PC Hardware in a Nutshell, 2nd Edition

3.1 Motherboard Characteristics

Several characteristics differentiate motherboards, including physical characteristics, which in combination are called the form factor; the chipset used, which defines the capabilities of the motherboard; the processors the motherboard supports; the BIOS it uses; and the internal and expansion buses that it supports. The following sections examine each of these factors.

3.1.1 Form Factor

Motherboards differ in size, shape, position of mounting holes, power supply connector type, and port types and locations. Together, these differences define the form factor of the motherboard. Form factor is a critical issue when you upgrade a system, because the replacement motherboard must physically fit the case and use the existing power supply connectors. Form factor doesn't matter when you're building a new PC. You simply select the best motherboard for your needs, and then buy a case that fits it. Many motherboard manufacturers build similar motherboards in different form factors. Here are the form factors you may encounter:

AT, Baby AT (BAT), and LPX

All of these form factors are based on the motherboard used in the original 1984 IBM PC AT and are now obsolescent or, in the case of LPX, obsolete. Although it is still possible to buy AT and BAT motherboards, selection is limited (e.g., Athlon and Pentium 4 motherboards are not available in these form factors) and most AT and BAT motherboards do not include the latest chipsets and other technologies. The only reason to buy an AT or BAT motherboard is to upgrade an existing system without replacing the power supply and case. Good cases with power supply are available for $65 or so, and the power supply in an older case is likely to be near the end of its life, so we do not recommend buying a motherboard in any of these form factors.

ATX and variants

Nearly all current motherboards use the ATX form factor, or one of its smaller variants, the miniATX, ATX, FlexATX, and NLX. For a complete discussion of these form factors, including case and power supply issues, see Chapter 25 and Chapter 26.

Proprietary

Although it is much less common nowadays to find new systems that use proprietary motherboards, such boards were relatively common a few years ago, particularly in systems sold by major manufacturers like Compaq and IBM. A cynical observer might believe that the major manufacturers did this to lock customers in. In fact, it probably had more to do with the fact that the then-current Baby AT standard had reached the end of its useful life and the ATX standard had not yet become the obvious successor. To reduce manufacturing costs and increase reliability, manufacturers that had the in-house engineering talent to do so designed their own motherboards. Some of these are actually very elegantly designed. They all share one fatal flaw, however: they aren't standard. Any system that uses a proprietary motherboard form factor is effectively not upgradable.

3.1.2 Chipsets

Just as the motherboard defines a PC, the chipset defines a motherboard. The chipset determines the main characteristics of the motherboard—what processors it supports, what RAM types it can use, what bus types and speeds it supports, whether it supports standards like AGP and USB, and so on. Chipsets are so named because they usually comprise two relatively large chips. Some chipsets contain three or more chips. A few chipsets, most of which are intended for low-cost systems, have all functions on one physical chip.

Figure 3-1 shows a block diagram for the 815E, Intel's flagship chipset for the Pentium III and Celeron processors. Block diagrams are helpful in understanding the functions of a chipset. Like most chipsets, the 815E comprises a Northbridge chip, labeled GMCH, and a Southbridge chip, labeled ICH2. The Northbridge interfaces the CPU, memory, and AGP video—all high-bandwidth components. The Southbridge can conveniently be thought of as a peripheral controller. The Southbridge manages low- and moderate-bandwidth components such as the PCI bus, IDE interface, USB ports, and so on. Some motherboards supplement Southbridge functions by adding another chip, often called a Super I/O controller.

Figure 3-1. Block diagram of the Intel 815E chipset (graphic courtesy of Intel Corporation)

Chipset makers often use one Southbridge with different Northbridges to create chipsets for different processors. For example, Intel uses the 82801BA ICH2 Southbridge in numerous chipsets, some of which support the Pentium III and Celeron, and others the Pentium 4. It's also possible to use the same Southbridge to create chipsets for processors from different makers. For example, VIA Technologies produces chipsets for the Intel Pentium 4 and AMD Athlon—two processors with very different architectures—by combining the same Southbridge with a Northbridge that supports the processor in question.

There are scores of motherboard manufacturers, but only a handful of chipset manufacturers. This is because designing a chipset requires significant engineering resources, but building a motherboard around that chipset is straightforward. Chipset manufacturers want motherboard manufacturers to buy their chipsets, so they provide detailed specifications and engineering drawings, which makes building motherboards more a matter of production than design.

If you examine twenty motherboards made by twenty different manufacturers, all using the same chipset, you'll find that the similarities outweigh the differences. There may be minor variations in features and layout, but the chief differences are the quality of the components used and the quality of construction. Accordingly, it is quite possible for one manufacturer to make a terrible motherboard and another manufacturer a superb motherboard, even though both use the same chipset. The motherboard market is so competitive that price is almost invariably an excellent predictor of the quality of a motherboard.

Understanding chipset basics is important whether you are upgrading an existing PC or building a new one. If you are upgrading, understanding the chipset tells you what can and cannot be done within its limitations. It also helps you avoid useless or counterproductive upgrades. For example, the Intel 430TX Pentium chipset supports 256 MB RAM but caches only 64 MB, so adding more than 64 MB to a 430TX system can actually reduce performance. If you are building or buying a new PC, its chipset is the most important factor in determining motherboard performance and future upgradability. Understanding the differences between competing chipsets lets you make rational decisions about which computer or motherboard to buy. Because chipsets are so important, we've devoted a great deal of space to explain what you need to know about chipsets.

3.1.2.1 CPU Family Support

Chipsets support only one of these CPU families:

Intel 486 and similar fourth-generation CPUs

Intel Pentium and compatible fifth-generation CPUs

Intel Pentium Pro/Pentium II/Celeron/Pentium III sixth-generation CPUs

Intel Pentium 4 seventh-generation CPUs

AMD Athlon/Duron sixth-generation CPUs

This is because processor families differ greatly in how they access and manage main memory, cache, and other major system components. For example, while fifth-generation CPUs use separate L2 cache memory on the motherboard, sixth-generation CPUs have L2 cache within the processor package or integrated on-die.

