IDC Opinion IDC believes that 64-bit architecture is an inevitable evolution in technology that will be important but not dominant in the second half of the decade. The transition from 16- to 32-bit system architecture was quite different from the upcoming transition from 32- to 64-bits. In general, there is not the pent-up demand for the features of 64-bit architecture as there was in the transition to 32-bits, most notably in address space requirements. On the supply side, markets where 64-bit capabilities can provide a competitive advantage, (i.e, high performance technical systems and midrange systems) will lead the migration. Digital and SGI/MIPS are seen as the leaders in this evolution. IDC believes that competitors in these markets have a window of two to three years to respond. In the desktop markets, however, volume and cost issues enter the equation much more rapidly than before. When 32- and 64-bit processors and systems converge in price for the mass market, the transition will be accelerated. IDC expects this to happen a few years after Intel's P7 becomes available _ when its lower-cost siblings hit the street. On the user side, IDC expects the high-performance technical users to tap into 64-bit architectures first, due to the leading-edge nature of their applications. The transition of commercial users will take much longer, however, due to the more conservative nature of this group and a reliance on higher-level software tools, which will take time to evolve to take advantage of 64-bit features. We believe ISVs will tend to support the architectures from which they derive revenue, regardless of the 64- or 32-bit question. For instance, major ISVs will port to Digital's Alpha platform not because it is a 64-bit architecture but because they believe that Digital is a major player in the market. The majority of ISVs will port existing 32-bit software to the 64- bit platforms without taking advantage of the 64-bit features. What Does 64-Bits Mean? In order to differentiate and categorize processors as 32-bit or 64-bit (or any other size, for that matter) we must first realize that there is a hardware definition and a software definition. To illustrate the difference, many PCs utilize an Intel 486 chip, a 32-bit chip by any hardware definition. However, most of these PCs run DOS or Windows, which is a 16-bit software environment. Therefore, these PCs would be considered 32-bit according to a hardware definition but 16-bit by a software definition. There are many components of a system that could be classified as 64-bit. These include: Integer register size * Floating point register size * Virtual address space (determines maximum memory any one process can use) * Physical address space (determines maximum physical memory that can be addressed) * Data bus size * Ability to manipulate 64-bit data types * System software environment IDC does not believe that the data bus size alone should be used in determining whether a system is 64-bit or not. Likewise, we do not believe that the ability to manipulate 64-bit data types makes a system 64-bit. Many 32-bit architectures allow for manipulation of 64-bit data types. Sixty-four-bit floating point registers are common throughout 32-bit architecture machines and therefore are not part of our 64-bit definition. Sixty-four-bit data types provide a performance boost as well as increased precision, but by themselves do not make an architecture 64-bit. Some supercomputer architectures utilize this type of technology and are often regarded as 64-bit but do not meet IDC's definition. We do believe that the integer register size and the virtual and physical address size, in conjunction with the software that controls the system, determine whether a system is 64-bit. IDC uses a "purist" technology-based approach to categorizing processors and operating systems. We stipulate that 64-bit hardware must have all of the following characteristics: * 64-bit integer registers * Significantly greater than 32-bit flat, unsegmented virtual address space * Significantly greater than 32-bit flat, unsegmented physical address space * No "hoops" to jump through to obtain these capabilities (i.e., no segmentation) The reason for the "significantly greater than" terminology is to allow architectures flexibility to utilize specific bits for special purposes. Sixty-four-bit integer registers are necessary to deal with 64-bit logical and physical addresses efficiently. Our definition of 64-bit operating systems and software is based on our definition of 64-bit hardware. We stipulate that 64-bit software must: * Allow access to the 64-bit hardware features (especially virtual addressing) * Utilize 64-bit data types (i.e., 64-bit integers, compilers manipulate 64-bit quantities, etc.) * Not impose "hoops" to jump through to obtain these capabilities (i.e., no segmentation) A similar definition can be made for 16- and 32-bit systems. Comparing Transitions When comparing the transition from 16- to 32-bit computing, with the upcoming transition from 32- to 64-bit computing, the following differences are apparent: * There was a need for greater than 16-bit addressing before 32-bit systems were generally available. * The increase in address space varies dramatically in going from 16- to 