Rasmussen Consulting Group April, 2005 Intel and the RISC Wars One of the largest threats to Intel and its dominance of the PC microprocessor world occurred in the early 1990s with the emergence of the first commercial RISC microprocessors and a large computer industry driven initiative called the ACE consortium. Nearly every Intel ally, from operating system partner Microsoft to leading customers including Compaq was actively planning to switch server architectures away from the aging x86 to a new platform. The ACE consortium created clear danger to Intel’s high-end business and a strong, potential threat to their entire personal computer franchise. This paper covers the scope of those threats and how Intel was able to cull together a strategy to effectively respond, reducing the threat of ACE and actually strengthening its position in the long run. In 1985 Intel made a gut-wrenching strategic shift away from memories as its primary product line and towards microprocessors. The popularity of the PC-AT and similar clones based on the 8086, 80286, 80386 and 80486 (1989) had minimized the threat of most other microprocessor architectures including the Motorola 68000 family upon which Apple computers and Sun Microsystems UNIX machines were based. Intel and Motorola were increasingly dominating the microprocessor market. Intel and Microsoft had successfully staved off competition in the PC space from AMD, Cyrix, Chips and Technologies and IBM1. Motorola 680x0 family was at the heart of machines ranging from Apple’s Macintosh to Sun workstations. By 1986, 90% of the PCs shipped were based on Intel processors and 90% of workstations were based on Motorola processors. One key factor of the continued success of the Intel microprocessor architecture was backward code compatibility, allowing software written for older processor generations to run on current machines2. This compatibility and the resultant longevity of software proved to be a boon to consumers. However, each successive generation of these Complex Instruction Set Computer (CISC) architectures contained circuitry that executed all instructions from previous generation while adding new instructions to improve performance. This requirement to run ‘legacy’ code created problems and complexity when designing new processors. Thus, the time between processor introductions was growing and the incremental speed improvement between each generation (on a percentage basis) was slowing. 1 One of the main reasons why IBM chose the 8088 as the basis for the IBM PC was they had manufacturing rights. This code base of legacy software became the core message for a marketing program in the mid 1980s, internally known as “Project Crush”, designed to gain design wins for Intel over competing architectures of the time. 2 Rick Rasmussen prepared this note for the course BCEMBA 7017-200 Strategic Management of the Enterprise as taught by Professor Paul Tiffany during Winter Quarter 2005 as a basis for class discussion rather than to illustrate either effective or ineffective handling of an administrative situation. Copyright © 2005 by the founder and principal of Rasmussen Consulting Group. To order copies or request permission to reproduce materials, call 1-408-813-1299 or write Rasmussen Consulting Group, San Jose, CA 95125. No part of this publication may be reproduced, stored in a retrieval system, used in a spreadsheet, or transmitted in ay form or by any means – electronic, mechanical, photocopying, recording, or otherwise—without the permission of Rasmussen Consulting Group. 1 Intel and the RISC Wars RISC Microprocessors The theoretical basis for Reduced Instruction Set Computers (RISC) was the 80/20 rule – 20 percent of the operations were run 80 percent of the time. These 20 percent should be in hardware (silicon) while the balance can be done in software. Standardized operating systems such as UNIX, based on a computer language called “C”, created an environment where benchmarks could be developed and instruction sets optimized. The RISC “less is more” philosophy was believed to be an elegant way to overcome the trend that microprocessors were becoming increasingly difficult and complex to design and fabricate. The complexity was in large part due to the fact that each new generation had to contain circuits that continued to run the instructions of the last generation. The requirement to retain compatibility with legacy implementations increased transistor counts (and therefore cost) as well as complexity (and therefore design time and system speed). RISC microprocessors were simpler and more streamlined because they relegated much of the lesser-used and more complex operations to software. Thus, sophisticated compilers could translate legacy code to run on newer and faster machines. UC Berkeley's RISC project started in 1980 under the direction of Professor David Patterson. The RISC-I processor in 1982 contained just over 44,000 transistors compared with averages of about 100,000 in newer CISC designs and yet completely outperformed any other single-chip design. They followed this up with the 40,760 transistor RISC-II in 1983, which ran over three times as fast as RISC-I. Even to this day, almost all modern RISC processors are based on the principles of the RISC-II design. In 1981, Professor John Hennessy3 of Stanford University started a similar project called MIPS (for Microprocessor without Interlocked Pipeline Stages4). The project saw first silicon in 1983 and was soon followed by MIPS-X, a more powerful and robust architecture. Based on the promised shown by these projects, Hennessy left Stanford in 1984 to starting a company called MIPS Computer Systems. MIPS’ first design, named the R2000, differed in several respects from the Stanford work (including a floating point coprocessor) but was true to the basic theme of marrying silicon and compiler technology to maximize overall system efficiency. The first processors shipped in 1986 to great commercial acclaim. Within the next two years, similar RISC development efforts produced the Motorola 88000, the INMOS Transputer, the Acorn Advanced RISC Machine (ARM), the HP Precision Architecture (HP-PA) and the AMD 29000. UNIX Operating System Much of the RISC efforts were focused towards optimizing the speed workstations running the UNIX operating system. UNIX has a long and varied lineage. The original work occurred at in the early 1970s at AT&T’s Bell Laboratories by Dennis Ritchey and Ken Thompson. AT&T made UNIX available to universities and commercial firms, as well as the United States government under licenses. In the mid and late 1970s, Berkeley and other institutions used UNIX as a development platform to teach operating system design. Sun Microsystems chose UNIX as the operating system for its line of workstations based on a commercial derivative of the work done by Sun founder Bill Joy while he was attending UC Berkeley. One of the key strengths of UNIX is its portability as evidenced by its appearance on a large number of different platforms. Each vendor (and often each product line from each vendor) ran a version of UNIX such as Domain/OS for Apollo Computers or IRIX for Silicon Graphics workstations, tailored for that machine. The benefit to this circumstance was the ability for UNIX to evolve since each supplier had incentive to innovate, adding extra features as needed to satisfy customer requirements. 