undefined

System Chipset and Controllers

The system chipset and controllers are the logic circuits that are the intelligence of the motherboard. They are the "traffic cops" of the computer, controlling data transfers between the processor, cache, system buses, peripherals--basically everything inside the computer. Since data flow is such a critical issue in the operation and performance of so many parts of the computer, the chipset is one of the few components that have a truly major impact on your PC's quality, feature set, and speed.

What exactly is a "chipset"? It sounds like something very complex but really is not, although many of the functions it performs are. A chipset is just a set of chips. (He ducks to avoid the flying vegetables.  ) At one time, most of the functions of the chipset were performed by multiple, smaller controller chips. There was a separate chip (often more than one) for each function: controlling the cache, performing direct memory access (DMA), handling interrupts, transferring data over the I/O bus, etc. Over time these chips were integrated to form a single set of chips, or chipset, that implements the various control features on the motherboard. This mirrors the evolution of the microprocessor itself: at one time many of the features on a Pentium for example were on separate chips.

There are several advantages to integration, but the two primary ones are cost reduction and better compatibility (the more things that are done by a single chip or group of chips from one manufacturer, the simpler the design is, and the less chance of a problem). Sometimes the chipset chips are referred to as "ASICs" (application-specific integration circuits), which I suppose they are, although there are many other types of ASICs as well.

Note: Intel also calls their chipsets "PCIsets" and "AGPsets", refering to the system bus technologies the chipsets implement.

The system chipset in most cases does not integrate all of the circuitry needed by the motherboard. Most motherboards have the following controllers on them:

• The system chipset itself.
• The keyboard controller, which manages not only the keyboard but also the integrated PS/2 mouse
• The "Super I/O" chip, which handles input and output from the serial ports, parallel port, floppy disks, and in some cases, the IDE hard disks as well
• Additional built-in controllers that are normally found in expansion cards: video, sound, network and SCSI controllers being the most common.

Note: The term "chipset" is also used to refer to the main processing circuitry on many video cards. The name is used because the concept is similar: a highly-integrated circuit used to perform a set of functions. However, this is a totally different type of chipset, and is not the same as a motherboard (system) chipset.

Chipset Functions and Features

This section describes the various functions performed by the system chipset, and thevarious PC features that the chipset plays a key role in supporting. In doing so, ittouches upon the various system features that are enabled by the choice of chipset made bythe motherboard designer. These capabilities are broken into several categories, asdescribed by the links in the index frame on the left. The descriptions ofchipset functions and features in this section are representative of modern Pentium andPentium Pro class machines. The structure of 486 class and earlier chipsets can be quitedifferent, especially for those machines that use a different local bus than PCI.

Chipset Processor Support

One of the most important decisions made by anyone choosing or building a new PC iswhich processor is desired. The key to making the decision of what type, speed and evenwhat number of processors to use is the motherboard, and in particular the chipset thatcontrols it.

This section describes in more detail the support factors that determine what sorts ofprocessors can be used by the chipset.

Processor Class Support and Optimization

A chipset is designed to work with a specific set of processors in mind. In general,most chipsets only support one "class" or generation of processors: mostchipsets are geared specifically for 486 type systems, Pentium class systems, or PentiumPro / Pentium II systems. The reason for this is simple: the design of the controlcircuitry must be different for each of these processor families due to the different waysthey employ cache, access memory, etc. For example, the Pentium Pro and Pentium II havelevel 2 cache within the CPU package itself, so obviously these need different logic thanthe Pentium, which has level 2 cache on the motherboard. These issues are discussed in more detail in this sectionon processors.

Most good motherboards that support Pentium processors also support their equivalents(or near-equivalents) from AMD (the K5 and K6) and Cyrix (the 6x86 and 6x86MX). Since theywere designed specifically to be Intel alternatives, they work with Intel chipsets (inmost cases), although they sometimes need a different jumper setting. Another optimizationfactor to consider is that these compatibles are not always identical to the Intel chipthey are intended to replace, and in some cases they add additional performance featuresthat can only be taken advantage of by the chipset.

