsoft is planning to include the optimizations in an upcoming version of Visual C++ for Windows NT, which is scheduled to ship within weeks of NT's ship date. But even when such compilers are available, the Pentium market is still far too tiny to make it worthwhile for software vendors to recompile their applications for the sake of the Pentium alone. In addition, the performance gains are probably too small to be noticeable on the other chips. Not until the next major upgrade will most vendors even consider optimizing their code. In fact, some software publishers are just beginning to confront the issue of optimization. "Intel's statements that applications should be recompiled for the Pentium took us somewhat by surprise," says Darin Richins, public relations manager at WordPerfect. "We don't have a good idea yet of what needs to be done to make our applications run well on the Pentium." Microsoft and Borland, the two leading compiler vendors, say that the first of their applications to be optimized will be 32-bit programs for Windows NT. Autodesk, whose AutoCAD package is a natural candidate for the Pentium, won't even comment. It'll also be a while before optimized compilers include all the possible optimizations. In general, these optimizations affect the ordering of the instructions generated by the compiler. By replacing complex instructions with simple instructions (i.e., those that don't require microcode) and by arranging them in ways that minimize dependencies (so an instruction doesn't depend on the result of the instruction it immediately follows), these compilers generate code that executes in parallel through the Pentium's twin integer pipes. Keeping both of those pipelines primed will greatly increase the chip's throughput, which means your programs will run faster. A few optimizations produce code that runs more slowly on the 486 and 386. For that reason, Intel recommends a "blended optimization strategy" — a mix of optimizations that runs very well on the Pentium and either slightly faster or at about the same speed on earlier chips. Future compilers will offer multiple levels of optimization, so programmers can make their own compromises. System-Design Challenges Another easily overlooked aspect of Pentium performance is the design of the computer systems that are wrapped around the chip. There's much more to designing a well-rounded Pentium system that just sticking the chip into a socket and adding some memory (see the box "Pentium Design Hurdles"). Indeed, the Pentium poses so many challenges that some engineers and analysts suggest it will shake out the PC clone industry. The main design problem is engineering a reliable motherboard that's fast enough to keep up with the chip. This isn't a Pentium-specific problem; it has already been encountered by engineers designing systems around the 50-MHz 486DX. (The clock-doubled 66-MHz 486DX2 has a bus that runs at only 33 MHz, so it presents less difficulty in that regard.) But it's a particular problem with the Pentium because even the slower version is clocked at 60 MHz. At that speed, there's no margin for careless design or sloppy manufacturing tolerances. This could have a big impact on users. If the Pentium's high clock rate — soon to climb to 100 MHz and beyond — deters the low-end, high-volume system makers, the retail prices of Pentium computers won't drop as fast as they did when 386 and 486 computers were new. "Fifty MHz was painful, and 66 MHz is turning out to be more painful," says Steve Phillips, the manager of advanced development for the Workstation Products Group at Unisys (Blue Bell, PA). "It's like a barrier. Many of those [clone] manufacturers will be out of business, or else they're going to have to garner experience they haven't demonstrated in the past." When designing a system to run reliably at high clock speeds, Phillips explains, an engineer has to carefully manage all "boundary conditions": Thermal output, voltage variations, signal losses, RF interference, and TTL signaling all become significant. Thermal considerations are especially critical. The Pentium is a hot chip in more ways than one. A single fan for the whole computer may not be enough, and the placement of fans and cooling vents isn't a haphazard affair. Even before sample silicon was available, Intel gave systems designers thermal models so they could begin planning how to keep their computers cool. "This is where our big-system experience really becomes important," says Phillips. "We're used to dealing with issues such as thermal modeling and high-speed electrical design, whereas many PC manufacturers don't have that experience. It's like PCs are finally growing up." Although the design challenges aren't insurmountable, they will weed out some of the low-end competition, says Dataquest's Lowe. "I don't think these things will permanently bar the no-name companies from participating [in the Pentium market], but it'll take them longer to master the design problems. It's one more barrier that's contributing to a general consolidation of the clone industry," he says. Engineers must also pay more attention to performance bottlenecks, because a fast CPU often exposes hidden weaknesses in other components. For example, many Windows users have discovered that even a fast 486 gets bogged down without an accelerated video card or local-bus video. Likewise, Pentium-based computers will need to exploit every trick in the book to deliver the performance users expect. Some systems will double the secondary cache to 512 KB, and others will put drive controllers as well as video on the local bus. Intel claims that today's VL-Bus — a widely adopted local-bus standard created by the Video Electronics Standards Association (San Jose, CA) — is not fast enough for the Pentium. Intel says Pentium systems should use PCI (Peripheral Component