PC manufacturers seem to have a mantra. We hear it's "Faster, better, cheaper." It particularly applies to the makers of the brain of a computer, which is known as the central processing unit (CPU). The two main processor manufacturers are Intel and Advanced Micro Devices (AMD), and their rivalry is spurring announcements almost every month. Processors are speeding up, slimming down, gaining memory, and getting cheaper all the time. For this reason, more processors are in the "What's Hot" than in the "What's Not" category. PCs are getting faster, better, and cheaper because PC processors are doing likewise. When you hear about computers in your walls, clothes, and body, people are talking about processors small and powerful enough to do things that were previously impossible. With that in mind, here are some trends in PC processing.
What's Hot. Speed. AMD's 800 MHz Athlon | In the processor world, everyone is competing to be the fastest. AMD announced an 800MHz Athlon processor in January in response to Intel's 800 MHz Pentium III processor, which was announced on December 20, 1999. Both have said they intend to reach one gigahertz (GHz) by the end of 2000. Some industry analysts think that AMD may even get their 1 GHz processor out by the summer. In 1999 a company called Kryotech managed to run an Athlon at 1 GHz. It accomplished this by supercooling the processor with an advanced compressor to reduce excess heat. But the Holy Grail, says Kevin Krewell, an analyst at MicroDesign Resources (MDR), is to attain one GHz with an air-cooled processor. Clock speeds. Today, even home users want superior performance and clock speeds for 3D gaming, entertainment software, and the Internet. "Frequency is the tide that lifts all boats," according to one AMD presentation. As processors rev up the other components must improve to keep up with them. While many applications don't need sheer speed, the rising-tide effect makes performance something to watch. Also developing rapidly is the processor's internal architecture. Processors are using sophisticated techniques to do operations in parallel. Long gone is the era when computers executed one command at a time in sequence on a series of punch cards. To gain a better understanding of the technology behind CPUs, imagine the processor reading several instructions into memory. It scans for tasks such as calculations and hands off the ones it finds to be processed separately. If it encounters a statement with two possible outcomes, it predicts the likeliest outcome and tackles that first. Taking on several commands at once is called superscaling. Breaking them into several parts is called superpipelining. As an example of a superpipelined microarchitecture, AMD's Athlon has nine executable pipelines: three for address calculations, three for integer calculations, and three for floating point and other calculations. AMD's K6-III processor | Related to parallel processing is the use of Single Instruction, Multiple Data (SIMD) techniques. SIMD is an evolution of MMX technology, which Intel introduced a couple of years ago. By improving the PC's computational capabilities, manufacturers deliver better audio and video in multimedia applications and on the Internet. Both Intel and AMD are adding SIMD instructions and enhancing MMX. Intel bundles this technology under the name Streaming SIMD Extensions (SSE) and AMD calls it 3DNow!, but the idea is similar. Both companies aim to add as much functionality as possible in the fewest number of instructions since each instruction takes time to execute. Bottlenecks. In PC processing one of every three operations involves memory, so access to system's random access memory (RAM) is a huge bottleneck. Manufacturers get around this by placing memory caches near the processor. When the processor fetches a block of code, it puts it in a nearby cache. If the processor finds the instructions it needs in the cache next time, it proceeds. If not, it queries the RAM again. Typically a processor has a first-level cache (L1) built onto the chip and a second-level cache (L2) nearby connected with a bus. Intel and AMD use different cache configurations on their processors, but both are improving the caches constantly. The trend is to increase the cache size, put more cache on the die with the processor, and run the cache at the processor's clock speed. AMD's TriLevel Cache design even offers an optional Level 3 (L3) cache. The processor's fetches to RAM are slow because the system bus (the channel between the CPU and RAM) is slow. A typical system bus can convey 64 bits (eight bytes) at a time at 100 MHz, or a peak of 800 MB per second. Buses haven't edged much beyond 133 MHz, though the Athlon has a 200-MHz front-side bus (FSB) and AMD has proposed a 266-MHz FSB. System buses have several channels for data transfer, so a processor can send another request before it gets the results of the last one. New buses will allow split and out-of-order transactions. In the latter case, the processor will make requests in any order and match the responses when they arrive. It won't have to do jobs sequentially anymore. Manufacturing. There are some of the improvements to note, but none of them come free. The "cascade of tricks," as one writer put it, will require