Of course, since Intel is the largest manufacturer of Pentium and Pentium Pro chipsets,that puts AMD and Cyrix at a significant disadvantage. This disadvantage is diminishing,especially with regard to socket 7 motherboards, as non-Intel chipsets become moreprevalent.

Note: The ability of themotherboard to accept dual-voltage processors such as the Pentium with MMX is dependentonly on the voltage regulators, not the chipset. MMX is an instruction set extension anddoes not require a chipset change compared to the standard Pentium.

The introduction of AMD's K6 and Cyrix's 6x86MX processors has clouded the definitions of"processor generation" quite a bit. Functionally, these two processors arereally sixth-generation, but they are designed to fit into a fifth-generation (Pentiumclass) motherboard.

Faster processors require chipset control circuitry capable of handling them. Thespecification of the processor speed is done using two parameters: the memory bus speed,and the processor multiplier. These are discussed in detail in the section on processor architecture. In short,the processor bus and memory bus connect the processor, chipset and memory together.

The memory bus speed is the processor's "external" speed, the speed it talksto the rest of the computer at (as opposed to its internal speed). The memory bus speedalso (normally) dictates the speed of the PCI local bus, which in most motherboards runsat half the memory bus speed. Typical modern memory bus speeds are 50, 60, 66 and 75 MHz.The multiplier represents the factor that the processor multiplies the memory bus speed inorder to obtain its internal speed. Multipliers on modern PCs are normally 1.5x, 2x, 2.5x,3x, 3.5x, 4x, 4.5x or 5x.

The range of the processor speeds supported by the chipset is indicated, generally, bylooking at the range of supported memory bus speeds and multipliers. For example, atypical classic Pentium chipset will support bus speeds of 50 to 66 MHz with a multiplierrange of 1.5x to 3x. This yields speeds of 75, 90, 100, 120, 133, 150, 166 and 200 MHz(along with some in-between values that don't correspond to any processors actually on themarket; for example, 50 MHz and 2.5x yields 125 MHz, a pair of settings not normallyused.)

Note: In some cases thelimiting factor on processor speed is not the chipset, but the physical socket. A Pentium166 MHz will not fit into a Socket 5, so a motherboard using that socket is limited tolower-speed Pentiums (or OverDrive processors).

Note: SomeIntel-compatibles use "equivalence ratings" in naming their chips, which do notrepresent the true speed of the processor. The Cyrix 6x86-PR166+ chip, for example,actually runs at 133 MHz. See here for details on"P ratings".

Some chipsets support the ability to create motherboards with two or four processors onthem. The chipset circuitry coordinates the activities of the processors so that theydon't interfere with each other, and works with the operating system software to share theload between the CPUs for maximum efficiency. The only current standard formultiprocessing in Pentium and Pentium Pro PCs is Intel's version of SMP (symmetricmultiprocessing). It only works with Intel processors.

There is more to running multiple processors than the chipset supporting it, thoughthat is one piece of the puzzle. In addition, the processors must be compatible, and youroperating system must be able to take advantage of the extra CPUs for there to actually bea performance improvement.

Chipset Cache Support

This section discusses chipset features related to the system or secondary cache. Thecache buffers recent memory accesses by the processor, which improves performancedrastically since it operates much faster than the system memory.

Secondary Cache Size

The chipset determines how much level 2 (secondary) cache is supported. Most modernchipsets support cache of either 256KB or 512KB. Note that this is not relevant forPentium Pro machines, which have level 2 cache in the same package as the processor.

Secondary Cache Type

There are three major types of cache currently in use;in increasing order of performance, they are: asynchronous, synchronous burst, andpipeline burst. Each requires different control circuitry, and therefore must beexplicitly support by the chipset.