32- to 64-bit systems. Table 1 illustrates the relative size of the physical and virtual address spaces for 16-, 32-, and 64-bit systems: Note that in going from 8- to 16-bits, as shown in Table 1, the address space is 256 times larger. In going from 16- to 32-bits the address space is a very comfortable 64,000 times larger. In going from 32- to 64-bits the address space is a whopping 4,000,000,000 times larger than before! Therefore, address space demand had to increase 256-fold for 32-bits to become a requirement. Similarly, demand must increase by 64,000 times before a 64-bit architecture is needed (when the 33rd bit is required). The transition from 16- to 32-bits was quite different from the upcoming transition from 32- to 64-bits. The single biggest difference is that for the 16- to 32-bit transition there had been a great need for more than 16- bits of addressing _ both virtual and physical. Note that the demand for 32-bits occurs essentially when the 17th address bit is needed. Likewise, demand for 64-bits occurs when the 33rd bit is required. Many different schemes or kludges were developed to get around these issues when need for the next bit or two began to surface. Examples include overlays, segmentation, and use of special hardware to translate virtual addresses to physical addresses. When 32-bit systems were first introduced (e.g., VAX/VMS, DG MV, etc.) there was a market waiting for them. The desktop PC market is, in many ways, different from the minicomputer/mainframe market that preceded it. However, the transitional issues are remarkably similar. Similar mistakes were made and similar kludges were done _ the time frame is just extended. IDC expects similar schemes will be used to delay transitions to 64-bit technology. As illustrated in Figure 1, IDC believes that demand for 64-bit addressing begins to be real in 1995, led by high-performance systems, several years after the technology is available. Note that this is the first time that the technology is ahead of the need. What Does 64-Bits Enable? From an architectural standpoint, a 64-bit architecture provides advantages in two areas: computation speed and addressing. The former occurs simply because computational instructions can work on data types (integers, floating point, pointers, etc.) that are twice as large within one clock cycle. This increase in speed not only helps technical applications, which are typically more demanding computationally, but also commercial applications that perform numerous data comparisons, translations, and block data movements. While 64-bit technology can provide some performance boosts, it can also generate additional overhead. Any performance differences are application-specific. Sixty-four-bit technology does not imply a universal doubling in performance. The advantages of 64-bit addressing capabilities are not apparent when looking at the address space requirements of the average program. For the vast majority of these, 4GB of addressing will be sufficient for many years. Instead of driving 64-bit functionality into existing applications, we believe that the practical application of 64-bit architectures will be for emerging applications with specific characteristics such as: * Large database, file, and I/O processing (e.g., imaging) * Complex layered software (e.g., object management) In order to properly move up the application food chain, we believe the use of larger address spaces and other 64-bit features will need to be implemented in system software first, particularly operating systems and database management software. The system software functionality that would need to be addressed to solve each of these problems is related, but different. Large Database and I/O Multiuser operating systems provide a logical view of memory and I/O via virtual addressing, paging, device mapping, logical I/O interfaces, and file systems. All of these place additional overhead on a system. This overhead also tends to get exponentially larger as the system capacity (memory, disk) gets larger, and/or the number of programs vying for these resources increase (see Figure 2). This is why many suppliers have needed to rewrite the memory management and I/O management segments of their operating systems in order to support the larger memories (512MB+), storage (300GB+) and process counts (1000+) associated with larger capacity systems. Complex Layered Software An increasing trend for building applications is to use modular components that are pieced together and customized to match specific requirements and functions. A problem with this approach is that message passing among software components is time consuming and absorbs operating system resources. This slows the performance of the application and the entire system. Microkernel-based operating systems will provide the architectural foundation from a software standpoint for making software components modular, but will face similar performance issues. If such architectures are to become accepted, fast memory-based message passing must be used. The key bottlenecks in system software, which slow the execution of these functions, are translation schemes (i.e., multiple table look ups, logical mapping, etc.). As these tables contain more information the overhead increases