from FaxBACK(TM) document #7046 THE iCOMP(TM) INDEX: A NEW WAY TO COMPARE PROCESSOR PERFORMANCE An Intel Corporation Backgrounder The personal computer and its shrink-wrapped software applications have brought convenience, greater productivity and many other benefits to the working world, but along with all of this has come a certain amount of confusion. Many tasks, from writing to managing business accounting to creating complex visual presentations, are easier than ever before. But the price of this power is that it takes more time and effort to understand the technology and choices offered by the marketplace. Deciding which microprocessor is best is only one of a number of choices, ranging from hard disk capacity to display devices, facing the PC buyer, but it's an important and little-understood one. The power of the central processing unit (CPU) places an upper boundary on how fast the system can run its software, and therefore on how much work it can do for the user. Seven years ago, there was one 32-bit member of the Intel architecture, the Intel386(TM) DX microprocessor. Today, there are more than 30 Intel386 and Intel486(TM) microprocessors, and the first of a new generation of CPUs, the PentiumTM processor, has just been announced. This proliferation of processors has been good for users, who now have more choices of both CPUs and systems than at any time during the history of the Intel architecture, but it also poses a problem: how can the average non-technical PC buyer, uninterested in a high level of technical detail, quickly evaluate the differences in performance among the many CPUs? PC buyers typically are barraged by several indicators--megahertz (MHz), or the operating frequency of the processor; any of a number of technical benchmarks such as Whetstones, Dhrystones, Linpacks and BAPCo SYSmarks; and the standard but misleading measure of millions of instructions per second (MIPS) rating. Each performance indicator provides worthwhile information, but each has disadvantages for the non-technical user. Research Results To better understand the level of knowledge about the CPU in the marketplace, Intel conducted research among end users, MIS managers, system manufacturers, channel salespeople and others. Result: there is a widespread, and incorrect perception among non-technical PC purchasers that a CPU's megahertz rating is a linear measure of its performance. The proliferation of the Intel architecture, with its DX, SX and other designations, has further muddied the waters, as has the appearance of new technology in later members of the CPU family, including floating-point units and clock-doubling technology. Although the megahertz rating has important consequences for the design of a PC system, and is important to the PC design engineer, it is not a good measure of CPU performance, especially when comparing one family of CPUs to another. Yet in a recent Intel-sponsored study, some two-thirds of the PC buyers polled thought there was a direct correlation between a CPU's megahertz rating and its performance. The solution that Intel has undertaken to address this problem is to develop a single number that does reliably measure relative CPU performance. It is called the iCOMP index, for Intel COmparative Microprocessor Performance. The iCOMP formula was designed to reflect the full capabilities of a CPU--those which are executed not only by today's desktop systems and software applications, but the software expected to be in use throughout the three to five year average life of a system bought today. It is an index that reflects the relative performance of one Intel microprocessor to another, not system performance. Creating another performance measure may not seem like an ideal solution. There are already a lot of numbers floating around, all vying for the attention of the PC buyer. However, new research confirms what intuition suggests is true, that users at different levels of technical sophistication need to know different things. For example, large corporate MIS departments spend weeks testing a handful of systems using home-grown performance suites. These organizations usually have their own testing lab for private performance testing. MIS shops with a smaller budget might use the publicly available technical benchmark reports and packages, such as SPEC, PC Labs, BAPCo's SYSmark92 and others to thoroughly understand every aspect of CPU and system performance. Meanwhile, an individual technical user may spend a dozen hours understanding the technology of the system he or she is planning to purchase. In stark contrast, the non-technical user may spend just a few minutes becoming familiar with a PC's microprocessor and its performance. For this user, Intel has ensured that the iCOMP index is not complicated, but a simple, single rating for each microprocessor, measuring its overall performance. The iCOMP index is rich in information because it embodies in one number a weighted average of eight aspects of CPU performance, presently using four industry-standard benchmarks. Thus, the non-technical user who doesn't have the interest or time to read many separate benchmark reports still derives benefit from them. This rating is not intended to replace megahertz, or any benchmark. It is additional information that the PC buyer can use to better understand the differences among Intel CPUs. Buyers should still consult other sources of information, in order to evaluate more completely the systems they are considering purchasing. They still need to make the other choices associated with the PC--peripherals, expansion slots, total cost, service and support, to determine what's needed to run their chosen software with the performance they need and expect. Intel hopes that, armed with iCOMP ratings of the many Intel microprocessor choices, buyers will find it easier to understand important differences among Intel's broad selection of CPUs. The iCOMP Index The iCOMP index is a simple numerical index of relative performance for making straightforward