Secondary Cache Write Policy

There are two different "policies" for controlling the way writes to memoryare done in a cached system: a "write through" cache means that memory writesare sent to memory as soon as the processor sends out the information; a "writeback" cache means the processor writes the information only to the cache, which laterwrites the information to memory at the appropriate time. Write back is in most casessuperior to write-through

System Memory Cacheability

Chipset characteristics control the maximum amount of memory the system can cache. Thisis a very different question than the amount of memory the system can hold. Many chipsets(unfortunately) can hold more memory than they can support with the level 2 cache. The amount of cacheable memory depends on thechipset control circuitry and the amount of tag RAM on the board. It does notdepend on how much memory you currently have in the system.

Note: Some chipsets, likethe Intel 82430HX "Triton II", can cache either up to 64 MB of RAM, or 512 MB.The larger amount is available if an "optional" tag RAM chip is added to the motherboard. In mostcases, the motherboard manufacturer builds this chip into the board, since being able tocache 512 MB is one of the HX chipset's advantages over its other Triton"cousins".

Using more memory than the system can cache will lead to very significant performance degradation.

Chipset Memory Support

There are many different styles and sizes of system memory (RAM), and many differenttechnologies in use today. The chipset dictates many of the allowable characteristics ofthe memory used on your motherboard.

Maximum Memory Support

The chipset dictates the maximum amount of RAM you can have on the motherboard. Formodern systems, this can be as low as 64 MB, or as high as 4 GB. This number can be higher(in some cases much higher) than the maximum amount of memory that can be cached,and this is something to be aware of.

DRAM Technology

The chipset controls whether your motherboard can use FPM, EDO, BEDO, or SDRAM memory (they are all explained here). Changing the memory typeimpacts the way that memory is read and written to, which is controlled by the chipset. Inaddition, some chipsets are better than others with certain types of memory; a chipset canbe optimized to provide faster access to a certain kind of memory, while being lessefficient at using a different kind.

DRAM Packaging and Size Support

Memory for desktop computers comes in two basic types of packages: the SIMM (singlein-line memory module) and DIMM (dual in-line memory module). SIMMs come in two sizes: 30pins and 72 pins. DIMMs are 168 pins. These packages differ not only in physical size butalso in the "width" of the memory they canoutput at once. 30-pin SIMMs produce 8 bits of data at a time, 72-pin SIMMs 32, andDIMMs 64 bits.

The width of memory needed by the motherboard depends on how the chipset accessesmemory, and on the data bus width of the processor. In general, 486 class machines require32-bit wide memory and Pentium and Pentium Pro machines require 64-bit wide memory, butthe chipset design has an influence on this. Some Pentium motherboards will handle 32-bitwide memory (slowly) by changing the way memory is accessed.

The design of the chipset is also a limiting factor on how many SIMM or DIMM slots themotherboard can have. Better chipsets support more slots, giving you more flexibility insetting up and upgrading your PC.

Parity and Error Correction Support

Error correction logic is provided as part of thememory control circuits of the chipset. On modern PCs, both parity and ECC functions areprovided by the use of parity memory; some chipsets support ECC only, using ECC memory. Achipset without at least parity support has no ability to detect or correct memory readerrors.

Chipset Timing and Flow Control

One of the chipset's most important functions is controlling memory reads and writes,and transfers to the local bus (usually PCI and/or AGP) by the processor. The designquality of the chipset, the type of cache, the speed of your system memory, and the typeof processor used all play a part in determining how efficiently information can betransferred to and from the processor, which will dictate, in part, the overallperformance of your machine.

Address Decoding

The chipset performs the function of translating the processor's requests forinstructions and data into addresses that represent the locations in memory where thisinformation can be found.

Cache and Memory Data Transfer

When the processor requests information from the memory, the cache is first checked tosee if it contains the data (since it is much faster than the memory). If it does, thedata is read from there. Otherwise, it is read from the memory and given both to theprocessor for computation, and to the cache to store it in case the processor needs itagain soon. The chipset controls the timing of these transfers. Better chipsets can dothis work more efficiently than others

Bus Buffering and Data Flow

The chipset controls and manages the flow of information from the local bus (PCI on aPentium or later system) to memory, as well as from PCI directly to the processor. On AGPsystems, the chipset also coordinates transfers between the CPU and AGP video card. Forhigher efficiency, there can be several transactions pending at the same time. Since theprocessor, memory, and PCI bus are all operating at different speeds, buffers (temporarydata storage areas) are used to hold data during transfers. Chipsets vary in the amount ofthese buffers that they provide; more buffers mean more ability to support concurrentoperation of the processor, PCI bus, and memory.