3 Hennessy later returned to Stanford and became President of the University in September, 2000 The term “pipeline” refers to each step or stage in the computation process. Processors have multiple steps or “stages” that, if kept full, mean the processor is running at high efficiency. 4 2 Intel and the RISC Wars One strong impediment to growth was the lack of a single “industry standard” UNIX platform that would allow Independent Software Vendors (ISVs5) to develop one software package that could run on machines offered by multiple vendors. As a result, ISV had to create custom versions of their software applications for each new hardware vendor. It was increasingly clear that a unified UNIX platform would greatly increase the market potential for workstations and possibly form a bridge to the desktop world. AT&T initiated the effort to establish a standard UNIX operating system in 1987 by merging features from Xenix6, BSD7, SunOS and System V into System V Release 4 (SVR4). This new release incorporated most of the previous features into one package in an attempt to spell the end of competing versions. Unfortunately, this attempt to unify only galvanized the competition in workstation operating systems. Sun Microsystems Sun Microsystems (Stanford University Network) was founded in 1982 by Bill Joy, Andy Bechtolsheim, Vinod Khosla and Scott McNeely, and began shipping its first Motorola 680x0-based workstation products later that year. The company grew rapidly by offering a range of “standard” UNIXbased machines with much better price-performance than the minicomputers of the day. However, the company quickly became constrained by the price/performance and introduction pace of the Motorola processors. Furthermore, these off-the-shelf devices were available to competitors such as Apollo Computers. Sun felt it needed a path to greater independence. In 1985, Sun began an internal project with Fujitsu to commercialize UC Berkeley’s RISC-II design. The project, called SPARC for Scalable Processor ARChitecture, first went public in July 1987 when Sun introduced its first SPARC-based system, the Sun-4/260, with 10 MIPS performance. Sun's use of a RISC chip in their new machines that demonstrated that the benefits of the RISC approach were real and interest in RISC-based machines skyrocketed. Sun passed Apollo Computer in 1987 as the leading workstation vendor and had aspirations to produce the same type of success that the IBM-PC had seen eight years earlier. At the time it was “common wisdom” that the reason the market for the IBM-PC grew so rapidly was that the PC was an “open” architecture – that is, any vendor who wished to enter the market could purchased standardized components and operating system software from third-parties, begin manufacturing and seeking new customers. This open-ness in turn drove ISVs to port or develop new software applications on the platform which further drove demand for the underlying hardware. In 1988, Sun and AT&T (the originator of UNIX) together with Motorola, Fujitsu, Toshiba, and Unisys announced a new consortium called UNIX International intended to drive the definition and development of protocols to allow different versions of UNIX (e.g. AT&T’s System V and SunOS) to run the same application software. As a public demonstration of commitment, AT&T announced it would acquire 20% of Sun Microsystems. Sun ignited the UNIX workstation market in 1989 with the introduction of SPARCstation I. This platform at 12.5 MIPS8 offered a huge delta in price performance over existing workstations, becoming the first true UNIX workstation to sell for under $9000. To many observers, it appeared that UNIX could begin to compete not only at the server level, but also threatened to become adopted for desktop applications. 5 ISV or Independent Software Vendor is a company that makes application software (e.g. word processor, database, spreadsheet) that runs on top of another company’s operating system. 6 Xenix was Microsoft's UNIX-based operating system for x86-PCs, first announced in 1980. 7 BSD or Berkeley Software Design is the UNIX code base that came from the work done at UC Berkeley 8 Here MIPS is used in the traditional form, Millions of Instructions per Second, a benchmark of performance. 3 Intel and the RISC Wars Sun was working hard to bring new versions of its SPARC processor to market at all priceperformance levels. Sun enlisted the help of several of their major semiconductor vendors to put their own design resources on the SPARC teams in return for volume commitments from Sun as well as rights to market their chips to other systems vendors. 1n 1989, Sun announced the formation of SPARC International that would own the rights to the SPARC trademark and development responsibility for the SPARC processor instruction set (in essence, the architecture). Several disparate groups became active in SPARC International, most notably semiconductor licensees including Texas Instruments, LSI Logic, Cypress Semiconductor, Ross Technology9 and Fujitsu Microelectronics. System manufacturers including Solbourne, Tatung and Tadpole brought early “SPARC clones” to market. Digital Equipment Corporation Founded in 1957 by Ken Olsen, Digital Equipment Corporation (DEC) was based in “the old mill” in Maynard, MA. DEC became the pre-eminent manufacturer of a new class of machines known as minicomputers (as contrasted to the mainframes made by IBM and others). Rather than being ensconced in glass rooms and touched only by IT professionals as mainframes were, minicomputers found their way into the engineering labs of many companies and universities around the world. Olsen was an engineer and the company was run for engineers, by engineers. By the late 1980s, DEC’s PDP (Programmable Data Processor) and the successor VAX (Virtual Address eXtension) series propelled DEC to become the second largest computer manufacturer behind IBM with nearly 100,000 employees. In 1984, DEC entered the workstation10 market. Their proprietary products were designed to be hardware and software compatible with DEC’s large installed bases of VAX minicomputers. In 1988, the VAXstation 2000 was the highest volume workstation in the industry but the company was losing sales to Sun and Apollo whenever standard UNIX operating systems were specified, particularly in DECs core engineering and government markets. With the VAX line of machines aging and UNIX on the rise, DEC was looking to move to a new architecture and committed a ‘bet the company’ play on RISC technology. DECs internal RISC effort (called Prism) was faltering and for the first time in the company’s history, DEC needed