on system resources, especially CPU cycles. With a full 64-bit hardware architecture (64-bit data manipulation and addressing) the additional bits can be used to map devices and system software via logical addresses as shown in Figure 2. To gain even higher performance, the entire I/O subsystem could be mapped to look like a single, large set of multilevel caches (see Figure 3). In such a scheme, table lookups will be less frequent, freeing the CPU to do the computational work it was meant to perform. Inhibitors for the Movement to 64-Bit The aforementioned methods of solving particular problems with 64-bit system software make a bold assumption: the world is ideal. The realities of today's market dictate that customers will require a high level of consistency with existing applications and environments. Therefore, suppliers will need to spend increasingly valuable resources ensuring this consistency and compatibility in order to keep customer bases happy and products stable. A list of the potential inhibitors for 64-bit computing includes the following: * Compiler technology * Standards and the growing number of APIs/ABIs * Legacy systems and system software The advancement of compiler technology is the first hurdle to overcome, and probably the easiest. Beyond the ability to effectively utilize 64-bit features within the hardware and software architecture, the challenges facing compiler makers will be in maintaining compatibility with existing architectures and compilers (i.e., hiding the intricacies from the programmer) and installing effective intelligence on when and where to use 64-bit capabilities. We do not see these as large problems to overcome, and thus do not envision compiler technology issues dramatically inhibiting the movement to 64-bit computing. Compliance with industry standards and the ability to convince software developers to write to specific APIs for 64-bit architectures looms as another potential problem. Most formal standards are created out of existing APIs. At the high-level programming interface level, developers will not see a difference in most system calls and libraries. However, implementations of these standard interfaces on 64-bit architectures will have some differences in order to take advantage of the more advanced features. Thus code would not be completely portable. Some of this phenomenon occurs today among different interface implementations, but we believe the problem will be exacerbated with the proliferation of 64-bit architectures. Perhaps the largest problem will be in providing compatibility with existing systems and software. This is a problem for both suppliers and users. Suppliers have a choice of either providing complete upward compatibility when moving to a 64-bit architecture (e.g., what Sparc International is promising), or changing the architecture and pushing the installed-base through a migration (e.g., Digital with Alpha). The tradeoffs of these options are: * Provide upward compatibility. Minimize the risk to the installed base, but face a slower implementation of 64-bit features and added R&D costs in maintaining existing interfaces. * Change the architecture. Expose the installed base to change (which could be to another supplier), yet bring 64-bit functionality to market faster and have this as a technological differentiator. Users, in general, prefer to minimize their risks and continue to receive steady technological improvements in the platform, while maintaining compatibility with the architecture they use today. This is one of the primary reasons why we believe the movement to 64-bit computing will be a slow, incremental movement rather than a rapid turnover. Market Dynamics for 64-Bit Acceptance We believe the dynamics that will dictate the movement to 64-bit computing will be determined differently among computer suppliers, ISVs, and users. Each group has its own set of motivations, facilitators, and hurdles that will affect the movement to 64-bit architectures. Computer Suppliers Among the computer suppliers, the factors that will drive 64-bit computing are: * The increased need to maintain some level of differentiation. An early jump to a 64-bit platform will provide performance edges, a perception of technology leadership, and the ability to offer greater headroom for future platforms. * Greater engineering expertise. Development and release cycles for microprocessors and systems have shortened due to a greater level of engineering talent and expertise. Thus, the ability for suppliers to introduce a new 64-bit architecture is greater today than during the transition from 16- to 32-bits. Computer suppliers also face a number of obstacles that will hamper the movement to 64-bit architectures, of which the most noteworthy are: * Generating volume. The only way to generate large volume for a microprocessor is the desktop market. However, the commercial desktop market is only just beginning to fully utilize 32-bit systems (e.g., OS/2, Windows NT, Unix), and thus will not require 64-bit architectures for some time. It is difficult to financially justify an investment in developing a 64-bit microprocessor with limited hope of penetrating the volume desktop market. As with the movement to 32-bits, IDC expects that 64-bit hardware will be