Interconnect), the local-bus standard invented by Intel. It's true that today's VL-Bus is only 32 bits wide. But VESA executive director Tom Ryan says VL-Bus was designed for future expansion to 64 bits and will keep pace with the Pentium, as well as with RISC chips such as the Silicon Graphics/Mips R4000. "Obviously, with the 64-bit extension, that's not a problem," says Ryan. "It's something that's in committee at the moment." Because the VL-Bus chip set and connectors are much less expensive than PCI's, it's likely that VL-Bus will be the dominant standard on low-end 386 and 486 clones for the next year or two (see "Fast Transit," October 1992 BYTE). Pentium systems, which by definition will be considered high-end systems for at least the next year, will probably adopt PCI instead. Keeping Up with the Joneses The Pentium presents a challenge of a different sort to Intel's competition. These competitors fall into two groups: those that make 80x86-compatible chips, and those that are promoting entirely different architectures, usually RISC-based ones. So large is the worldwide market for 80x86 processors that the first group can survive by undercutting Intel's prices and providing computer manufacturers with an alternative supply source (see "Make the Right CPU Move," December 1992 BYTE). AMD (Sunnyvale, CA) and Cyrix (Richardson, TX) are the primary players in this group. Although neither of these two companies has a Pentium-class processor ready to ship, neither admits to being overly concerned about it. "As a company, we focus on the volume markets, and the volume markets right now are the 386 and the 486," says AMD spokesman David Frink. "In fact, we expect the 486 to be the volume product through 1994 and into 1995. When you're talking about a market of 35 million to 40 million PCs that will be sold in the next year, and Intel will sell maybe 100,000 Pentiums in that period, then the Pentium is really insignificant — really nowhere." If AMD and Cyrix truly aren't worried about the Pentium, they should be, because both companies are having trouble cloning the 486. AMD's 486 project suffered a major setback last December when a federal judge ruled against the company in a long-running court battle with Intel. The judge decided that a 1976 licensing agreement between the two companies did not entitle AMD to use Intel's microcode in future microprocessors, including the 486-compatible chip that AMD had planned to ship in January. As a result, AMD was sent back to the breadboards, forced to rewrite its microcode. The introduction of the Am486 was pushed back to June. Cyrix introduced three "486" processors during 1992: the Cx486SLC, Cx486DLC, and Cx486S2/50. But despite their names, none is considered a true 486-class chip. Although they're compatible with the 486 instruction set, they lack integrated FPUs and have only 1 or 2 KB of internal cache. Intel's 486SX doesn't have an FPU, either, but it has an 8-KB cache and is otherwise identical to the 486DX. Cyrix won't introduce its 486DX clone, code-named the M7, until sometime during the second half of this year. Cloning the Pentium will prove an even greater challenge, not just because it's faster, but also because it's architecturally more complex. There are two general approaches to the problem: Either copy the architecture as closely as possible or depart from the Pentium's internal design and use other techniques to make a chip that delivers the same level of performance. Both AMD and Cyrix are taking the second approach. AMD is working on a new 80x86-compatible core that inherits little from existing designs. It's an ambitious project, because AMD is simultaneously working on a new 0.35-micron process technology and wafer-fabrication facility that will allow engineers to cram as many as 10 million transistors onto a single chip. That's enough density to propel AMD into the Pentium class and beyond. The first chips based on the new submicron process are expected in 1995. To spread out the burden of these enormous development costs, which Dataquest analyst Charles Bouchet estimates to be $800 million to $1 billion, AMD has formed an alliance with HP. HP will apply the technology to its next generation of PA-RISC processors, as well as to other chips. Cyrix, too, is diverging from the Pentium's architecture, and spokesman Mike Bruzzone claims the company's first Pentium-class chip will be ready this year. "We'll release a competitor — not an equivalent — to the [Pentium] during 1993," he explains. "It's targeted at 100 to 150 MIPS and will calculate three instructions per cycle. It will be optimized for calculating integer instructions and will not need recompiled software. We don't design our chips based on what Intel does; we design our chips based on what 80x86 software requires in a CPU. So we think that one of the big advantages of our superscalar chip, our [Pentium] competitor, is that it'll achieve the most robust performance levels without recompiled code." However, this design philosophy raises questions about compatibility and performance. If future chips from AMD and Cyrix don't exactly duplicate the Pentium's parallel pipelines and other architectural features, they may cause software written with Pentium-optimized compilers to behave differently. The software may run faster, slower, or at the same speed, but there's also the possibility, however remote, that in certain circumstances it may not run as expected. For example, the Pentium has special logic to detect and handle self-modifying code, which can play havoc with cache coherency. Programmers aren't supposed to write self-modifying code, but they always have and probably always will. If a Pentium-class chip doesn't implement this logic in precisely the same fashion as the Pentium, the results may be unpredictable. If enough users start worrying about these questions, it's