outside help. DECs management began a series of high-profile moves to position DEC as a player in the emerging open standards environment. In 1988, in response to the AT&T-Sun alliance and the formation of Unix International, IBM, DEC, HP, Apollo and several other major computer companies form the Open Software Foundation to set a UNIX counter-standard to Unix International. In September of that year, DEC signed an agreement with MIPS for $15M for access to all current and future MIPS architectures, designs and related system software. It also invested an additional $10M for a 5% equity stake. Soon thereafter, DEC launched its first “open systems’ platform – the DECstation 3100 workstation, based upon the brand-new MIPS RISC R3000 processor. MIPS Computer Systems Stanford Professor John Hennessey took a leave of absence in 1984 to start MIPS Computer Systems, Inc. The company’s original business plan was to emerge as a fabless semiconductor company commercializing the Stanford MIPS project and had a desire to become a microprocessor company of the likes of Intel or Motorola. The MIPS R2000 (1986) and R3000 (1988) hit the market to huge technical acclaim and claimed several early design with leading edge workstation companies including Silicon Graphics, Inc Ardent Computers, and Pyramid Technologies. However, volumes generated by these early 9 Ross Technology, Inc was founded by Motorola designer Roger Ross, later purchased by Cypress, then sold to Fujitsu in 1993 10 Workstations are single cabinet, microprocessor-based computers, networked and intended for multiple users. 4 Intel and the RISC Wars sales could not generate sufficient monies to recoup the company’s large development costs. By 1987, they had burned through all but $1 million of their $22M venture financing and were facing a $16M deficit. At that point, the board decided to embark on a major shift in direction and hired computer industry veteran Robert Miller11 as chairman and CEO. Miller struck out on a strategy to make the MIPS architecture an industry standard. He quickly raised $20M from Japan’s Kubota Corporation (a farm equipment manufacturer looking to diversify into high technology) for a 20% stake plus $5M in development funds and a commitment to build a $100M factory for MIPS in Japan. Realizing that others could do a better job at actual semiconductor manufacturing and sales, he signed manufacturing/resale licenses with domestic chip suppliers Performance Semiconductor (then a foundry for MIPS) plus LSI Logic, IDT and BIT12. These smaller companies were soon joined by Japan’s NEC and Germany’s Siemens. In addition to allowing MIPS to focus on system-level product development, these licenses (selling for $1M to $10M plus royalties) provided a much-needed source of revenue. The MIPS strategy struck a chord with computer manufacturers that had relied on their own proprietary systems for sales in the 1970s and 1980s but were seeing sales and market share falling as their own installed base began to wither in the face of open systems architectures from Sun and others. MIPS began courting every hardware company (except for Sun) that was interested in offering a new line of RISC-based workstation machines and offering whatever assistance it could to bring these products to market. MIPS flexible business model meant customers could engage at the component level (buying microprocessors from silicon vendors and compilers from MIPS), purchase boards or remarket entire systems. This model appealed to management at many former computer giants including Control Data, Olivetti, Bull, Prime and Wang, which began to manufacture or OEM MIPS-based systems to their legacy customer base. In addition to early system sales, MIPS sold multi-million dollar architectural licenses to DEC, Sony, Olivetti and others that gave these companies some measure of security and control over their own destiny. Based largely on these license sales, MIPS became profitable and Miller took the company public in 1989. AIM (Apple, IBM, Motorola) IBM’s choice of the Intel 8088 microprocessor as the heart of the original IBM-PC was the single most important event in Intel’s history. The two companies worked closely together and continued to cooperate through the various processor generations. However, IBM had its own internal research and development programs outside of the PC group. IBM’s first RISC efforts started in 1975, preceding the work at Berkeley and Stanford. The “801” started as a pure research project at the Thomas J. Watson Research Center in building 801. They were looking for ways to improve performance of their existing machines, studying traces of programs running on System/370 mainframes and looking at the compiler code. From this project came the idea that it was possible to make a very small and efficient microprocessor running only a few instructions. The 801 effort was first commercialized as the ROMP (1982) launched as part of the, underperforming, high-priced RT PC in 1986. However, the lessons learned through the 801 and ROMP projects, formed the basis of knowledge for the development of the POWER PC architecture which was first used in the popular R/6000 line of workstations running IBM’s version of UNIX called AIX, In 1991 IBM approached Apple to collaborate on a project known as the Common Hardware Reference Platform (CHRP). The intent was to create a new generation of machines to compete with (and supercede) the IBM PC. While CHRP never materialized, the cooperation ultimately led to the Apple 11 Robert “Bob” Miller had spent years at IBM and was a key executive at Data General. IDT is the acronym for Integrated Device Technology, Inc – an SRAM supplier and chief competitor to Cypress Semiconductor. BIT, or Bipolar Integrated Technologies was a fabricator of a high-speed, high-power specialty process known as ECL used in the R6000 MIPS processor. 12 5 Intel and the RISC Wars considering a switch from its existing Motorola 68000 family processors over to IBM’s POWER architecture. Motorola, the leading microprocessor vendor of the early 1980s, had a strong history of success with its 8-bit 6800 and 16-bit 68000 processor lines. The company was second behind Intel as the 68000 family was used by nearly every workstation manufacturer as well as Apple’s Macintosh. Motorola’s RISC effort wasn’t anywhere near as successful. Motorola was struggling with its 88000 RISC design and made the strategic mistake of not providing any backward compatibility with their own famous 68000 series, forcing customers to port to a brand new architecture. Furthermore, by splitting its resources between two architectures Motorola was now losing on both fronts due to time-to-market and performance issues. As Motorola's largest customers of desktop-class microprocessors, Apple brought Motorola into the discussions because Apple felt that Motorola had more experience manufacturing high-volume microprocessors than IBM and was seeking a second source. Thus, Motorola was willing to drop it’s own internal