available and even plentiful on the desktop long before software begins to take advantage of the capabilities. * Installed-base disruption. Many 64-bit systems may not be completely compatible with existing 32-bit systems. Installed-base migration programs are typically long and costly to suppliers. Software Vendors ISVs typically are driven by two factors: volume and value-added functionality. New functionality or support of specific platforms is added in a reactive fashion when enough customers demand it. Porting to a particular platform may also occur if a revenue opportunity exists. We believe ISVs will tend to support the architectures from which they derive revenue, regardless of the 64- or 32-bit question. For instance, major ISVs will port to Digital's Alpha platform not because it is a 64-bit architecture, but because they believe Digital is a major player in the market and any platform the company produces will generate enough volume for ISVs to get return for the porting investment. As a result of extensive interviewing, IDC believes that the majority of ISVs are not seeing demand among their users for specific 64-bit functionality, nor do they see many inherent capabilities of which their software products can take advantage. The majority of ISVs will port existing 32-bit software to the 64-bit platforms without taking advantage of the 64-bit features. In fact, most vendors porting to Digital's Alpha systems are using OpenVMS, a 32-bit OS. In the future, as more customers demand functionality that can be more efficiently provided through 64-bit features, ISVs will modify products to better utilize the 64-bit platforms. However, some software vendors may choose to use 64-bit as a differentiator at some time in the future. For details on the impact of 64-bit technology in the CAD/CAE/CAM markets see 64-Bit Architecture: What It Will Mean for CAD/ CAE/CAM (IDC #6445, March 1992). Users The demand for 64-bit computing will originate with high-performance technical users rather than in the commercial market. It is the technical users who typically lead the way to advanced technologies. Technical users also tend to turn over systems fairly quickly, porting or recompiling to reap the benefits of more advanced platforms as introduced. These users tend to model "Grand Challenge"-type problems, which typically can have huge data set sizes and therefore can always use more address space. Commercial users tend to be more conservative in nature, valuing service, support, and stability more so than the technical user that buys more for advanced features. The commercial user also uses more high level software (e.g., middleware) to piece together applications. Thus, 64-bit platforms would not have value to the mass commercial market until the higher-level software utilized 64-bit features. Who Will Be Selling 64-bit Technology and When? 64-bit technology will rollout from most of the major vendors over the next three years. Table 2 illustrates the major players and our expectation of the rollout. Silicon Graphics/MIPS - MIPS R4000 SGI/MIPS was the first vendor to ship 64-bit hardware-based systems. SGI's Iris Crim-son workstations utilize the MIPS R4000, the first 64-bit chip. It runs Irix, SGI's version of Unix, and runs in 32-bit mode. IDC expects Windows NT, also a 32-bit OS, to be made available on SGI's R4000- based machines as well as R4000-based systems from other manufacturers. Digital's DECstation line utilizes MIPS R3000 processors and is scheduled to get R4000 upgrades in 1993. These systems will also run 32-bit operating systems and will not get the 64-bit version of OSF/1 that DEC plans for its Alpha AXP systems. IDC does not expect a 64-bit operating system to be available for the R4000 until the 1995 time frame; that OS is likely to be an Irix evolution from SGI. With the demise of the ACE initiative, prospects for proliferation of the R4000 have lessened. However, there is still significant support by European and Asian vendors (especially Sony and NEC) even though U.S. vendor interest (most notably Compaq and Digital) has waned. Digital _ Alpha AXP Digital is the company that is most aggressively marketing 64-bit as an innovation and differentiator. As such, we have the most information about Digital's 64-bit strategies and products. However, the real impetus for Alpha is RISC, not 64-bit, as Digital needed to revamp its entire line due to the VAX running out of gas. In the process of designing a new architecture for the next 25 years, Digital correctly identified that sometime in that 25-year time frame 64-bits will be a requirement. That sometime is not quite here today. Rollout of operating systems for Alpha is as follows: * Digital's OpenVMS will ship with the first Alpha systems in December 1992 but will be largely restricted to run in 32-bit mode. Access to 64-bit data types will be provided but virtual address space will remain 32-bit. Digital has not stated when OpenVMS will be fully 64-bit, but that it will be in the future. IDC expects this functionality in the 1995 time frame. * Microsoft's Windows NT will be available in approximately the same time frame as it is made available for Intel (sometime in 1993). At this point, Microsoft has not made any statements about its 64-bit directions, except