A particular chipset may support most or all CPUs within a family, or may support only one of them. For example, some chipsets support the Pentium II, Celeron, and Pentium III CPUs, while others support only one or two of those. Most chipsets designed for the Pentium also support compatible processors like the AMD K5 and K6 and the Cyrix 6x86 and 6x86MX, which were designed as pin-compatible, although using one of these CPUs may require different motherboard jumper settings and chipset CMOS parameters than those used for a genuine Pentium. Also, chipsets designed for late variants within a generation may not support earlier variants within that generation. For example, Intel 815-family chipsets do not support the Pentium II and early variants of the Celeron.

There is also the issue of hybrid processors, which straddle generations. For example, recent Pentium-compatible CPUs like the AMD K6-2/III and the Cyrix 6x86MX have sixth-generation features that are not supported by older chipsets, particularly Intel chipsets. These processors are sixth-generation internally, or largely so, but physically fit fifth-generation motherboards. To encourage migration to the Pentium II, Intel stopped developing fifth-generation chipsets. Third-party chipset manufacturers, notably VIA, SiS, and ALi, took up the slack by introducing enhanced fifth-generation chipsets called Super Socket 7 chipsets. Using these chipsets allowed manufacturers to build motherboards that supported features not supported by Intel fifth-generation chipsets, such as AGP.

3.1.2.2 CPU Speed Support

A chipset that supports a particular CPU may support all or only some of the speeds that that CPU is available in. In general, faster CPUs require faster chipsets, so you might imagine that a chipset rated to handle the fastest version of a particular CPU could also handle all slower versions of that CPU. That's not always the case, however. Modern motherboards accommodate different processor speeds by varying two settings, memory bus speed and CPU multiplier.

Memory bus speed

This is the speed at which the CPU communicates with system memory. Intel Pentium-class chipsets support Intel's official memory bus speeds of 50 MHz, 60 MHz, and 66 MHz. Third-party chipsets also support these standard speeds, and may also support other speeds, including 75 MHz, 83 MHz, 90 MHz, 95 MHz, 100 MHz, and 112 MHz. All sixth-generation chipsets support a 66 MHz memory bus speed, which is called front-side bus (FSB) speed on sixth-generation systems. Later versions may also support 100 MHz and 133 MHz, along with various intermediate speeds. Early Athlon motherboards supported only the 100 MHz FSB (although it is "double-pumped" and so referred to as 200 MHz). Later Athlon motherboards also support the 133/266 MHz FSB, and the latest also support the 166/333 MHz FSB.

For many motherboards, including most Intel motherboards and third-party motherboards, setting the memory bus speed also specifies the PCI bus speed. These motherboards, called synchronous motherboards, divide the memory bus speed by a fixed factor to determine PCI bus speed. For 50 MHz, 60 MHz, and 66 MHz memory bus speeds, the divisor is 2.0, which runs the PCI bus at 25 MHz, 30 MHz, and 33 MHz, respectively. For a 100 MHz bus, the divisor is 3.0, which runs the PCI bus at 33.3 MHz. For 133 MHz, the divisor is 4.0, which also runs the PCI bus at 33.3 MHz.

Some motherboards, called asynchronous motherboards, allow memory bus speed and PCI bus speed to be set independently, usually by means of jumpers on the motherboard. They do this to allow using faster memory bus speeds while limiting PCI bus speeds to 33 MHz, which is the fastest reliable setting for the PCI bus. Otherwise, for example, with a processor designed to use a 100 MHz FSB, setting the memory bus to 133 MHz would run the PCI bus at 44.4 MHz (one third of 133), which is much too fast for reliability.

CPU multiplier

This specifies the multiple of memory bus speed at which the CPU runs internally. Modern chipsets may support CPU multipliers from 1.5x to 10.0x or higher. For example, a 90 MHz Pentium CPU uses a 60 MHz memory bus speed and a 1.5x CPU multiplier, while a 1.2 GHz Pentium III uses a 133 MHz memory bus speed and a 9.0x CPU multiplier. All recent AMD and Intel CPUs have a locked multiplier, which means the only way to run them faster than their rated speed is to set the FSB for something above nominal. For example, setting the FSB to 150 MHz rather than the nominal 133 MHz for a 1.2 GHz Pentium III runs the CPU at 1350 MHz (150 x 9.0). That results in a small performance increase, but also reduces system stability, because the PCI bus is running at 37.5 MHz (150/4) rather than its design speed of 33.3 MHz.

On fifth-generation and early sixth-generation motherboards, you can generally determine the range of CPU speeds supported by a chipset and motherboard by examining the different memory bus speeds and CPU multipliers that it supports. For example, a typical Pentium motherboard may provide nominal memory bus speeds of 50, 60, and 66 MHz (actually, 50.0, 60.0, and 66.6 MHz), and CPU multipliers of 1.0, 1.5, 2.0, 2.5, and 3.0x. This motherboard will support Pentiums running at 75 MHz (50 x 1.5); 90 MHz (60 x 1.5); 100 MHz (66 x 1.5); 120 MHz (60 x 2.0); 133 MHz (66 x 2.0); 150 MHz (60 x 2.5); 166 MHz (66 x 2.5); and 200 MHz (66 x 3.0). The motherboard could also be configured for speeds—e.g., 180 MHz (60 x 3.0x)—for which no processors are available.

You may also have a choice of two or more speed/multiplier combinations to use. For example, you could install a 100 MHz CPU on this system board using either 66 MHz with a 1.5 multiplier or 50 MHz with a 2.0 multiplier. Generally, use the combination that uses the fastest memory bus speed, which provides faster performance. However, sometimes it makes sense to choose a slower memory bus speed. For example, you may install a 450 MHz CPU and memory that is allegedly rated for 100 MHz, but find that the system is unreliable using 100 MHz and 4.5x. Changing to 90 MHz and 5.0x runs the processor at the same speed, but the memory at only 90 MHz rather than 100 MHz, which may allow the system to run reliably with only a small performance hit due to the slower memory speed.