development efforts in favor of joining an IBM, Apple PowerPC partnership. This three-way collaboration became known as AIM, for Apple, IBM, Motorola. The relationship between IBM and Apple grew beyond a microprocessor partnership when they formed Taligent in 1992, an independent company chartered with defining new methods for software development and an entirely new nextgeneration operating system. Microsoft In 1975, Paul Allen and Bill Gates founded “Micro-soft” in Albuquerque, New Mexico. The company’s first product was software for the Altair 8800, a hobbyist computer manufactured by their former local employer MITS. Microsoft's defining moment came in the late 1970s when IBM chose MSDOS as one of the operating systems released with the IBM Personal Computer (PC) in 1981. The early 1980s saw a flood of IBM PC clones, and Microsoft was quick to use its position to dominate the operating system market. In 1985, Microsoft and IBM began jointly working on a successor to the MS-DOS operating system, a project called OS/2. IBM wrote the core instructions while Microsoft worked on the user interface. Together they brought the first product to market in 1987 to limited fanfare. At the time, it became clear that the two companies had differing technical views and marketing priorities. Microsoft and IBM disbanded their joint activities and Microsoft focused its mainstream efforts on Windows13, a user interface residing atop MS-DOS. Because it was a DOS-based platform, Windows had inherent performance and functionality limitations. Microsoft also recognized the threat of UNIX and believed the openly warring UNIX factions created a potential opportunity. In August 1988, Microsoft hired architect Dave Cutler away from Digital where he led DECs Seattle lab developing new chip and operating system architectures. Cutler was to head Microsoft’s internal OS/2 3.0 development effort, renamed OS/2 NT for New Technology. The NT group eventually grew to 200 engineers and scrambled to bring a demonstrable product to market. Bill Gates imposed two technical goals which dramatically increased the project scope: the operating system had to run DOS and Windows applications without modification and it had to be portable – capable of running on Intel as well as other processor architectures. In 1990, Microsoft released Windows 3.0 and later 3.1 to great market acceptance. With the success of Sun in the marketplace, analysts began to question Microsoft’s server strategy. Microsoft first demonstrated Windows NT on an Intel platform at the August 1991 Windows Developer’s Conference. Windows NT was being developed as a high-end server-based operating system. Intel x86, while the 13 Version 1.0 of Windows was shipped as early as 1983 but the program was not deemed commercially mainstream until version 3.0 was released in 1990 and version 3.1 in 1992. 6 Intel and the RISC Wars standard for desktop machines, was not viewed as having enough horsepower for high-end use. It was increasingly clear that the new generation of RISC architectures outperformed existing or planned Intelbased microprocessor implementations. With DEC joining the ranks of the MIPS camp and high-end PC companies (particularly Compaq) showing strong interest in the workstation market, it was time for Microsoft to make a bold move. Intel Corporation In the late 1980s, Intel had already survived several attempts on its core microprocessor business and was trying to use various ploys to starve companies out of the I/O and graphics chip markets. Sun SPARC workstations were being well received in the marketplace. The MIPS R3000 had been garnering industry notice and was a favorite among workstation vendors such as Silicon Graphics. Intel was concerned these new offerings easily outperformed their upcoming 80486 -- and the advantages of RISC were likely to give other processor vendors a technical advantage over Intel's older CISC-based designs. In August 1988, Intel began positioning the unannounced 80486 against the RISC offerings of the day using software as their historic strength. As reported in EE Times: “Even though Intel is months behind the competition and will not begin showing the device to computer makers until early next year, analysts think Intel could still become a force in the workstation and minicomputer market. The timing is a disadvantage, but being Intel is an advantage, and being able to run MS/DOS, OS/2 and Unix is a big advantage.” In November 1988, it was announced that the 80486 would not meet performance expectations: Earlier reports had given the impression that the chip would operate four times faster than the 80386. But John Crawford, chief architect of the 386, pronounced that the 80486 would perform at “twice what you see today in the 80386”. At that performance, the 80486 would still be significantly slower than a number of RISC processors on the market, including SPARC, MIPS, AMD’s 29000 and Motorola’s 88000, most of which had a three of four times performance lead over the 386. Attempting to heap disdain on the ever-growing list of competitors. Intel took to calling these offerings YARPs, (Yet Another RISC Processor). Still, customers, analysts, and the press continued to hound Intel to come up with an answer to the impending threat of these new RISC offerings. At that point, Intel’s tactics became more aggressive. In January 1989, a deliberate campaign of Fear, Uncertainty and Doubt (FUD14) starts with the announcement of a ‘mystery chip’ dubbed the N10: Intel's terse description of the chip's performance states that it can carry out 150 million operations per second. That would be about 30 times more powerful than the company's high-end 80386 microprocessor and five to 10 times faster than the impending 80486. Analysts gushed even without many of the details, "It is unquestionably a terrific chip," said Adam Cuhney, an analyst at Kidder Peabody & Co. "It will make them a major factor to have to contend with in high-end machines." The N10 was unveiled in February 1989 as the i860. The i860 was actually a research project that was never intended to be a stand-alone processor. It was publicly transformed from a proof-of-concept, high-end 80486 math coprocessor into a stand-alone RISC processor with all the promise in the world. However, without any software or operating system, the chip was useless to customers. Intel was using the i860 as a defensive tactic in the RISC wars. Intel quickly attempted to grab the specsmanship high ground by announcing the i860 as the “highest performance RISC processor on the market” with no less than Andy Grove saying, …”this isn’t just a coprocessor chip. This could be the central processor of a super technical workstation.” 14 See Appendix for a detailed definition and historic examples. 