Processor Design: There Is a Choice

CISC AND RISC:

MORE IS NOT ALWAYS BETTER Most processors in mainframe computers and personal computers have a CISC (complex instruction set computer) design. A CISC computer's machine language offers programmers a wide variety of instructions from which to choose (add, multiply, compare, move data, and so on). CISC computers reflect the evolution of increasingly sophisticated machine languages. Computer designers, however, are rediscovering the beauty of simplicity. Computers designed around much smaller instruction sets can realize significantly increased throughput for certain applications, especially those that involve graphics (for example, computer-aided design). These computers have RISC (reduced instruction set computer) design. The RISC processor shifts much of the computational burden from the hardware to the software. Proponents of RISC design feel that the limitations of a reduced instruction set are easily offset by increased processing speed and the lower cost of RISC microprocessors.

PARALLEL PROCESSING: COMPUTERS WORKINGTOGETHER

In a single processor environment, the processor addresses the programming problem sequentially, from beginning to end. Today, designers are building computers that break a programming problem into pieces. Work on each of these pieces is then executed simultaneously in separate processors, all of which are part of the same computer. The concept of using multiple processors in the same computer is known as parallel processing. In parallel processing, one main processor examines the programming problem and determines what portions, if any, of the problem can be solved in pieces (see Figure 2-7). Those pieces that can be addressed separately are routed to other processors and solved. The individual pieces are then reassembled in the main processor for further computation, output, or storage. The net result of parallel processing is better throughput.
Computer designers are creating mainframes and supercomputers with thousands of integrated microprocessors. Parallel processing on such a large scale is referred to as massively parallel processing (MPP). These super-fast supercomputers have sufficient computing capacity to attack applications that have been beyond that of computers with traditional computer designs. For example, researchers can now simulate global warming with these computers.

Describing the Processor: Distinguishing Characteristics

How do we distinguish one computer from the other? Much the same way we'd distinguish one person from the other. When describing someone we generally note gender, height, weight, and age. When describing computers or processors we talk about word size, speed, and the capacity of their associated RAM (see Figure 2-8). For example, a computer might be described as a 64-bit, 450-MHz, 256-MB PC. Let's see what this means.

Word Size: 16-, 32-, and 64-Lane Bitways

Just as the brain sends and receives signals through the central nervous system, the *"" **"»"*il 4M.MAAH3 processor sends and receives electrical signals through its common electrical bus aword at a time. A word describes the number of bits that are handled as a unit within a particular computer system's bus or during internal processing.
Twenty years ago, a computer's word size applied both to transmissions through the electrical bus and to all internal processing. This is no longer the case. In some of today's computers, one word size defines pathways in the bus and another word size defines internal processing capacity. Internal processing involves the movement of data and commands between registers, the control unit, and the arithmetic and logic unit (see Figure 2-5). Many popular computers have 64-bit internal processing but only a 32-bit path through the bus. For certain input/output-oriented applications, a 64-bit computer with a 32-bit bus may not realize the throughput of a full 64-bit computer.
The word size of modern microcomputers is normally 64 bits (eight 8-bit bytes). Early PCs had word sizes of 8 bits (one byte) and 16 bits (two bytes). Workstations, mainframes, and supercomputers have 64-bit word sizes and up.

Processor Speed: Warp Speed

A tractor can go 12 miles per hour (mph), a minivan can go 90 mph, and a slingshot drag racer can go 240 mph. These speeds, however, provide litde insight into the relative capabilities of these vehicles. What good is a 240-mph tractor or a 12-mph mini-van? Similarly, you have to place the speed of computers within the context of their design and application. Generally, PCs are measured in MHz, workstations and mainframes are measured in MIPS, and supercomputers are measured in FLOPS.

MEGAHERTZ: MHz

The PC's heart is its crystal oscillator and its heartbeat is the dock cycle. The crystal oscillator paces the execution of instructions within the processor. A micro's processor speed is rated by its frequency of oscillation, or the number of clock cycles per second. Most modern personal computers are rated between 300 and 600 megahertz, or MHz (millions of clock cycles). The elapsed time for one clock cycle is 1/frequency (1 divided by the frequency). For example, the time it takes to complete one cycle on a 400-MHz processor is 1/400,000,000, or 0.000000025 seconds, or
Many field sales representatives carry notebook PCs when they call on customers. This Merck Human Health-U.S. representative uses computer-based detailing to make a presentation on cholesterol reducer to a cardiologist. Computer data help her better target information to physicians. Her notebook has a 64-bit processor with a speed of 300 MHz and a RAM capacityof 1 28 MB.
The CRAY T90 supercomputer is the most powerful general-purpose computer ever built. General purpose computers are capable of handling a wide range of applications. The system is capable of crunching data at a peak rate of 64 GFLOPS (gigaflops). One GFLOP equals one billion floating point logic operations per second. Courtesy of E-Systems, Inc.
2.5 nanoseconds (2.5 billionths of a second). Normally several clock cycles are required to fetch, decode, and execute a single program instruction. The shorter the clock cycle, the faster the processor.
To properly evaluate the processing capability of a computer, you must consider both the processor speed and the word size. A 64-bit computer with a 450-MHz processor has more processing capability than does a 32-bit computer with a 450-MHz processor.

MIPS

Processing speed may also be measured in MIPS, or millions of z'nstructions per second. Although frequently associated with workstations and mainframes, MIPS is also applied to high-end PCs. Computers operate in the 20- to 1000-MIPS range. A 100-MIPS computer can execute 100 million instructions per second.

FLOPS

Supercomputer speed is measured in FLOPS—/loating point operations per second. Supercomputer applications, which are often scientific, frequently involve floating point operations. Floating point operations accommodate very small or very large numbers. State-of-the-art supercomputers operate in the 500-GFLOPS to 3-TFLOPS range. (A GFLOPS, gigaflops, is one billion FLOPS; a TFLOPS, or teraflops, is one trillion FLOPS.)

RAM Capacity: Megachips

The capacity of RAM is stated in terms of the number of bytes it can store. Memory capacity for most computers is stated in terms of megabytes (MB). One megabyte equals 1,048,576 (220) bytes. Memory capacities of modern PCs range from 32 MB to 512 MB. Memory capacities of early PCs were measured in kilobytes (KB). One kilobyte is 1024 (210) bytes of storage.
Some high-end mainframes and supercomputers have more than 8000 MB of RAM. Their RAM capacities are stated as gigabytes (GB)—about one billion bytes. It's only a matter of time before we state RAM in terms of terabytes (TB)—about one trillion bytes. GB and TB are frequently used in reference to high-capacity secondary storage. Occasionally you will see memory capacities of individual chips stated in terms of kilobits (Kb) and megabits (Mb). Figure 2-9 should give you a feel for KBs, MBs, GBs, and TBs.