that NT does utilize 64-bit file pointers (also on the Intel version). In fact, Microsoft is boldly promoting the fact that NT is 32-bit, which is considered "advanced" in the markets where Micro-soft competes. * Digital's DEC OSF/1 will be the first 64-bit OS available for Alpha and should ship to customers in the first half of 1993. A developer's version will ship in conjunction with the first Alpha shipments in December 1992. * Other Unix offerings: Encore also announced that it will develop a 64- bit version of Unix SVR4 ES/MP for its high-end Alpha-based commercial systems. Availability for these systems and the 64-bit version of SVR4 has not yet been disclosed. Olivetti will also be providing a version of SVR4 for its Alpha-based systems. IDC expects that Digital will announce lower-cost Alpha systems in 1993. Lower-cost systems, the availability of Windows NT, and the ability of the company to leverage its relationship with Microsoft, are key to the future success of Alpha beyond the VAX installed base. This potential is largely based on a 32-bit operating system (NT) for the foreseeable future. Success in the Unix space is dependent on Digital's rationalization of its Unix strategy. For a closer look at Digital's Alpha operating system strategies see Is Alpha DEC's Ticket to Regain Lost Unix Market Share? (IDC #6699 June 1992). Sun Microsystems and Others _ Sparc Sparc International has approved Sparc version 9, which extends the Sparc architecture to 64 bits. The new version was influenced by startup HaL Computer, expected to be the first Sparc manufacturer to introduce a system utilizing 64-bit Sparc technology. IDC expects this introduction in 1993. HaL will likely ship a 32-bit version of SVR4 and will make 64-bit enhancements over time. Sun Microsystems, by far the largest and most influential Sparc system manufacturer, will likely follow HaL but lag by at least a year. IDC believes that Sun has other much higher priority projects, such as making substantial progress in commercial servers and introducing object technologies. On the software side, IDC does not expect a 64-bit version of Solaris until 1995 at the earliest. Although Sun is the volume leader in workstations, HP is currently the performance leader and therefore likely to be the workstations utilized in the most demanding environments _ those likely to need 64-bits first. IBM/Apple/Motorola _ Power, PowerPC PowerPC, an evolution of IBM's Power architecture utilized in the RS/6000, is a joint project between IBM and Motorola. Today's RS/6000s are 32-bit but have 52-bit addressing via a segmentation scheme. We expect that the PowerPC chips will be 32-bit and that IBM will continue to enhance its Power derivative with 64-bit capability targeted at the high-performance technical market. The first PowerPC chips should see the light in 1H93 with an MP version available in 2H93. The 64-bit derivative from IBM will likely debut in 1994, followed in the 1995 time frame by a version of AIX that will exploit the 64-bit capabilities. Hewlett-Packard, PA-RISC HP's PA-RISC is the architecture utilized in the company's 9000 Series 700 workstations as well as its 9000 Series 800 commercial Unix systems and its 3000 line, which runs MPE/iX. In addition, PA-RISC will be the first RISC architecture to run native NetWare. Given HP's performance leadership position at the high end of the technical workstation market, IDC expects HP to see demand for 64-bits before most other manufacturers. PA-RISC is a 32-bit architecture. In our opinion, HP has approximately three years before 64-bit demand becomes significant. Therefore, the company has enough time to adapt PA-RISC to 64-bits and to bring product to market. If HP does not articulate its direction vis-a-vis 64-bit within the next 12-18 months, the company stands to lose its perception of technology leadership. IDC expects that HP will enhance HP-UX, its Unix OS, to take advantage of 64- bit functionality within a year of making available systems based on 64-bit architecture. Intel, P7 IDC expects that Intel will rollout its P6 processor (a 32-bit chip) in the 1994 time frame, followed closely by P7, a 64-bit chip, in the 1995 time frame. Both are members of the X86 family. Systems utilizing each will follow processor introduction by a year at most. Most P7 systems will run 32-bit software environments through the end of the decade. We expect that 64-bit Unix operating systems will make their way onto the X86 in the 1997 time frame. Outlook/Assumptions Our outlook (see Figure 4) is derived from careful consideration of the following factors: General market assumptions * This outlook is based on the market dynamics we've discussed from both the supply and demand side. * We believe the history of the 16 to 32-bit migration that began in the late 1970s serves as a useful guideline for the 32- to 64-bit transition, with adjustments made to accommodate differences as outlined. * This outlook is based on the hardware definition of 64-bits and does not take into account the software environment. * This outlook covers systems shipped in all IDC size classifications: PC, workstation, small-scale, medium-scale, and large-scale. Thus, the unit outlook will be heavily weighted towards the impact in the PC