7 Intel and the RISC Wars In September 1989, Intel brought another ‘red herring’ out of the labs and into the press with the announcement of the “superscalar15” RISC-based i960MC. BiiN - a company formed by Intel and Siemens to develop fault tolerant high-performance multi-processor computers, designed the i960. The whole project was considered within Intel to have been so poorly managed that the company name was considered to be an acronym for Billions Invested In Nothing. However, the resulting i960MC CPU served its purpose as a RISC “strawman” showcase for Intel even though it was repositioned for use as a controller in hardware applications such as laser printers and network appliances. As marketing tools, the i860 and i960 served their purpose and quieted the analysts for the time. 1990 was a good year for Intel as it launched the much-anticipated 80486 in April. The steady stream of system announcements and performance improvements in the pipeline served to stave off the RISC threat for a few months. However, beginning in December 1990, Intel started leaking advanced information of the “586” (code name P5), in order to take some of the attention away announcements to come. ACE Consortium In January 1991, MIPS and the semiconductor partners announced the R4000 to great fanfare. As the first “super-pipelined16” CPU, the processor represented a significant performance improvement over all competing offerings at the time. Even though the R4000 was not scheduled to become available in volume for a year, word began to leak about secret meetings held in California between Compaq, Digital Equipment Corporation (DEC), Microsoft and Santa Cruz Operations (SCO)17 to define a next generation PC specification based on the R4000. The Hamilton Group (named for the DEC building located on Hamilton Avenue in Palo Alto, CA where first meetings were held) was the seed for what was to become the Advance Computing Environment (ACE). In April 1991, a group of 21 major computer and software companies--led by Compaq, Microsoft, MIPS, DEC and SCO -- publicly announced the development of a set of specifications to define an "advanced computing environment" for the 1990s and beyond. Besides the headlining quintet, the other ACE members were Acer, Control Data, Kubota, NEC, NKK, Olivetti, Prime, Pyramid, Nixdorf, Silicon Graphics, Sony, Sumitomo, Tandem, Wang, and Zenith Data Systems. ACE members agreed to define and develop “ACE-compatible” machines that would run Microsoft’s OS/2 3.0 as well as other UNIX-type operating systems (initially OpenDesktop from SCO). ACE-compatibles would ideally be microprocessor independent but initially based on the Intel and the MIPS R4000 architectures. ACE’s primary benefit is “binary software compatibility” amongst differing hardware platforms. Up until the time, each new application (e.g. word processors, spreadsheets, email programs, etc.) had to be re-compiled18 for each different hardware/operating system combination. ACE 15 Superscalar refers to a chip implementation with multiple execution units on the same chip, thus the potential exists to have more than one operation occurring at the same time. 16 Super-pipelined means that primary operations are subdivided into smaller steps (a deeper pipeline) so clock speed can be increased. Of course, each instruction will now take more cycles to complete (latency), but the processor will still be completing one instruction per cycle (same throughput), and there will be more cycles per second, so the processor will complete more instructions per second (higher actual performance). 17 Santa Cruz Operations, better known as SCO, was the leading shipper of standardized UNIX operating systems for Intel platforms, based upon AT&T’s System V version 4 (SVR4). SCO was partially owned by Microsoft. 18 Compilation or re-compilation is the last critical step in constructing a program to run on a computer. It is the translation of human-readable source code to the “1s and 0s” binary machine language that actually runs on a computer. In order to sell programs on each system manufacturer’s hardware, Independent Software Vendors (ISVs) had to produce a new and different compiled version for each unique OS/hardware combination, a complex process fraught with logistical problems, that had the effect of limiting the number of software programs available for any given machine. 8 Intel and the RISC Wars promised an API (Application Program Interface) that ease software development and cross-platform compatibility with the goal of bringing "shrink-wrapped" software to RISC workstations. Some viewed ACE as primarily a political action: the establishment of a new group of companies, many of them competitors, banding together to challenge the power of Sun Microsystems and, to a lesser degree, Intel, IBM, and Hewlett-Packard. For Compaq, it was also a business strategy for moving beyond commodity PCs into RISC workstations. For Intel, ACE was a direct threat to the monopoly it enjoyed as the only hardware architecture that would run Microsoft operating systems. Intel was doubly threatened as the competing microprocessors (e.g. MIPS and the PowerPC) were outperforming all then-current and then-planned implementations of the x86 architecture. Publicly ACE draws Intel’s disdain. VP of the microprocessor division David House scoffs, "[ACE] is a tempest in a teapot". CEO Andy Grove says, “RISC is a technology of the have-nots”. Rikki Kirzner, senior industry analyst at Dataquest Inc differs: "ACE is a slap in the face to Intel. What ACE really signifies is that PC maker and ACE cornerstone Compaq Computer Corp. has recognized its need to grow rapidly into higher-performance workstations in order to evade its PC competitors and has embraced MIPS's RISC architecture to do it. In essence, Compaq has decided that [Intel's] plans for next-generation 586 and 686 chips are great, but Intel isn't getting there fast enough." Intel begins to scramble. To combat the RISC threat, Intel press announcements (and leaks) at the time promised the P5 would be out by early 1992, with volume production in fall of that year. The chips would drop right into the current 80486 sockets (in some flavors). Clock speeds would start at 66 MHz with higher speeds soon to follow and would be two times faster than 80486 (integer) and 4-10 times faster than the 486 (floating point), far better performer than the MIPS R3000 and comparable to the new R4000. Intel attempted to inject fear and doubt into consortium members. The main public theme was to have buyers think, "Why should I design in a competitors chip, when Intel's will be far faster, and is only a couple months away?" Of course, if Intel fails to release on time, or fails to deliver promised performance, Intel is still better off than doing nothing so there is almost no risk or downside. Against this backdrop, Intel launched the initial “Intel Inside” campaign in July 1991. By the end of the year over 300 companies had agreed to participate. Intel was particularly upset with Compaq. Compaq’s mainstream PC business was growing rapidly as the recession of the time was abating. Compaq volume products depended heavily on Intel for microprocessors sold at favorable prices. Furthermore, Compaq engineering required timely access to specifications and samples for advanced 80486 models as well as the P5. ACE placed great strain on the Intel/Compaq relationship, both public and private. There was no evidence of illegal activity, just implied threats, or little actions and hints -- but Intel’s marketing and sales people did get the message across. Public Moves • Pre-announce P5 (Pentium) • Over promise P5 schedule • Large push in press • Intel Inside Private (non-public) moves • Lower production allocations for current products • Raise prices (or delay price cuts) • Withhold technical information on next generation products. Compaq management weighed the reward of ACE against the risks of losing favorable treatment from Intel and eventually renounced their membership in the consortium. With Compaq out, Intel was able to lessen customer enthusiasm for the ACE consortium and eventually get many members to pull out altogether. The ACE Consortium suffered a fatal blow when Compaq withdrew. By the end of 1992, the alliance quietly disbanded. 