Not all motherboards allow these two parameters to be set independently. Many motherboards, including early sixth-generation Intel models, allow you to set only CPU speed, which in turn selects a predetermined combination of memory bus speed and CPU multiplier. This is done to prevent overclocking , or running the CPU at higher than its rated speed, a practice that Intel naturally discourages. Late sixth-generation and seventh-generation Intel motherboards boards have no user-accessible speed settings at all, depending on the CPU to identify itself to the motherboard and setting the FSB speed and CPU multiplier accordingly.

Other motherboards, including third-party models that support Intel sixth-generation CPUs, may provide a means to set both FSB speed and CPU multiplier, either by means of a jumper setting or a menu selection in BIOS Setup. Setting FSB manually is ineffective for nearly all Intel sixth-generation and later processors, all AMD Duron processors, and all Socket A Athlon processors because they use multiplier locks.

Intel locks the CPU multiplier internally, which effectively means there's no way to change the multiplier on an Intel processor. AMD uses a method that makes the multiplier setting accessible externally. It's possible to change the CPU multiplier setting on AMD Athlon and Duron CPUs by using so-called "magic fingers" or even a graphite pencil to connect different traces on the face of the processor. We don't recommend doing that, but if you must experiment you can find detailed instructions on enthusiast sites like AnandTech (http://www.anandtech.com).

FSB speeds, on the other hand, are externally determined and not locked. That means you can change the processor speed by setting the FSB speed to some value other than nominal, assuming that the motherboard gives you that option. For example, a 1.2 GHz AMD Athlon is designed to operate with a 133 MHz FSB speed and a 9.0x multiplier. There's nothing you can do to boost the multiplier (short of getting out your pencil), but you can set the FSB speed to a faster value, say 140 MHz. That causes the processor to run at 1260 MHz rather than 1200 MHz, a gain so minor that it will not be noticeable.

Some motherboards limit CPU speed by the CPU socket they use. For example, 166 MHz and faster Pentium CPUs do not physically fit Socket 5 motherboards, so although the chipset may support faster speeds, the socket prevents faster processors from being installed. Also note that the nominal speed of some Intel-compatible CPUs is inaccurate. This is not due to error, but because some manufacturers, notably Cyrix, label processors with performance ratings rather than actual speeds, intending to convey an approximate Pentium-equivalent performance level rather than the actual chip speed. For example, the Cyrix 6x86 PR166+ CPU supposedly provides performance similar to that of a 166 MHz Intel Pentium, but actually runs at 133 MHz, and should be jumpered accordingly. Running it at 166 MHz results in unreliable operation at best.

In 2001, AMD revived the hoary PR system for Athlon XP processors. Because the Athlon cannot compete with the Intel Pentium 4 on raw clock speed, AMD labels Athlon XP and MP processors with numbers that are higher than the actual clock speed but that supposedly indicate relative performance. For example, the Athlon XP 2000+ processor actually operates at 1.67 GHz. There is no danger of misconfiguring an Athlon XP system, however, because all motherboards that accept that processor configure the proper clock speed automatically.

3.1.2.3 Multiple CPU Support

Some chipsets support motherboard configurations with multiple CPU sockets. These chipsets coordinate operations between the multiple processors and memory, which is necessary but not sufficient to allow symmetric multiprocessing (SMP) operation. In addition to chipset support, SMP requires:

CPU SMP support

Non-Intel CPUs other than the AMD Athlon and Duron lack the circuitry to support SMP. Intel fifth- and sixth-generation CPUs support SMP, except Coppermine128-core or later Celerons and most Coppermine-core Pentium IIIs at 1 GHz and faster. The Pentium 4 does not support SMP except in its more expensive Xeon variant. AMD produces a special version of the Athlon called the Athlon MP that supports SMP applications in conjunction with the AMD-760MP or AMD-760MPX chipset.

Operating system SMP support

DOS, Windows 3.X, and Windows 9X do not support SMP. Running one of these operating systems on an SMP PC uses only one of the CPUs. Windows NT/2000/XP Pro supports SMP, as do most Intel-based Unix implementations, including Linux. Note that Windows XP Home does not support SMP.

If you run an SMP-aware operating system, consider buying a motherboard that supports dual CPUs. The incremental cost is generally small, and the increased performance is significant. Prior to 2001, Intel's pricing strategy often made buying two mid-range CPUs less expensive than buying one top-of-the-line CPU. With the collapse in CPU pricing that occurred during 2001, that is no longer true. However, CPUs are now so inexpensive that adding a second CPU is a cheap way to improve system performance. Using two mid-range CPUs usually provides much better overall performance under Windows NT/2000/XP or Linux than using one fast CPU.

If you need to upgrade an existing single-processor Intel system to dual CPUs and the exact CPU you currently use is no longer available, see the detailed Intel Processor Specification Update (http://developer.intel.com/design/processor/) for a matrix of which currently available processors can be mixed and matched. In general, the safest course is to use two processors with identical S-Specs (Intel's identifying number for minor variations of a processor), but processors with different S-Specs can sometimes be used together in an SMP system, with some restrictions. For AMD systems, we recommend using only identical CPUs.

3.1.2.4 Memory and Cache Support

Chipsets vary significantly in the support they offer for memory and cache, as follows:

Maximum system memory

Every chipset places an absolute limit on the amount of RAM that it supports. For recent and current chipsets, this limit ranges from as low as 64 MB to 4 GB or more. Chipsets also limit the number of memory sockets and the number of SDRAM chips or RDRAM devices, which in conjunction with available memory densities may enforce a usable memory limit much lower than the theoretical limit.