market since PCs account for the vast majority of system units. Supply side factors (i.e., vendor shipment): * Expected introduction of 64-bit technologies into the market by major vendors (refer back to Table 2 for details). * As this is a complete industry outlook, it is heavily influenced by the volume leader, Intel. IDC expects that Intel will rollout its P7 processor (the first 64-bit x86 compatible), in the 1995 time frame. If Intel's schedule is not as we have assumed, this outlook would need to be revised accordingly. * In the PC market, the transition will accelerate at the point in which 32- and 64-bit microprocessors converge in price. IDC expects this to happen a few years after the P7 is available _ when its lower- cost siblings hit the street. This will be very late in the decade. * In the midrange market, microprocessor cost is a much smaller component of the system pricing, thus the difference in prices between 32 and 64-bit microprocessors will have little impact. Instead, the movement to 64-bit in the midrange will be dictated by competitive needs and R&D resource allocation among suppliers. Demand Side Factors (i.e., user adoption) * The 64-bit migration trend will be led by high-performance workstations and mid-range systems in the technical environment, where applications requiring high performance will make 64-bit solutions more attractive. * PC users will transition to 64-bit systems much more slowly. This is a function of the lack of applications requiring 64-bit solutions, as well as the fact that 32-bit systems are only now beginning to hit their stride. * When 64-bit software environments become widely available (in approximately the 1996 time frame), IDC expects that a significant hurdle will be overcome. Growth in 64-bit applications will accelerate, which will then accelerate the growth of the overall 64-bit market. To summarize, we believe the transition from 32- to 64-bit architectures will happen at different speeds within different market segments (see Table 3). The net effect on the overall market outlook for 64-bit systems, shown in Figure 4, is a slow transition; this is due mainly to the overwhelming influence the Intel PC market has on shipments in the overall market. IDC believes that 64-bit architecture represents an inevitable, evolutionary progression in processor technology; this will be an important, but not dominant technology in the latter half of the 1990s. We don't usually forecast out beyond a five year time frame, but our best estimate is that by 1997 approximately 10% of all computer systems shipped worldwide will have a 64-bit processor (not necessarily running a 64-bit Operating System). Sixty-four-bit architectures should comprise almost 20% of all unit shipments by the turn of the century. Filing Information: Date: November 1992 IDC #: 7175 Volume: 1.Unix Tab: 3.Operating Systems/ Architectural Issues Sixty-Four Bit Computing: A Bit Ahead of Its Time? Analysts: David M. Smith and John Morrell The Bottom Line IDC believes that 64-bit architecture is an inevitable technology evolution. High-performance technical users will be the first to take advantage of the features but the overall market will be slow to absorb the new technology. In general, there is not the pent-up demand for the features a 64-bit architecture provides as there was in the transition to 32-bits, most notably in address space requirements. We believe ISVs will tend to support the architectures from which they derive revenue, regardless of the 64- or 32-bit question. The majority of ISVs will port existing 32-bit software to the 64-bit platforms without taking advantage of the 64-bit features. Copyright 1992 International Data Corporation. Reprinted and distributed electronically by Hewlett-Packard Company with permission of International Data Corporation. For additional copies please contact Linda Rich, 508-935-4389. Source: International Data Corporation, 1992 Figure 1 Approximate Demand for Addressing and Availability of Hardware Source:International Data Corporation, 1992 Figure 2 Mapping Schemes Source: International Data Corporation, 1992 Figure 3 Architectural Advantages of 64-bit Computing Source: International Data Corporation, 1992 *OSF/1 only initially (see text for details) **Likely to be HaL, not Sun Source: International Data Corporation, 1992 Source: International Data Corporation, 1992 Source: International Data Corporation, 1992 Table 1: Address Space Size Bits Address Space Size # of times larger than previous size 8-bit 256 bytes _ 16-bit 64,000 bytes 256 32-bit 4,000,000,000 bytes 64,000 64-bit 16,000,000,000,000,000,000 bytes 4,000,000,000 Table 2 Anticipated rollout of 64-Bit Products from Major Suppliers Architecture Sponsor Introduction First System 64-Bit OS First 64-Bit (64-Bit) Shipment Applications R4000 SGI/MIPS 1992 1992 1995 1995 Alpha Digital 1992 1993 1993* 1994 Sparc Sun, others 1993 1993** 1995 1996 Power IBM 1994 1994 1995 1996 PA-RISC HP 1994 1995 1996 1996 P7 Intel 1995 1996 1997 1997 Table 3 Transition Times to 64-bit Architectures Within Different Market Segments Commercial Technical Overall PC Very, very slow Very slow Very slow Workstation Slow, cost issues Fast Fastest Midrange Medium, 64-bit SW lag Fast Medium Large scale Medium Very fast Fast