9 Intel and the RISC Wars Post Mortem: 1993 – As written in the industry journal “Microprocessor Report”: In response to the RISC threat from ACE, Intel accelerated its P5 program [later named the Pentium] and began showing up at every PC industry conference touting the P5 as being only a few months behind the R4000 and offering higher performance. The PC industry seemed to buy Intel's story. As a result, the ACE effort collapsed, and the P5 -- still a paper tiger -appeared to have triumphed. In the meantime, however, the P5 slipped... by the time the P5 is shipping, [MIPS was] shipping the R4400 with perhaps 50% better performance than the P5. Intel's preemptive strike against the ACE initiative turns out to be, at best, a result of wishful thinking, and at worst, a fraud. (MPR 061503) Intel: The P5, renamed Pentium, was formally announced in October 1993. The earliest Pentiums were released at the clock speeds of 66 MHz and 60 MHz. Pentium architecture chips offered just under twice the performance of an 80486 processor per clock cycle. The fastest Intel 80486 parts were almost the same speed as a first-generation Pentium, and AMD 80486s performed roughly equal to the Pentium-75. Although the official minimum clock speed was 66 MHz, they were only able to build 60MHz devices for the first several months. Delivery problems were compounded when in October 1994, it was discovered that early Pentiums had a problem in the floating-point unit that, in rare cases, resulted in reduced precision of division operations. The Pentium FDIV bug was first denied then downplayed by Intel. Later, they created an exchange program to replace the faulty processors with corrected ones. The Pentium production ramp proved to be slower than anyone’s expectations and dramatically underperformed when compared against earlier public announcements19. Volume shipments of the first buggy chips only started by the end of 1994 and working chips volume shipments weren’t available until the end of 1995. Also, Pentium Overdrive chips (compatible with 80486s) arrived over 18 months later than promised. In 1994, Intel renewed it’s Intel Inside campaign specifically to encourage demand for the Pentium processor, fueled by a $150M war chest – a marketing budget unheard of up to that time for any technology company. In June 1994, Hewlett-Packard and Intel announced a joint research-and-development project aimed at providing advanced technologies for end-of-the-decade workstation, server and enterprisecomputing products. This effort includes development of a full 64-bit instruction set and compiler optimization (IA-64). After several false starts, they were able to deliver the initial chips in 2001 (code named Merced). Intel entered the RISC business a decade after the ACE consortium disbanded. In 1999, the Federal Trade Commission negotiated a settlement with Intel aimed to avoid full antitrust proceedings against the company. The crux of the settlement revolved around Intel’s repeated pattern of a dominant company using its intellectual property as a weapon or an excuse to force deals from customers or weaker partners. Unfortunately for Compaq and the ACE consortium, this settlement “closed the barn doors long after the horses had fled”. Microsoft: Microsoft released the first commercial version of Windows NT 3.1 in July 1993. As promised, it supported Intel and MIPS as well as DEC Alpha and IBM/Motorola PowerPC processors20. Sales on non19 The first P5 chips benchmarked 1.5 times faster than mid-range 80486's against promises of two times faster than cutting edge 80486s. For floating point operations, they benchmarked as 4x better than the 80486 against promises of 10x improvement. 20 Intergraph Corporation ported Windows NT to its Clipper architecture and later SPARC but neither version was sold to the public. 10 Intel and the RISC Wars Intel processors never came close to meeting expectations. Windows NT 4.0 was the last major release to support Alpha, MIPS, or PowerPC, though development of Windows NT 5.0 (Windows 2000) for Alpha continued until 1999 when Compaq (who had purchased DEC) stopped support for Windows NT on that architecture. In October 2001, Microsoft released Windows NT 5.1, better known as Windows XP as their sole and unified platform for desktops and servers. The letters "XP" originate from the word "Experience”. While Windows remains the dominant operating system for desktop machines, the Windows/UNIX battle still rages to this day. UNIX: By 1993 most UNIX vendors had changed their commercial variants of UNIX to be based upon AT&T’s System V Release 4 (SVR4) but many added features from BSD as well – so no single standard ever emerged. Shortly after SVR4 was placed into the market, AT&T sold all its rights to UNIX to Novell. In 1994, Novell decided to split and sell off their bundle of UNIX-related assets. The UNIX trademark and the certification rights were sold to the X/Open Consortium, an industry group created to define a "UNIX Standard". Finally X/OPEN and OSF/1 (a competitor to the SVR4 standardization) merged, creating the Open Group. Various standards by the Open Group now define what is and what isn't a "UNIX" operating system. In 1995, the business of administration and support of the existing UNIX licenses plus rights to further develop the System V code base were transferred to the Santa Cruz Operation (SCO). Novell retained the core copyrights, veto rights over future licensing activities of SCO, and 95% of the licensing revenue. Linux, a version of UNIX developed by Linus Torvalds, has seen a huge rise in popularity and although it is open source, companies such as Red Hat and SUSE (purchased by Novell) have emerged to provide commercially supported versions. In 2000, the Santa Cruz Operation sold its entire UNIX business and assets to Caldera Systems. In March 2003, Caldera, renamed the SCO Group (SCO) filed a lawsuit against IBM claiming that IBM had contributed portions of SCO's intellectual property to the Linux kernel in violation