Memory type

RAM is available in a bewildering array of form factors (e.g., SIMM, 168-pin and 184-pin DIMMs, 168-pin and 184-pin RIMMs), access types (e.g., SDR-SDRAM, DDR-SDRAM, RDRAM), logical types (e.g., registered, unbuffered), voltages, densities (e.g., 4Meg X 72, 32Meg X 64), error correction (non-parity or ECC), and so on. Different chipsets support different combinations of these characteristics. Some chipsets optimize their performance with one type of RAM at the expense of performance with other types, so make sure if your chipset supports different RAM types that you choose the type for which it is optimized. Most motherboards that support more than one type of RAM are "transition" motherboards—produced during a shift from one type of RAM to another—and are usually optimized for the earlier type of memory. For example, motherboards that support SDR-SDRAM and DDR-SDRAM may be optimized for SDR-SDRAM and provide little or no performance benefit with DDR-SDRAM. If in doubt, check the motherboard documentation.

Choose a chipset that supports the RAM characteristics you require. If you are buying RAM for an existing motherboard, make sure that the RAM is of a type supported by the chipset. In the past, when memory cost $50 a megabyte, making sure that a replacement motherboard could use your existing RAM was a major issue. Nowadays, with RAM selling at very low prices, this is much less important.

Secondary (L2) cache size and type

Even the fastest standard memory is much too slow to keep pace with a modern CPU. To prevent the CPU from spending most of its time waiting for data to be read from main memory, modern chipsets support secondary (also called level 2 or L2) cache. L2 cache is a relatively small amount of very fast (and expensive) memory that sits between the CPU and main memory, buffering requests. Sixth- and seventh-generation CPUs like the Pentium III, Celeron, Athlon, Duron, and Pentium 4 embed L2 cache in the CPU package itself or actually place it on-die with the CPU. Fifth-generation CPUs, including the Pentium, and hybrid CPUs like the AMD K6 and Cyrix 6x86 use discrete cache installed on the motherboard.

Cache size

Most fifth-generation chipsets support 256 KB or 512 KB of L2 cache, although some support only 128 KB. All other things being equal, the larger the L2 cache, the faster the system will be. Many motherboards whose chipsets support 512 KB L2 cache have less than that installed. If you are doing a minor motherboard upgrade, e.g., replacing a 100 MHz Pentium with a 200 MHz Pentium, upgrading the cache at the same time is inexpensive and can increase performance noticeably, particularly if the system currently has little or no cache RAM installed. Do note, however, that spending even $25 to upgrade L2 cache may not be a wise choice, given the low cost of current motherboards, processors, and memory. Size is not an issue for sixth- and seventh-generation chipsets, which support the cache size embedded on the processor.

Cache type

Pentium-class chipsets support one or more cache types. In increasing order of performance and desirability, they are asynchronous, synchronous burst, and pipeline burst. The policy that L2 cache uses to write data also differs. Write-through cache writes changes immediately to both the L2 cache and main memory to preserve coherence. With write-back cache, the CPU writes changes to L2 cache, but not directly to main memory. This provides higher performance, at the expense of requiring additional circuitry to maintain coherence between main memory and L2 cache. Cache type is not an issue for sixth- and seventh-generation chipsets, which support the cache type used by the processor.

Cacheable memory

Sixth- and seventh-generation chipsets all support a cacheable memory area larger than the amount of memory that can be physically installed with current technology. Fifth-generation chipsets differ in the maximum amount of main memory they can support with the L2 cache. This figure is unrelated to the maximum amount of main memory that the chipset supports, but may vary with the amount of L2 cache installed. For example, the three Pentium-class chipsets that you are most likely to encounter are the Intel 430 HX (Triton II), the 430VX (Triton III), and the 430TX. The 430HX supports 512 MB main memory, all of which it can cache; the 430VX supports 128 MB, but can cache only 64 MB; the 430TX supports 256 MB, but can cache only 64 MB. Although the 430HX chipset itself can theoretically cache up to 512 MB of main memory, the amount that it can cache in practice depends on the amount of cache RAM installed on the motherboard. Most 430HX motherboards include the full amount of TAG RAM needed to cache 512 MB, but some do not. Never install more RAM on a motherboard than it can cache. Doing so can actually make the motherboard slower than it was before you added RAM.

3.1.2.5 Chipset Buffers

One important duty of the chipset is to arbitrate transfer of data between the CPU, memory, and the local bus. Many factors determine how efficient such transfers are, including the quality of the chipset itself, the CPU type and speed, and the size and type of the L2 cache. Because many components that operate at widely different speeds are involved, managing these transfers efficiently is critical to fast transfers and overall system performance.

When the CPU requests data from main memory, the chipset first determines if that data is already available in L2 cache. If so, it retrieves it from cache—which is much faster than main memory—and sends it to the CPU. If the data is not cached, the chipset retrieves it from main memory, sends it to the CPU, and also writes it to the cache (in case the CPU requests the same data in the near future). The timing of these transfers is controlled by the chipset, and some chipsets are better than others in any particular memory/cache/CPU environment.

Besides CPURAM transfers, the chipset also manages CPUPCI transfers and PCIRAM transfers. On systems with AGP video, the chipset also handles direct transfers between the CPU and the AGP video adapter. Nor are these transfers purely sequential. At any moment, the chipset may be arbitrating many transfers simultaneously. With many components doing different things at the same time and at different speeds, it's up to the chipset to juggle all of these conflicting demands for attention. One way it accommodates these demands is by using buffers to store data temporarily while other operations are in progress. Chipsets vary in the number of buffers they have available for various purposes, and how efficient those buffers are.

3.1.2.6 Memory Timing

Because there is a huge disparity between CPU speed and main memory speed, the CPU must sometimes wait one or more clock cycles for memory to supply data it requests. These enforced periods of CPU inactivity are called wait states, and the goal of any chipset is to minimize them. Wait states cannot be completely avoided, however, unless memory is fast enough to keep pace with the CPU, an impossible state of affairs given the extremely high cost of very fast memory and the absolute limitations on the speed of conventional memory. Using fast main memory and a large, fast, efficient L2 cache can minimize wait states, but can never completely eliminate them.