Memory System Timing

Since memory is in most cases much slower than the processor it serves, the processoroften must wait for the memory to provide it with information it needs. A "waitstate" is a clock cycle (or "tick") where the processor is idle because itis waiting for the system memory. The chipset's goal is to reduce this waiting as much aspossible; it inserts these wait cycles where necessary to make sure the processor doesn'tget ahead of the system cache or memory. The faster your system memory and cache, thefewer of these wait states need to be performed, which increases performance (very fewwait states are needed for the cache, compared to the system memory, which is kind of thepoint of cache.  ). All of this is also a function of the chipset memory accesscircuitry, and is discussed in detail in the section on memory timing.

Cache on a modern system is stored in 32-byte "lines", meaning information isread from and written to the cache 32 bytes at a time. Since the memory is normally read 8bytes (64 bits) at a time, this means it takes 4 memory reads (or writes) to fill an"entry" in the cache. On the first of these reads or writes, the addressinformation must be provided to the memory, to tell it which location must be used. Afterthis, the next three reads or writes are from consecutive locations, so the speed is muchhigher, because there is no need to send the address for the last three accesses (sincethey are consecutive with the first one). This, of course, greatly improves performance.The delay in accessing the first memory location is referred to as "latency".

Reflecting this technology, cache and memory access timing is often specified usingterminology like this: F-S-S-S, where "F" is the number of cycles for the firstaccess, and "S" is the number for each subsequent consecutive access. An exampleof this speed specification would be "5-2-2-2" which means the first accesstakes 5 clock cycles, and three following it take 2 each. You may also see a speedparameter in your BIOS settings like "x-2-2-2", or "5-x-x-x", becausethe timing for the first access, and for the subsequent accesses, can be set andcontrolled independently on many systems. Remember that the first number doesn't representjust "wait states", part of the reason it takes longer is specifying the addressto read from, as stated above.

Memory Autodetection

Most modern chipsets can automatically detect the type of memory and its speed, andadjust its wait states accordingly to ensure maximum performance without data loss. Ifthis isn't supported, the user must enter the BIOS setup program and set the timing valuesthere.

Chipset Peripheral and I/O Bus Control

Most modern computers use two buses: the ISA (industry standard architecture) bus forslower peripherals, and for compatibility with older components, and PCI (peripheralcomponent interconnect), a high-speed "local bus" for hard disks, video cardsand other high-speed devices. The newest PCs also use the new AGP port for video.

The chipset controls these buses, and transfers information to and from them and theprocessor and memory. The chipset's capabilities determine what kinds of buses the systemcan support, what speed they can run at, and what additional features they may have.

Bus Types

The chipset dictates what sort of buses the system can support; in fact, Intel callsits chipsets "PCIsets" and "AGPsets". Most modern PCs support the ISAand PCI buses, but older chipsets (on 486 class machines) support the VESA Local Bus (VLB)instead of PCI. There are even some 486 PCs that support all three buses on the samemotherboard. The newest PCs also add AGP for video.

Bus Bridges

A "bridge" is a networking term that refers to a piece of hardware thatconnects two dissimilar networks and passes information from the computers on one networkto those on the other, and vice-versa. In an analogous way, the chipset must employ busbridges to connect together the different system bus types it controls. The most common ofthese is the PCI-ISA bridge, which is used to connect together devices on these twodifferent buses.

IDE/ATA Hard Disk Controller

Almost all motherboards now have integrated into them support for four IDE (ATA) harddisks, two on each of two channels. The IDE/ATAinterface is discussed in considerable detail here. Integrating this support makessense for a number of reasons, among them the fact that these drives are on the PCI bus,so this saves an expansion slot (and hence reduces cost).