more transistors, longer signal paths, and larger chips. That means a more elaborate production process and a greater chance of defects. In other words, more expense. That's why progress in manufacturing methods is important. The industry is moving from 0.25-micron to 0.18-micron technology and from aluminum to copper circuitry. The 0.18-micron standard refers to the width of the wires between transistors; narrower pathways let more circuits fit on a chip. Copper's advantage is that it's a better conductor of electricity. More data can flow through it in a given amount of time and space. Together these developments will make chips faster, cooler, more energy-efficient, and cheaper to manufacture. Intel is already putting 28 million transistors on its Pentium IIIs with 0.18-micron technology. AMD is using similar methods to produce its Athlons and plans to use them for its K6-2+ line this year. There is one caveat worth noting. Processor and system vendors have persuaded consumers that speed is the only thing that matters. This is not the case. Speed does matter, but two machines with the same megahertz can produce widely varying results. Factors such as parallel processing, on-chip caches, and graphics enhancements cause these variations. Most people can't test these features individually, but they should be made aware of them. Don't just buy the model with the biggest numbers. Do your homework; read reviews in Smart Computing and other magazines and see what the experts say about performance. Mobile computers show similar tendencies; this is especially the case where miniaturization is concerned. One difference is the need for less power consumption. In upcoming notebook technology, such as Intel's Geyserville and AMD's Gemini, the system will be able to slow the processor and reduce its voltage when it's running from the battery. That will conserve the unit's power and prolong its operating life. However, the system will run at desktop system speeds when plugged into a power outlet.
What's Not. Although processors provide more bang for the buck than ever, the market for inexpensive versions is curiously tame. The days when most people needed only a basic computer for word processing and spreadsheets are past. Even casual users are demanding power for advanced gaming, multimedia, and Internet functions. Less powerful processors. Perhaps that is why the demand for low-end processing has ebbed. Clock speeds less than 400 MHz and 500 MHz are becoming passé. Why buy a processor barely able to cope with applications when a more potent one is affordable? Manufacturers such as Cyrix and IDT's Centaur group have suffered because of this trend. Furthermore, Intel has released upgrades of its low-end Celeron processor and has aggressively cut prices. All this has made it hard for cost-conscious companies to compete. Via Technologies has bought Cyrix and Centaur and hopes to revitalize them using their best assets. It faces several obstacles; this includes feuds over patent rights and fierce competition from Intel and AMD. Until it proves itself, Via's prospects in the CPU market remain uncertain. Intel also had some difficulties when rolling out its Coppermine Pentiums based on 0.18-micron technology. The glitches affected only a small percent of these processors, but they forced Intel to delay its Coppermine release announcement from summer to fall. MMX. Another technology that's no longer hot is Intel's MMX instruction set, which enhances integer calculations and data movement. MMX was a buzzword when it appeared. It was a technical term that Intel marketed heavily and everyone thought they had to have. In the end, however, it didn't have much of an impact. Advanced Graphics Port (AGP). Like MMX, the Accelerated Graphics Port (AGP) was another hot item that fizzled. AGP is a high-bandwidth fast lane for graphics data, but few programs take advantage of its benefits. Systems are appearing with AGP4X (four times the speed of the original AGP), but some fast games don't even utilize AGP2X. Intel's Pentium III processor | Integrated 3D. A related issue is 3D graphics. Intel has integrated 3D graphics into its mainstream chip sets, the 810 and 810e. The integrated approach is itself dubious. With their higher bandwidth, graphics accelerators can render 3D graphics faster than integrated circuits can. But computer buyers care "only about megahertz, not actual performance," says Linley Gwennap of MicroDesign Resources. Personal and business users don't run many 3D applications and don't realize what they're missing. CPU ID. One area that's definitely not sizzling is the Pentium's processor serial number. Intel claims this ID number is there to help information technology (IT) departments manage their systems. The number was supposed to aid e-commerce, protect digital content, and prevent counterfeiting and theft. However, when Intel announced it last year, industry pundits noted that remote operators could use it to track systems across the Internet regardless of whether they were stolen. Worse, malicious users could steal an ID number and fleece online merchants with it. Though corporations may appreciate the processor serial number, it offers no clear benefits to most consumers, small business, and home offices.