When L2 cache is temporarily unable to supply the data needed by the CPU, the chipset inserts a wait state—basically a command to the processor to twiddle its thumbs until the data it requested becomes available. Cache reads and writes data in 32-byte chunks called lines, but main memory is accessed in 64-bit (8-byte) increments, so it takes four separate memory read or write operations to transfer a cache line. The first operation is the slowest, because the memory address must be looked up and accessed before the first transfer can occur, a delay called latency. The next three transfers are much faster, because data is read from or written to sequential contiguous memory addresses based on the address used for the first transfer, eliminating latency.

These differing memory timings are specified in the form X-Y-Y-Y, where X is the number of clock cycles required for the first transfer (with latency) and Y the clock cycles required for the remaining transfers. This pattern varies according to the particular chipset, the memory bus speed, and the type of memory. For example, the Intel 430VX chipset using FPM RAM uses 5-3-3-3 timing at 60 MHz, and 6-3-3-3 timing at 66 MHz, while the later 430TX uses 4-3-3-3 and 5-3-3-3, respectively. With EDO RAM, the 430VX timings are 5-2-2-2 (60 MHz) and 6-2-2-2 (66 MHz), and the 430TX timings are 4-2-2-2 (60 MHz) and 5-2-2-2 (66 MHz). With BEDO or SDRAM, the 430VX 60/66 MHz and 430TX 60/66 MHz timings are 6-1-1-1 / 7-1-1-1 and 5-1-1-1 / 5-1-1-1, respectively.

Some BIOS chipset setup programs allow you to select only among predefined memory timings. For these setup programs, you choose a single memory timing setting. Other setup programs allow you to specify timing for the initial access independently of the timing for subsequent accesses. For these setup programs, you choose two separate settings. The first, in the form X-2-2-2, allows you to specify a value for X that sets timing for the initial access. The second, in the form 4-Y-Y-Y, allows you to specify a single value that will be used for all subsequent accesses. With either type of chipset setup program, choose the fastest timing supported by your chipset and memory.

Recent chipsets, including all sixth- and seventh-generation chipsets, automatically detect the type, size, speed, and optimal timing settings for installed memory, and configure it accordingly, a process called memory autodetection. To do so, however, they require help from the memory itself. RAM supports autodetection using a feature called Serial Presence Detect (SPD). SPD uses an EEPROM that is physically a part of the RAM module to store information about the type and configuration of that module. All current sixth-generation chipsets for Intel and AMD processors are optimized for SDR-SDRAM or DDR-SDRAM, except the Intel 820 and 840, which are optimized for RDRAM. The first seventh-generation chipset, the Intel 850, supports the Pentium 4 with RDRAM. The later Intel 845 chipset supports the Pentium 4 with SDR-SDRAM or DDR-SDRAM.

3.1.2.7 I/O Bus Support

The chipset determines what I/O bus standards the motherboard can support. The chipset manages the I/O buses, arbitrating data transfer between them, the CPU, and system memory. The chipset features determine which I/O buses the system supports, the speed at which the buses operate, and what additional related system features are supported. Depending on how you count, half a dozen or more I/O bus standards have been in common use since the first PCs. In order of their appearance, they include:

Industry Standard Architecture (ISA)

Used in 8-bit form in the PC and XT and 16-bit form in the PC/AT. Obsolete, but most motherboards made through 2000 provided at least one ISA slot for legacy cards. By late 2000, many new motherboard models shipped without ISA slots. Good riddance, we say.

MicroChannel Architecture (MCA)

An IBM standard that never caught on. Obsolete.

Extended Industry Standard Architecture (EISA)

An open standard developed by Compaq and eight other PC makers (the so-called "Gang of Nine") to compete with the propriety MCA, but which achieved only limited acceptance, primarily in servers. Obsolete.

VESA Local Bus (VLB)

An open standard that was widely used in 486 systems, but whose technical shortcomings made it inappropriate for Pentium and later systems. Obsolete.

Peripheral Component Interconnect (PCI)

An open standard developed by Intel, used on late-model 486 systems and almost universally for Pentium and later systems. Older PCI systems use the PCI 2.0 standard, which limits upgrade possibilities. Newer motherboards use PCI 2.1, and current models use PCI 2.2. The PCI 2.3 Specification was approved in summer 2001, and most motherboards shipping in 2002 will be PCI 2.3 compliant. PCI 3.0 will likely be finalized in 2002, and boards using that specification will begin shipping in late 2002 or early 2003.

Accelerated Graphics Port (AGP)

A dedicated I/O port used on late-model fifth-generation and most sixth- and seventh-generation motherboards to provide high-performance graphics. Not technically a bus because it supports only one device, but thought of as a bus because it is implemented as an expansion slot. Note that many current motherboards support only AGP 2.0 1.5V video adapters, and cannot accept legacy 3.3V AGP video adapters.

All but the most recent motherboards provide a mix of expansion slots, usually four or five PCI slots and one shared ISA/PCI slot that can accept either type of expansion card. The chipset provides bridging functions between these I/O buses (the PCIISA bridge) and between I/O buses and other system buses, including the memory bus. The very latest motherboards have no ISA slots at all, and contain only PCI slots and perhaps an AGP slot.