There are several features related to the IDE interface and its use of the PCI bus thatare controlled by the chipset. The data transfer rate of IDE drives is based on theirusing programmed I/O (PIO) modes, and use of the fastest of these modes depends on supportfrom the PCI bus and chipset. The ability to set a different PIO mode for each of the twodevices on a single IDE channel, called independentdevice timing, is also a chipset feature. Without this, both devices must run at theslower of the two devices' speed.

DMA Controller and DMA Mode Support

Direct memory access (DMA) provides a way for devices to transfer information directlyto and from memory, without the processor's intervention. This was originally conceived asa more efficient means of transferring information to and from memory than programmed I/O,as it allows the processor to do other work (when using an operating system that supportsit). It is still used by many devices, although newer transfer modes are now used forhigh-performance devices like hard disks. DMA is controlled by the chipset's DMAcontroller, and the better (and newer) the controller, the more DMA modes its supports.

Bus mastering is an enhancement of DMA whereby the remote device not only can send datato the memory directly, it actually takes control of the bus, and performs the transferitself instead of using the DMA controller. This cuts down on the overhead of having the(relatively slow) DMA controller talk to the device doing the transfer, further improvingperformance. Bus mastering support is alsoprovided by the chipset.

Interrupt Controller

The interrupt controller provides the means by which I/O devices request attention fromthe processor to deal with data transfers. This work is performed by a pair of Intel 8259interrupt controllers (now integrated into the chipset).

USB Support

USB (Universal Serial Bus) is a new technology intended to replace the currentdedicated ports used for keyboards and mice. It is still unclear as to whether thisstandard will catch on and become popular, because it's been around for a while now but Istill rarely see anyone using it. Support for USB is implemented as part of the chipset.

AGP Support

AGP (Accelerated Graphics Port) is a new "bus" specifically designed by Intelto connect processors to high-speed graphics cards, especially ones performing 3Doperations. It isn't really a bus because it only supports two devices.

Plug and Play

Plug and Play (PnP) is a specification that uses technology enhancements in hardware,BIOSes and operating systems, to enable supported devices to have their system resourceusage (IRQs, I/O port addresses, DMA channels) set automatically. Intended to helpeliminate some of the problems with getting peripheral devices to work without stepping oneach other's toes, Plug and Play requires support from the chipset as well.

Chipset Power Management Support

Most recent chipsets support a group of features that work together to reduce theamount of power used by the PC during idle periods. These initiatives came as a result ofefforts in two main areas: those concerned about the power consumption of PCs which areleft running for as much as 90% of the time even though idle, and laptop PC owners tryingto get more life from a battery charge.

Power management works through a number of BIOSsettings that dictate when to shut down various parts of the computer when it becomes idle.In general, the settings are progressive, so that you have the option to shut down moreparts of the PC as the idle time increases. There are a number of different protocols thatwork together to make power management work.

Note: While powermanagement is a great idea, in practice it isn't always as great as it sounds

Energy Star

Energy Star is the program started by the EPA to "certify" PCs that areconsidered energy efficient and incorporate power management or power use reductionfeatures. Most PCs these days are Energy Star compliant, and display its distinctive logoon the screen when the BIOS boots up.

Advanced Power Management (APM)

Advanced Power Management or APM is the name given to the component in someoperating systems (such as Windows 95) that works with the BIOS to control the powermanagement features of the PC. For example, APM allows you to set parameters in theoperating system to control when various power management features will be activated.

Display Power-Management Signalling

This standard, developed by VESA, specifies a set of signals that can be sent bycompliant video cards to compliant monitors to instruct them to go into power-conservingmodes.

System Management Mode

System Management Mode or SMM is a power-reduction standard for processors,that allows them to automatically and greatly reduce power consumption. It alsoincorporates features such as suspend/resume.

Hard Disk Spindown

Both IDE/ATA and SCSI hard disk accept a command to "spin down" wheninstructed, their contribution to the power conservation effort. The savings here ofcourse is quite minimal, because modern hard disks use little power.