of IBM's license to use UNIX. Additionally, SCO sent letters to a number of companies warning that their use of Linux without a license from SCO may be actionable, and claimed in the press that they would be suing individual Linux users. To date, there has been no proof of SCO's claims of copied code in Linux. IBM/Motorola: By early 1993, IBM had completed the development of a full line of PowerPC based desktops but had to wait since Microsoft’s Windows NT was still months away from first release. Accordingly, and further because IBM had developed animosity toward Microsoft, IBM decided to rewrite OS/2 for the PowerPC. It took IBM two years to rewrite OS/2 for PowerPC, and by the time the operating system was ready, the market for OS/2 on PowerPC had evaporated. Apple continued work on a new line of Macintosh computers based on the PowerPC, and eventually released the 601-based Power Macintosh in March 1994. Apple continues to successfully use the PowerPC to this day in all their servers, desktops and laptops -- their machine designations tied to the successive generations of the PowerPC architecture (G3, G4, G5 with G for “generation”). The IBM/Apple Taligent operating system joint venture continued on. As time went by, priorities shifted and the project focus changed from operating systems to “rapid application development”. The Taligent OS became a layer that could sit on top of any modern operating system and provide numerous services to applications software, thereby shortening the development cycle. Early in 1994, HewlettPackard, which had been doing its own work with object-oriented technology, made the decision to become 11 Intel and the RISC Wars a Taligent partner. After the tragic death of its CEO in late 1995 Taligent faced a reduction in force and in the spring of 1996, Taligent became a wholly owned subsidiary of IBM. In 1998, the company was formally dissolved with the engineering teams becoming IBM employees. Motorola had become less reliant on the microprocessor business as they began to focus on the production of cellular phones. In 2003 they announced they would spin off their semiconductor product sector as Freescale, which began publicly trading in July 2004. In addition, Motorola was unable to keep pace with IBM’s performance on the PowerPC architecture and in 2003, Apple turned to IBM to provide PowerPC chips for its new G5 desktop computers. Sun/SPARC: With annual revenues exceeding $11B, Sun continued to prosper as the largest supplier of engineering workstations in the market although financial results have suffered of late. SPARC remains the architecture used in nearly all Sun workstations to this day. Development continued as SPARC family grew to include SuperSPARC (1992), SPARClite (1992) for laptops and UltraSPARC (1995). The market for SPARC clones never really developed beyond the curiosity stage as Sun undercut pricing for any potentially large purchases. Also, while Sun promoted the concept of cloning Sun workstations with SPARC International, it failed as an "open standard" because Sun retained grant-back rights to any architectural or implementation improvements, eliminating the incentive for companies to invest efforts to differentiate themselves. One by one, the SPARC semiconductor partners lessened their commitments to SPARC and Texas Instruments is currently the sole foundry for the processor family. MIPS/SGI: MIPS continued to sign additional semiconductor licenses through the first part of 1991 but had largely saturated the market as the total number of licensees approached 10. These high-margin licensing deals had sustained the company while it attempted to build system sales. In the second half of 1991, the company began to sustain losses, which shook investor and customer confidence. Silicon Graphics, Inc (SGI) was one of MIPS' largest customers. The company designed and built high-performance three-dimensional graphics computers – for uses including the development of films such as Jurassic Park. As MIPS began to weaken, SGI became worried that it may not be able to depend on MIPS as a critical supplier of intellectual property for its line of workstations. In March 1992, MIPS and SGI announced a merger valued at about $400 million. MIPS sales continued to fall as customers questioned whether the company, after the merger, would continue to be an industry-neutral chip and software supplier. After the announcement, MIPS reported disappointing firstquarter earnings and its stock price fell sharply. Partly to reflect MIPS's poor performance in the first quarter and the dissolution of ACE, SGI and MIPS renegotiated the terms of the merger. In the end, SGI bought MIPS for about $200 million, half the value of the original deal. In 1999, newly appointed SGI CEO Rick Belluzzo believed that the proprietary hardware offered less of a competitive advantage and de-emphasized MIPS in favor of Intel-based platforms. SGI organized MIPS as a wholly owned subsidiary called MIPS Technologies Inc. (MTI). After a change in management in the mid-1990s combined with a decision to migrate to the Intel Itanium architecture, SGI decided to spin out MTI as a vendor of embedded microprocessors. MTI had an initial public offering on June 30, 1998. In 2000, SGI divested all its interest as stock dividend to its stockholders. Belluzzo left SGI in February 2001 to become the COO of Microsoft. Today, MIPS Technology, Inc is a profitable public company that licenses the Intellectual Property (IP) for it’s 32-bit and 64-bit microprocessor designs to be used as “cores” in chips for a wide variety for electronic systems. MIPS-based chips are used in applications raging from digital set-top boxes, 12 Intel and the RISC Wars digital televisions, DVD recorders, digital cameras, laser printers, network routers as well as the Sony PlayStation and Nintendo 64 game consoles. MIPS Semiconductor Suppliers: Because of the collapse of the ACE consortium as well as the purchase of MIPS by SGI, chip suppliers started losing interest in continuing to follow MIPS mainstream implementation path for the architecture as a mainstream CPU. Since they had rights to the underlying MIPS architecture, each vendor was free to choose their own path to optimize revenues and strategic position. Including NEC designing their own private versions for themselves and for semiconductor customer Nintendo, IDT and LSI Logic designing versions for the controller market and LSI Logic creating an ASIC cell for customer designs. Performance Semiconductor rapidly disappeared from the market. The semiconductor alliance partners each began to seek their own destiny. Siemens, Performance Semiconductor and BIT dropped the MIPS line. IDT began to make low cost versions for embedded applications such as laser printers. LSI Logic created a version of the MIPS R3000 in its ASIC cell library as well as customized versions for laser printers, network controllers, graphics terminals and others. NEC was able to leverage their memory business and become the primary semiconductor source for SGI. DEC/Compaq: In 1992, DEC abandoned their MIPS-based workstation line in favor of machines designed around their internal RISC project, code-named Alpha. Despite the fact that DEC’s Alpha generally outperformed most other RISC processors, Digital was never able to recapture the market share or profits it enjoyed with the VAX line of minicomputers. In May 1997 DEC sued Intel for allegedly infringing on its Alpha patents in designing the Pentium chips. The case settlement resulted in DEC's chip business being sold to Intel and its networking business being sold to Cabletron. In January 1998, Compaq purchased DEC for $9.6 billion, valued largely for its service business. At the time, this was the largest computer industry acquisition. Compaq, which also purchased Tandem, was only marginally successful at integrating and leveraging these new assets. Hewlett-Packard purchased Compaq for $20 billion after an extended shareholder battle that ended in May 2001. 