3.1.2.8 Embedded Feature Support

The chipset also provides various embedded low-level system functions and features. Many low-level system functions, e.g., the Programmable Interrupt Controller (PIC), are well standardized and have been so for years, so there is little to choose among different chipsets on that basis. But features differ between chipsets, not so much between competing chipset models as between generations. Chipsets of the same generation generally implement similar features in comparable ways, so these are not issues for choosing between current chipsets. The following features are important when evaluating the upgradability of a motherboard that uses an older chipset:

ATA interface

Any modern motherboard provides an embedded dual-channel ATA/ATAPI interface, which supports a total of four ATA/ATAPI devices, two per channel. But not all embedded ATA interfaces are equal. All current motherboards support at least ATA/66, most support ATA/100, and some support the proposed ATA/133 standard, which is likely to be rendered moot by the forthcoming Serial ATA standard. Current hard disks barely saturate an ATA/33 interface, but choosing a chipset that supports at least ATA/100 makes hard disk upgrades easier. If you are upgrading a system with an old motherboard, the chipset should support at least PIO-4 and independent device timing, which allows two dissimilar devices on one channel to both run at their optimal speed. If you are upgrading a system that uses an otherwise suitable older motherboard that does not support these features, you can disable the embedded ATA interface and install an ATA expansion card that supports recent ATA standards.

The current ATA standard specifies 28-bit addressing, which places an absolute upper limit of 128 GB on ATA drives that use the standard 512 bytes/sector. The Maxtor Big Drive Initiative extends ATA addressing to 48 bits, which allows drive sizes of up to 128 PB, which is to say a binary million times larger than the current limit. This initiative is backed by other major industry players, including Microsoft, and we expect that it will succeed in becoming the new ATA standard. We do not expect compliant motherboards to begin appearing much before mid-2002. However, we expect that inexpensive PCI adapters will be available from vendors like Promise Technology that will allow retrofitting Big Drive functions into existing systems.

DMA Controller and DMA mode support

Direct Memory Access (DMA) is a means to transfer data between devices without using the CPU as an intermediary, therefore increasing system performance. For example, a hard drive can use DMA to transfer data to and from memory without passing it through the CPU, allowing the CPU to do other things at the same time. DMA can be implemented using the DMA controller on the motherboard itself (called first-party DMA) or by using a DMA controller embedded on a device (called third-party DMA or bus mastering DMA). Older chipsets provide a limited selection of slower DMA modes and may make DMA available only on some expansion slots, which are typically differentiated by being a different color than non-DMA slots. Newer chipsets provide additional DMA modes, faster DMA transfers, and DMA capability on all expansion slots.

Plug-N-Play (PnP) support

The PnP standard is closely linked to the PCI standard. Systems that have chipsets, BIOSes, peripherals, and operating systems that are PnP-aware recognize and configure installed hardware automatically, eliminating IRQ, I/O base address, and DMA resource conflicts. A system that uses a non-PnP chipset is too old to upgrade effectively other than by replacing the motherboard.

Universal Serial Bus (USB) support

USB will eventually replace legacy I/O port standards, including serial, parallel, keyboard, and mouse ports. Most motherboards shipped since 1996 use chipsets that are USB 1.0 or 1.1 compliant. All current chipsets include USB support. In June 2002, Intel was the first manufacturer to begin volume shipments of motherboards that support the USB 2.0 standard, which increases maximum data rates forty-fold to 480 Mb/s. To use USB on an older system, install a $25 PCI USB card—the ADS Technologies USB Port (http://www.adstech.com) is one we've used successfully—or replace the motherboard with one whose chipset provides native USB support. To add USB 2.0 support to a current system, you can install an add-on PCI adapter, which is available from Adaptec and others.

Your motherboard may have "hidden" USB support. Most AT form factor motherboards made after 1996 have USB support, but only as sets of header pins on the motherboard. Using USB on these motherboards requires adding a port extender to route USB signals from the motherboard to the back panel. Such port extenders are readily available for $5 to $10 from motherboard makers and computer stores. However, note that the USB controllers used in motherboards of that vintage are often buggy, so you may be better advised to ignore the embedded USB ports and install a current third-party PCI USB controller.

Accelerated Graphics Port (AGP) support

AGP provides a fast, dedicated channel between the CPU and an AGP video adapter, moving video data off the memory bus. Because AGP is implemented as an expansion slot, any AGP-capable motherboard by definition uses an AGP-capable chipset and provides an AGP slot. But the AGP standard specifies different modes, including 1X, 2X, and 4X, not all of which are implemented in all chipsets. AGP 1X is obsolete, although you can use an AGP 2X or 4X adapter in an AGP 1X motherboard. Current motherboards provide AGP 4X support, but the additional potential throughput of 4X is of little or no benefit with current video adapters, few of which can saturate even 2X AGP. Intel has published an 8X AGP standard. Motherboards that support 8X AGP should begin shipping in 2002, although once again it's unlikely that video adapters will be available that require the additional throughput. A few motherboards, most of which are intended for workstations, have an AGP Pro slot. AGP Pro is a backward-compatible extension of AGP that provides the additional power needed by some very high performance video cards. Unless you plan to install an AGP Pro video adapter, you have no need of AGP Pro. When you buy a motherboard, make sure it supports AGP 4X. Anything more is probably unnecessary, and anything less means the motherboard is out of date.

3.1.2.9 Identifying Chipsets

To make upgrade decisions based on chipset, you must identify which chipset you have. If you are buying a new motherboard, this is straightforward. The motherboard specifications always list the chipset. When you upgrade an existing system, determining which chipset it uses can be difficult. You can identify the chipset in one of the following ways:

• If you have the documentation for the system board or the PC, the chipset it uses will be listed in the detailed specifications. If you do not have the documentation, but can identify the make and model of the PC or motherboard, the manufacturer's web site should list the chipset it uses.

• Use a third-party diagnostics program like CheckIt (http://www.checkit.com)or SiSoft Sandra (http://www.sisoftware.co.uk) to display system information. Figure 3-2 shows the results of running Sandra on an old Pentium III system. If you have no documentation, using a diagnostic utility is by far the easiest way to identify the chipset.

  Figure 3-2. SiSoft Sandra identifies the chipset as an Intel 440BX

  

  Windows-based diagnostics programs either do not run or provide very limited functionality under Windows NT/2000, both of which limit access of applications to underlying hardware. If your system runs NT/2000, the best alternative is to boot DOS on it and run a DOS-based diagnostic utility.