Popular Chipsets

This section discusses some of the more common chipsets used for various processor families. It also contrasts some of the more popular Intel chipsets to give you an idea of what their various strengths and weaknesses are--there are four members of the "430" series of chipsets and the differences between them can be somewhat confusing. This section also provides some concrete examples of the various chipset characteristics discussed in the preceding one, to hopefully give you a better feel for what all of this means.

The chipsets are grouped by "processor generation", much the way processors are in the Reference Guide chapter on processor families. Until 1997, this was a very clean and simple way to organize chipsets, but AMD and Cyrix have clouded the waters somewhat with their decision to make the AMD K6 and Cyrix 6x86MX compatible with Pentium-class chipsets and motherboards. Those two chips are really sixth generation in terms of technology, but they run on fifth-generation motherboards.

For the fifth-generation Intel chipsets, I describe them in "historical perspective". Intel produced six Pentium-class chipsets, and I find it useful to look at them based on what each chipset improved over its predecessor.

Fourth Generation (486 Class) Chipsets

There are no major or dominant chipsets for 486 class machines. There were many different types and speeds of 486s and several companies that produced popular chipsets for these machines. Some of the more popular 486 chipset producers are UMC, SiS and OPTi.

The VESA Local Bus was invented at around the time that the 486 was introduced. As a result, the majority of 486-based motherboards are based on VLB designs. There are some older ones that just use an ISA bus. The 486 is essentially obsolete at this time, although it has clung tenaciously to life through AMD and Cyrix's 5x86 chips (which are essentially high-end 486s running on 486 motherboards). The motherboards specifically designed to support these chips are relatively new, and are among the few 486 designs to support the PCI bus.

One interesting chipset design seen on some of these newer 486 motherboards is the so-called "VIP" motherboard, where VIP stands for "VLB, ISA, PCI". These boards actually incorporate all three buses! Usually they have 3 PCI slots, 3 ISA slots and one VESA Local Bus slot. The idea is to provide an upgrade path for 486 users without forcing them to throw away their VLB video cards. In practice, these boards often have trouble using both the VLB and PCI buses at the same time, and with the price of generic PCI video cards headed to the $25 range, the VLB slot is often not used.

Fifth Generation (Pentium Class) Intel Chipsets

The Pentium chipset world is a very different one than that of the 486. When it invented the Pentium, and at the same time the PCI bus for it to run on, Intel decided that it would make sense for it to get into the chipset business as well. From a technical standpoint, this makes sense, because in order to assure maximum performance, the coupling of the processor, cache and chipset need to become increasingly tighter and tighter.

Intel had (and still does have) a unique advantage in chipset development, because of its superior knowledge of the processors being used, and at the time, the PCI bus as well. As it continues to add new functionality to its processors, it can add support for them to its chipsets as well, and use its well-known name to continue its dominance in both markets.

Starting in 1997, however, the market seems to have taken an interesting turn. Intel has basically given up development of Pentium-class chipsets, since it has moved its focus exclusively to the Pentium II platform and sixth-generation designs. Alternative chipset manufacturers such as Via Technologies and SiS are poised to pick up the slack in the Pentium chipset field. It will be interesting to see what happens over the next year or so.

Intel 430LX ("Mercury")

The 430LX was Intel's first Pentium chipset (at least to my knowledge). It was used on the original Pentiums, which came in 60 and 66 MHz versions. These were 5V chips and were used on the old socket 4 motherboards; they were obsoleted very quickly when Intel moved to their newer 90 and 100 MHz 3.3V processors running on socket 5 motherboards.

Feature-wise, the 430LX was a very plain chipset by today's standards. It included the PCI bus of course and support for up to 128 MB of RAM. There was no support for EDO memory (which was introduced in the Triton series). Cache chips were soldered directly onto most motherboards, and I believe the size was 64 KB, but I am not sure. The 430LX went the way of the dodo very quickly, as did the motherboards it used and the hot, slow first-generation Pentiums they supported. Today it is extremely difficult to even find any information at all about this chipset!