13 Intel and the RISC Wars Appendix: FUD Fear, Uncertainty and Doubt (FUD) is a marketing technique used by companies with a large market share and used when a competitor brings a product to market such that the incumbent product is no longer competitive. The entrenched player whose market position is threatened then resorts to disinformation as a competitive weapon. IBM was the first to practice FUD on a large scale. Gene Amdahl defined the term after he left IBM in 1970 to start Amdahl Corporation, a company whose stated mission was to compete with in the entrenched IBM mainframe market with compatible machines that ran faster and cost less: "FUD is the fear, uncertainty, and doubt that IBM sales people instill in the minds of potential customers who might be considering Amdahl products. The idea, of course, was to persuade buyers to go with safe IBM gear rather than with competitors' equipment. This implicit coercion was traditionally accomplished by promising that Good Things would happen to people who stuck with IBM, but Dark Shadows loomed over the future of competitors' equipment or software." By spreading questionable information about the drawbacks of less well-known products, an established company can discourage decision-makers from choosing those products over its wares, regardless of the relative technical merits. This is a recognized phenomenon, epitomized by the traditional axiom of purchasing agents that "nobody ever got fired for buying IBM" equipment. The result is that many lower and middle level decision managers recommend products that they know to be technically inferior because upper management is more likely to recognize the brand and therefore be deemed less risky. FUD implementation comes in two phases: First, is a campaign of scare-mongering via 'gossip channels' intended to cast a shadow of doubt over the competitors offerings and make customers think twice before using it. Second, is to approach each customer and reassure them that the status quo is safe and to pre-announce a ‘next generation’ capability intended to trump the new announced competitive offering. FUD can be used to offhandedly 'smear' criticism or legitimate debate, even in cases where the allegations are without merit or are merely implied; this tactic is often used in cases where the initial publicity surrounding claims of FUD is likely to vastly overshadow any subsequent retraction. Such an arbitrary usage is a general type of logical fallacy known as Ad hominem circumstantial. Opponents claim that the practice of spreading of fear, uncertainty, and doubt is an unethical marketing technique that large corporations consciously employ. 14 Intel and the RISC Wars RISC Camps: Architecture MIPS (ACE Consortium) Motorola 88000 HP PA-RISC IBM PowerPC Sun SPARC (SPARC International) Semiconductor Suppliers Bipolar Integrated Technologies (BIT) IDT LSI Logic NEC Performance Semiconductor Siemens Toshiba Motorola Hitachi, Samsung IBM Motorola Cypress Semiconductor Fujitsu LSI Logic Ross International Texas Instruments System Adoptees Acer, Compaq, Control Data, Digital Equipment (DEC), Kubota, MIPS, NEC, NKK, Olivetti, Prime, Pyramid, Siemens, Nixdorf, Silicon Graphics (SGI), Sony, Sumitomo, Tandem, Wang, Zenith Data Systems Data General HP IBM Apple Sun plus: Force Computers, Fujitsu, Force Computers, Fujitsu, Solbourne, Tadpole, Tatung, Toshiba Relative Stock Prices: (Note: Use 1988 to 1992) 15 Intel and the RISC Wars Selected Microprocessor and Operating System Introductions: 1971 to 1999 Year Intel 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 4004 8008 8080 1984 1985 1986 Other Processors (RISC in BOLD) Microsoft AT&T: UNIX Motorola: 6800 Digital Research: CP/M IBM: Starts 801 8085 Motorola: 6809 8086 8088 80186 80188 80286 BSD Unix Motorola: 68000 Motorola: 68008 Berkeley: Berkeley: Motorola: Stanford: Motorola: RISC I RISC II 68010 MIPS 68020 80386 Stanford: MIPS-X MIPS: R2000 IBM: ROMP Motorola: 68030 Sun: SPARC AMD: 29000 Motorola: 88000 MIPS: R3000 1987 1988 1989 80486 i860 MIPS: R6000 HP: PA-RISC 1990 i960 1991 “Intel Inside” AMD: 29050 IBM: RS/6000 MIPS: R4000 Xenix MS DOS 1.0 1992 1993 Pentium 1994 1995 Pentium Pro 16 DEC: Alpha MIPS: R4400 Sun: SuperSPARC Sun: MicroSPARC IBM: PowerPC 601 MIPS: R8000 IBM: PowerPC 604e MIPS: R10000 Sun: UltraSPARC DEC: VAX/VMS Apollo: Aegis AT&T: System V UNIX MS DOS 2.0 MS DOS 3.0 Windows 1.0 Free Software Foundation OS/2 IBM: OS/2 Microsoft hires DEC architect Dave Cutler to start OS/2 3.0 AT&T buys 20% of Sun. Forms Unix International. IBM, DEC, HP, Apollo: OSF Sun: Sun OS Next: NextStep AT&T: Unix SVR4 Digital Research: DR-DOS Windows 3.1 MS DOS 5.0 Windows NT demonstrated 1996 1997 1998 1999 Other Operating Systems Sun: Solaris DEC: Open VMS ACE formed Linux first released MIPS and SGI merge ACE collapses MS DOS 6.0 Windows NT ships Windows 95 Windows NT 4.0 Pentium MMX IBM: PowerPC G3 IBM: PowerPC G4 Windows 98 Windows NT 5.0 (Win 2000) Apple: OSX server Intel and the RISC Wars Sources: Chow, Paul. “The MIPS-X RISC Microprocessor” Kluwer Academic Publishers 1989 Liu, Juliana. “Hennessy's technological experience at MIPS link Stanford to Silicon Valley” Stanford Daily 4/26/ 2000 Masanell, Yoffie, David B., Mattu “Intel Corporation: 1968 – 2003”. 2002 Case Study, Harvard Business School. Microsoft Corp “A Brief History of the Windows NT Operating System”. Fact Sheet. October 1998 Russinovich, Mark. “Windows NT and VMS: The Rest of the Story” Windows IT Pro Magazine, December 1998 Yoffie, David B. “Strategic Management in Information Technology”. Harvard Business School Prentice Hall, 1994 Websites: Every, David K. History of ACE - Alternatives to the PC - 1999 http://www.mackido.com/History/HistoryOfAce.html BYTE June 91 http://www.byte.com/art/9606/sec5/art1.htm MIPS Alliance on UNIX servers http://febcm.club.fr/english/mips_alliance.htm What is FUD? http://www.cavcomp.demon.co.uk/halloween/fuddef.html Wikipedia: FUD http://en.wikipedia.org/wiki/FUD History of OS2 Warp: http://www.os2bbs.com/os2news/OS2Warp.html 17