• Identify the chipset visually by examining the motherboard. Most chipsets have two chips, although some have only one and a few have three or more. Look for relatively large, usually square chips that are labeled with the name of a chipset manufacturer—Intel, VIA, SiS, ALi, Opti, etc. Record the string of numbers and letters that identifies each such chip. Then visit that manufacturer's web site and search for those strings to determine which chipset, if any, those chips belong to. For example, if you locate an Intel chip labeled 82438VX, a quick check of the Intel web site tells you that that chip is part of an Intel 430VX chipset (Intel calls most of its recent chipset models a PCIset or an AGPset). Note, however, that particularly on recent systems whose chipsets run at very high speeds, the chips that make up the chipset may have permanently installed heatsinks, which make it impossible to read the labels.

3.1.2.10 Fifth-Generation Chipsets

Systems that use Intel fifth-generation chipsets are now too old to be upgraded effectively at the CPU/motherboard level, although such systems can sometimes usefully be upgraded by adding memory, a faster hard disk, and so on. Systems based on late-model third-party fifth-generation Super Socket 7 chipsets, such as the VIA MVP3, are still marginally upgradable by replacing the existing processor with a faster model, although Socket 7 processors are becoming rare and will likely disappear entirely by the end of 2002. Even if you have such a system that is technically upgradable, it is usually not cost-effective to do so. Instead, retire that system in situ to less demanding duties, and build a new system based on current technology.

3.1.2.11 Sixth-Generation Intel and Intel-Compatible Chipset Characteristics

Intel has produced numerous sixth-generation chipsets in the 4-series and 8-series lines. Table 3-1 lists the characteristics of Intel 4-series desktop chipsets. These chipsets are obsolescent, although systems built on them may still be used productively and may even be reasonable upgrade candidates. In fact, motherboards based on the 440BX chipset were still being sold new as of June 2002. These chipsets do not support the most recent sixth-generation processors and lack modern features—support for PC133 SDRAM and the 133 MHz FSB, PCI 2.2, fast ATA modes, 256 megabit memory chips, and so on—but remain useful nonetheless.

Processor support varies between steppings of chipsets, CPUs, and motherboards. For example, early steppings of the 815E chipset do not support Tualatin-core Pentium III and Celeron CPUs, while later steppings do. Similarly, some sixth-generation Intel chipsets support Covington-core and Mendocino-core Celerons, but not CoppermineT-core Celerons. Even if the chipset supports a particular CPU stepping, a motherboard that uses that chipset may not support that CPU. For detailed information about CPU support, see the Intel Specification Update for the chipset (http://developer.intel.com/design/chipsets/) and CPU (http://developer.intel.com/design/processor/) in question.

Table 3-1. Intel 4-series sixth-generation chipset characteristics

	440EX	440LX	440BX	440ZX	440ZX66
CPU support
Pentium II (66 FSB)
Pentium II (100 FSB)
Pentium III (100 FSB)
Pentium III (133 FSB)
Celeron (66 FSB)
Celeron (100 FSB)
Number of CPUs	1	2	2	1	1
Host bus
66 MHz
100 MHz
133 MHz
Memory support
EDO	256 MB	1 GB
PC66 SDRAM	256 MB	512 MB	1 GB	256 MB	256 MB
PC100 SDRAM			1 GB	256 MB
PC133 SDRAM
RDRAM
ECC support
64/128/256 Mbit	/ /	/ /	/ /	/ /	/ /
EDO/SDRAM rows	4	6	8	4	4
RDRAM devices
Asynchronous setting
PCI
PCI bus version	2.1	2.1	2.1	2.1	2.1
Concurrent PCI
Southbridge	PIIX4E	PIIX4	PIIX4E	PIIX4E	PIIX4E
ATA support	33	33	33	33	33
AGP support
Integrated graphics
1X/2X/4X AGP	/ /	/ /	/ /	/ /	/ /

Table 3-2 lists the characteristics of obsolescent and current Intel 8-series sixth-generation desktop chipsets. Intel has produced some chipsets in two versions, one using the Integrated Controller Hub (ICH) and the other using ICH2. The major differences between these versions are that ICH2 supports ATA100 (versus ATA66 for ICH) and provides two USB controllers with four USB ports (versus one controller/two ports for ICH). The ICH2 models did not replace the ICH models immediately. ICH models continued in production because Intel sold them for a few dollars less than ICH2 models. Every penny counts (literally) to mass-market vendors, who build PCs for the lowest possible cost and sell on very slim margins.

Table 3-2. Intel 8-series sixth-generation chipset characteristics

	810	810E/ 810E2	820/820E	840	815/815E
CPU support
Pentium II (100 FSB)
Pentium III (100 FSB)
Pentium III (133 FSB)
Celeron (66 FSB)
Celeron (100 FSB)
Number of CPUs	1	1	1-2	1-2	1
Host bus
66 MHz
100 MHz
133 MHz
Memory support
PC100 SDRAM	512 MB	512 MB			512 MB
PC133 SDRAM					512 MB
PC600 RDRAM			1 GB	8 GB
PC700 RDRAM			1 GB
PC800 RDRAM			1 GB	8 GB
Dual memory channel
ECC support
64/128/256 Mbit	/ /	/ /	/ /	/ /	/ /
SDRAM rows	4	4			6
RDRAM devices			32	32 X 2
Asynchronous setting
PCI
PCI bus version	2.2	2.2	2.2	2.2	2.2
Concurrent PCI
Multiple PCI segments
Northbridge	82810	82810E	82820 (820) 82820E (820E)	82840	82815
Southbridge	ICH	ICH (810E) ICH2 (810E2)	ICH (820) ICH2 (820E)	ICH	ICH (815) ICH2 (815E)
ATA support	66	66 (810E) 100 (810E2)	66 (820) 100 (820E)	66	66 (815) 100 (815E)
AGP support
Integrated graphics	82810	82810E			82815
1X/2X/4X AGP	/ /	/ /	/ /	/ /	/ /

Intel intended the ill-starred 820 chipset to be the direct replacement for the venerable (and still respected) 440BX chipset. The 820 was a miserable failure, primarily because it was designed to use expensive Rambus RDRAM memory. With the 815 and the follow-on 815E, Intel finally got it right. The 815E is a superb chipset, fully worthy of inheriting the 440BX's crown as the best sixth-generation chipset available. The 815 and 815E chipsets are available in two variants. The 815 and 815E include an embedded graphics adapter. The 815P and 815EP do not.