bad news for AMD and Cyrix. But the two companies downplay these possibilities and point out that their chips undergo rigorous compatibility testing throughout their design cycles. But as the 80x86 architecture becomes more exotic, and as 80x86-compatible processors diverge more radically from the Intel chips they're trying to emulate, chances are that internal differences will translate into external idiosyncrasies. The Intel architecture won't end with the Pentium, of course. Intel is already working on the Pentium's successor, code-named the P6. Sources say the P6 will debut as soon as 18 months from now, will incorporate 10 million transistors, and will run at 200 MIPS. Aside from promising that an upgrade from the Pentium will be available, Intel isn't releasing any details. New Competing Architectures While Intel fends off competition from 80x86-compatible microprocessors, completely different breeds of chips are threatening the Pentium from another quarter. This camp comprises rival architectures, usually RISC-based, from such companies as DEC, HP, IBM, Motorola, Texas Instruments, Silicon Graphics/Mips, and others. At least in the 80x86 arena, Intel is fighting on familiar turf. But the threat from rival architectures is more serious because of the risk that someone will score a stunning price/performance breakthrough that renders the Pentium hopelessly obsolete. Although Intel staunchly defends the 80x86, it's not easy fighting off challengers with a basic architecture that was laid down when Jimmy Carter was president. The immediate threats to the Pentium are the DEC Alpha series and the Silicon Graphics/Mips R4400, a pair of hot new RISC chip designs. Because both will run Windows NT, direct comparisons with the Pentium are inevitable. Comparisons must be made thoughtfully, however, because it's not just performance but the price/performance ratio that counts. Historically, the prices of Intel microprocessors usually fall further and faster, but those patterns aren't necessarily carved in silicon. Motorola's 68000 series, a perennial 80x86 rival since the 1970s, appears to have fallen a whole year behind the Pentium. The next-generation chip in that series is the 68060, which isn't due until early 1994 — and Motorola says it will debut at only 50 MHz and 77 MIPS. A more likely contender is the PowerPC, a new line of RISC chips under development by Motorola and IBM in partnership with Apple. Derived from IBM's RISC System/6000 processor, the PowerPC series is intended to drive the next generation of Apple and IBM computers. The first entry in the four-chip PowerPC line is the 601. It is designed for moderately priced systems and is expected to debut by early 1994. Its performance is expected to be close to that of the Pentium, though perhaps not quite as fast. Later PowerPC chips will offer higher performance and lower power consumption. Meanwhile, Microsoft hints that it will port Windows NT to additional processors. Other operating systems are also migrating across platforms: Sun Microsystems' Solaris and Next's NextStep are coming to the 80x86; the PowerPC's PowerOpen is an operating system that's designed to be ported to different chips; Taligent, another IBM-Apple partnership, is developing a portable operating system that's due out in 1995; and IBM's OS/2 remains a dark-horse contender. PC users suddenly enjoy new alternatives. The relationship that for years has linked these users with the 80x86 architecture, MS-DOS, and Windows is being broken. As the Pentium emerges from the security of that relationship, the 80x86 architecture will face new challenges to its survival in the 1990s. Editor's note: BYTE news editors Patrick Waurzyniak and Ed Perratore, senior news editor Gene Smarte, West Coast bureau chief Andy Reinhardt, and executive editor Rich Malloy also contributed to this article. Sidebars:Pentium Design HurdlesHEAT DISSIPATION: The Pentium runs hot. Intel is providing system vendors with thermal models, but vendors will have to pay close attention to fan and component location.SPEED: The remainder of the motherboard — memory, video, system bus — must be able to keep up with the speed of the Pentium. TOLERANCES: With higher speeds, the margin for error diminishes dramatically. Small vendors with little or no experience in building high-speed systems might not be able to master the learning curve. RF INTERFERENCE: The faster the system, the more RF interference it produces. It will be tougher for vendors to meet the various RF emission standards with the Pentium. Figure 1: Intel sticks to a remarkably regular product cycle: Each new generation follows the previous one by about 44 months. The number of transistors on 80x86-series microprocessors has increased exponentially since the debut of the 8086 in 1978. During this period, the minimum feature size has shrunk from 3 microns on the 8086 to 0.8 micron on the Pentium. Future versions of the Pentium will probably use 0.65-micron technology for even higher densities. At 112 MIPS, the 66-MHz Pentium is nearly 150 times faster than the speediest 8086. Note that these MIPS ratings are for the fastest version of each chip; the Pentium is sure to get faster, too, as new versions are introduced. Figure 2: Based on Intel's estimates, Pentium-optimized applications will run about 30 percent faster than unoptimized programs on Pentium systems and 5 percent to 10 percent faster on 386/486 systems. Tom R. Halfhill is a BYTE senior news editor. You can reach him on BIX as "thalfhill." Letters / August 1993
I found Tom R. Halfhill's article on the Pentium ("Intel
Launches Rocket in a Socket," May) very informative but a
little biased. Halfhill apparently considers Windows NT
the only operating system that will run on the Pentium.
However, Next has been working closely with Intel in
developing a Pentium version of NextStep. |