comparisons of Intel CPU power. It provides consumers with useful information when they make a PC purchase. It is intended to supplement, not replace benchmarks. Because of the simplicity of the iCOMP index, it cannot capture every nuance of CPU performance. It is a tool for making comparisons between different Intel processors. Two systems with an Intel486™ DX2-50 CPU, for example, will have CPUs with exactly the same iCOMP rating. All systems with an Intel486 DX2-50 CPU will not perform the same--differences in system design and configuration will affect performance considerably. For instance, system vendors sell systems with a wide variety of disk capacities and speeds, system memory, system bus features and video and graphics capabilities, and all of these influence how the CPU and the system perform under actual conditions of use. However, given systems of comparable configuration and design, the one using the CPU with the higher iCOMP index will have more power and will run software faster. The iCOMP rating is an index of CPU performance for any Intel architecture processor regardless of the family. For example, iCOMP indices for Intel386 CPUs are less than those of Intel486 CPUs. Future generations of the Intel architecture, such as the Pentium(TM)processor, will have still higher iCOMP indices than those of the Intel486 CPUs. In addition, the iCOMP index is forward looking. Rather than focusing solely on the performance of existing 16-bit applications, the iCOMP index includes a 32-bit performance component to better capture the value of CPU performance over the typical three- to five-year lifetime of the processor. With the start of each new microprocessor design, Intel solicits extensive customer and industry comments concerning the capabilities that should be present in the next generation. The iCOMP formula also reflects that feedback. This, of course, requires making some predictions about the evolution of PC users' application mixes, as detailed below. The derivation of the iCOMP formula is public information (and can be seen at the end of this document and in Appendix A of the iCOMP report). The weighting of the iCOMP formula's components represents Intel's best guess as to how user software mixes will evolve over the next three to five years. Today, for example, most of the demands on the CPU are directed at its integer unit. However, other types of applications, especially in graphics and video, are gaining in popularity, and they will increasing call upon floating-point and specialized CPU capabilities. The current iCOMP formula takes most of these factors into account. Intel has conducted market research, including discussions with software vendors, OEM customers, MIS managers, the press and end users, but our ability to precisely predict the future is imperfect. The iCOMP formula represents Intel's best prediction of how the silicon technology will be used. As software and its demands on the CPU evolve, Intel will make adjustments to the formula as they are needed. As an example, over time, the 16-bit software performance component will shrink while the 32-bit software component will grow in importance. The base processor for the iCOMP index is the 25-MHz Intel486 SX CPU, which has been assigned a value of 100. All other rated CPUs will have a number that is either above 100, meaning a faster CPU, or below, meaning slower. The size of the disparity between any two indices provides a relative measure of how much more powerful one CPU is than any other. However, to using the analogy of the EPA's highway and city mileage rating, the actual performance the user sees in the system depends on the individual car (system) and driving habits (specific types of software) of the user, as well as on the CPU. The iCOMP index can help buyers select the right CPU for their systems, but they'll still base their system selection on the usual factors: disk capacity, video capability and other system features, price, the advice of magazine articles and friends, and system benchmarks. _______________________________________________________________________________ The Intel iCOMP(TM) Rating Index The Intel486 SL in a mobile configuration performs the same as an Intel486 DX in a mobile configuration. _______________________________________________________________________________ Source: iCOMP(TM): A Simplified Measure of Relative Intel Microprocessor Performance. If you'd like to receive a copy of this 30-page technical report, call 1-800-395-7009 and ask for document mumber 241461-001. Please allow 1-2 weeks for delivery.) _______________________________________________________________________________ The Intel iCOMP(TM) Rating Index for i486(TM) CPUs with Overdrive(TM) Processors _______________________________________________________________________________ Source: iCOMP(TM): A Simplified Measure of Relative Intel Microprocessor Performance. If you'd like to receive a copy of this 30-page technical report, call 1-800-395-7009 and ask for document mumber 241461-001. Please allow 1-2 weeks for delivery.) _______________________________________________________________________________ iCOMP Index Formula Components Performance Benchmark Weight 16-Bit Integer PC Labs 7.0 52 % 16-Bit Floating Point Whetstone 2 % PC Labs 7.0 1 % 16-Bit Graphics PC Labs 7.0 * 10 % 16-Bit Video PC Labs 7.0 * 5 % 32-Bit Integer SPECint92 15 % 32-Bit Floating Point SPECfp92 5 % 32-Bit Graphics SPECint92* 5 % 32-Bit Video SPECint92 * 5 % * Placeholder until a better benchmark for that performance type is developed. Only the CPU portion of PC Labs 7.0 is used _______________________________________________________________________________ Source: iCOMP(TM): A Simplified Measure of Relative Intel Microprocessor Performance. If you'd like to receive a copy of this 30-page technical report, call 1-800-395-7009 and ask for document mumber 241461-001. Please allow 1-2 weeks for delivery.) Intel386, Intel486, Pentium, OverDrive and iCOMP are trademarks of Intel Corporation.