Figure 3-3 shows the Intel 815E, which is a representative Intel 8-series chipset. Intel calls the 82815 Northbridge the Graphics and AGP Memory Controller Hub (GMCH) and the 82801BA Southbridge the I/O Controller Hub (ICH2). Figure 3-1 early in this chapter showed the block diagram for the Intel 815E chipset.

Figure 3-3. The Intel 815E chipset (image courtesy of Intel Corporation)

Intel makes the different 815 variants by mixing and matching the 82815 GMCH (Graphics and Memory Controller Hub) or the 82815EP MCH (Memory Controller Hub) with the 82801AA ICH (I/O Controller Hub) or the 82801BA ICH2 (I/O Controller Hub 2). Those variants with the MCH are designated "P" to indicate the absence of embedded graphics. Those that use the 82801BA are designated "E" to indicate the presence of the enhanced ICH2 Southbridge.

The 815E has only two real limitations. First, it supports a maximum of 512 MB memory. At the time the 815E was designed, 512 MB was a reasonable limit. Few people could afford more. But with the plummeting price of RAM, 512 MB began to look like a real limitation. Second, the 815E officially supports only one CPU (although at least one motherboard maker, Acorp, has produced a dual-CPU 815E motherboard). If the 815E supported 1 GB or more of memory and dual processors, it would be the perfect sixth-generation chipset for Intel processors. As it is, the 815E is merely the best available.

Although Intel originally claimed SDRAM support for the 820 chipset, they implemented that support via the Memory Translator Hub (MTH). After many motherboards had been shipped, Intel found that the MTH corrupted data. Intel recalled the CC820 motherboard, which was the only SDRAM 820-based motherboard they produced. Third-party motherboard makers also recalled their 820-based SDRAM motherboards. However, as with any recall, not everyone returned the boards, so you may still encounter 820-based motherboards that use SDRAM. If you discover such a board, stop using it immediately. Using it risks your data.

Other companies, including VIA Technologies, Inc., Acer Laboratories, Inc. (ALi), and Silicon Integrated Systems (SiS), make chipsets for Intel sixth-generation CPUs. In our experience, these third-party chipsets are slower, buggier, less compatible, and less stable than Intel chipsets. So why would any motherboard maker use them? Three reasons:

Price

Third-party chipsets generally sell for $5 or $10 less than Intel chipsets. If you are building a motherboard to a price-point, the difference between a $40 Intel chipset and a $30 third-party chipset is huge. Although we won't say that all motherboards with non-Intel chipsets are built with cost as the only priority, it is certainly a significant issue.

Feature sets

Intel is conservative about adding support for new technologies to their chipsets. So, for example, when DDR-SDRAM became all the rage, Intel did not produce a chipset that supported it, so motherboard makers who wanted to produce DDR-SDRAM motherboards had no choice but to use a third-party chipset. Similarly, the last mainstream SMP-capable chipset Intel produced was the 440BX, which lacks support for such essentials as the 133 MHz FSB. Motherboard makers who want to produce a modern dual Pentium III board have little choice but to use a third-party chipset like the VIA Apollo Pro133A or Apollo Pro266.

Availability

Despite their higher costs and more limited feature sets, most motherboard makers want Intel chipsets because they perceive (rightly) that Intel chipsets are superior to those produced by other manufacturers. Intel's production capacity, while large, is not unlimited. When Intel ships a new chipset, they are often unable to meet demand. Third-party manufacturers step in with chipsets that have comparable features and are available immediately in volume.

Figure 3-4 shows the block diagram for a typical third-party chipset, in this case the VIA Apollo Pro266T, which supports Intel sixth-generation processors.

Figure 3-4. Block diagram of the VIA Apollo Pro266T chipset (graphic courtesy of VIA Technologies, Inc.)

Of the third-party chipset makers, VIA Technologies is the largest and most prolific. VIA chipsets are found in motherboards made by nearly every major motherboard maker other than Intel, including such premier Taiwanese makers as ASUS, EPoX, and Gigabyte. Figure 3-5 shows the two chips that make up the VIA Apollo Pro266, a representative VIA chipset. The VT8633 chip is the Northbridge and the VT8233 is the Southbridge. As does Intel, VIA mixes and matches various Northbridges and Southbridges from within a given family to produce related chipsets with different features.

Figure 3-5. The VIA Apollo Pro266 chipset (image courtesy of VIA Technologies, Inc.)

Table 3-3 lists the characteristics of several VIA chipsets for Intel sixth-generation processors. The PM601 and Pro133 chipsets are obsolescent. The Pro133A and Pro266 are used in current motherboards. VIA produces variants of these chipsets, which differ chiefly in terms of processor support and whether or not embedded graphics are included. For example, the Apollo Pro133A chipset is also available as the Pro133T, which adds support for Tualatin-core Pentium III processors, and as the PM133, which adds embedded Savage video to the Pro133A chipset. Similarly, the PL133T is based on the Pro133A chipset, but includes Tualatin support and embedded Savage video. As always, the Northbridge determines which processors and memory types are supported, and the Southbridge determines which peripherals are supported.

Table 3-3. Characteristics of VIA chipsets for sixth-generation Intel processors

	PM601	Pro133	Pro133A	Pro266
CPU support
Pentium II (66 FSB)
Pentium II (100 FSB)
P