What's Next. In the near future AMD's Athlon processor will be "literally and figuratively hot," says MDR's Kevin Krewell. It will hold the high ground until Intel premieres its Willamette processor, the next generation of 32-bit processor in the x86 family, in 2000. As the successor to Intel's Pentium III Xeon, this processor may possibly carry the Pentium IV designation. Intel's Pentium III Xeon processor | AMD will respond with its K75 processors, which are codenamed Thunderbird and Spitfire and made using 0.18-micron technology. 2001 is expected to usher in AMD's Sledgehammer. This is an eighth-generation, 64-bit processor based on the x86 architecture. Meanwhile, Intel will perfect its 64-bit processor using a novel architecture known as IA-64 or Itanium. Understandably, companies won't reveal their future plans because of the fierce competition. However, some trends should continue. Among these are speed, miniaturization, convergence, and commoditization. MDR's Linley Gwennap claims each generation of processor roughly doubles the performance of the previous one. Whether the Athlon or Itanium is the first seventh-generation processor, people can expect more iterations during the next few years. If the pattern persists, processors may race at 8 GHz or faster by 2010. As Intel and AMD move to 0.18-micron processes, they're beginning to talk about whether 0.13- or 0.10-micron wiring will follow. Etching these paths will require an ultraviolet (UV) light with an extremely narrow wavelength. Further reductions may call for electron beams rather than UV light. These developments will make computers smaller and faster than ever. Simple processors will show up in wristwatches, eyeglasses, and hearing aids. More robust ones will begin the evolution of today's limited handheld devices into Star Trek's all-purpose tricorder. Another trend to note is convergence. Note that the computer is already the tool of choice for editing audio or video materials. Soon all content will be digital and physical media will fade away. Multimedia signals will arrive on a TV-like computer and be saved for playback. PCs will merge with equipment such as televisions, Web-based terminals, video-game consoles, digital audio and video recorders, and telephone answering machines. Tied to this is the commoditization of processors. Though the Big Two control the PC market, manufacturers are embedding intelligence in everything from TVs to telephones to game stations. Because of the ubiquity of these platforms compared to PCs, embedded processors should predominate. Peter N. Glaskowsky of MicroDesign Resources believes television "will take on an even greater role in the next decade," rendering cheap computers obsolete. "PCs will continue to sell to business buyers and home PC enthusiasts," says Glaskowsky, but they'll be a small part of the customer base. Most processors will come from vendors like Sony or Nintendo. A world without Intel inside? Stay tuned for further developments. by Robert V. Schmidt | What's HOT | What's NOT | What's NEXT | •Faster clock speeds •SIMD (SSE, 3DNow!) •More cache | •Low-end processors with low performance •Integrated 3-D graphics •Intel's processor serial number | •64-bit processing and architecture wars •Smaller and faster processors •Convergence and commoditization |
|
Terms To Know cache— Small, fast areas of memory near the processor where frequently-used instructions are kept. clock speed—The number of cycles a processor performs in the span of one cycle of the computer's system clock. Measured in megahertz (MHz), or millions of cycles per second. pipelining—A processor's ability to read a new instruction from memory before it finishes processing the current instruction. processor—A self-contained chip or unit that either does or controls a system's data processing. Also known as the microprocessor or central processing unit (CPU). Single Instruction, Multiple Data (SIMD)—Processing technique that applies one command to several data items within a clock cycle. Superscaling—Processor's ability to execute several sets of instructions at the same time. system bus—Set of conducting wires or conduits that transport data from the processor's core to other components. |
|