Intel 430NX ("Neptune")

The 430NX was the original chipset used for Intel's second generation Pentium chips, covering Pentiums with speeds from 90 to 133 MHz (which, of course, also work in later Pentium chipsets). The biggest improvement with this generation of technology was the newer Pentium itself: it ran much faster, cooler and more reliably than the first generation Pentiums did. In addition, the chipset itself introduced the following improvements over the Mercury (430LX) chipset:

• Dual processor support.
• Support for 512 MB of system memory (up from 128 MB for the Mercury).
• Support for up to 512 KB of asynchronous secondary cache.

While this was a good chipset, it didn't really introduce any major, ground-breaking new technologies, and it wasn't really well-publicized. Like the Mercury before it, it faded before the introduction of the Triton line, and about the only place to find information about it today is checking the motherboard manufacturers for spec sheets on their older boards.

Intel's first Triton chipset, the 430FX, "put Intel on the map" in the world of chipsets, so to speak. Although it is now dated, and feature-wise does not compare well to its newer siblings the 430HX, 430VX and 430TX, the FX was at the time a real advance in the chipset world.

The FX's primary contribution technology-wise was support for EDO RAM. The FX chipset started the ball rolling toward the eventual shift in dominance from FPM to EDO memory. It improved upon the 430NX, but also in some ways was a step down from it. (Given this, it is surprising that so many people were surprised when Intel did something similar with the 430TX, but then not many people know much about the 430NX or 430FX).

The advantages of the 430FX over the 430NX are:

• Support for EDO memory.
• Supported for pipelined burst cache and synchronous cache technologies.
• PCI level 2.0 compliance.
• Improved performance, in general.

The disadvantages of the 430FX compared to the 430NX are:

• Less memory support (only 128 MB total).
• No support for dual processing.

The 430FX chipset is still occasionally seen in older motherboards for sale at computer shows or by surplus stores, but it is definitely obsolete. It was replaced by the Triton II twins: the 430HX and the 430VX.

Intel 430HX ("Triton II")

The Triton II HX chipset was created by Intel to address some of the shortcomings of the original Triton. It was intended to be a higher-end chipset aimed at the business or professional market.

The HX chipset's primary advantages over the FX are:

• Support for parity and ECC memory.
• Dual processor support.
• Support for 512 MB of system memory instead of just 128 MB.
• Support for 512 MB of cached system memory instead of just 64 MB (if the optional larger tag RAM is installed.)
• Much improved performance, due to faster memory timing and more I/O buffers.
• PCI level 2.1 compliance.
• USB support.
• Independent device timing for IDE/ATA drives.

This is a full-featured chipset, and has been the most popular Intel Pentium chipset for those who are performance-oriented. Motherboards using it tend to be slightly more expensive than those using the FX or VX chipsets. Its biggest advantages are its faster performance, much larger maximum cacheable RAM, and parity / error correction support.

The 430HX is the only current Intel Pentium-class chipset to offer parity and error corrected memory support. This makes it the only choice (as far as this author is concerned) for mission critical applications, servers, etc., unless you want to use a non-Intel chipset. Unfortunately, the HX chipset has now been discontinued, which means that choices are becoming much more limited.

Note: Support for the full 512 MB of cached memory is only provided if the motherboard provides enough tag RAM; an 11-bit tag RAM is required. Most good motherboards do provide this support.

Warning: Beware the so-called "HX Pro" chipset, which is not the 430HX but a cheaper imitator trying to confuse the buying public by making it sound like an enhanced 430HX chipset.

Intel 430VX ("Triton II", a.k.a. "Triton III")

The 430VX chipset is technically also termed "Triton II" because Intel considers the HX and VX to be a "family" of sorts, two chipsets focusing on different segments of the market. While the HX is the more expensive solution intended for the business or power user, the VX is the cost-effective solution for the family PC or casual user. (The value of making these sorts of generalizations is totally lost on me personally, since my home machine often gets much more of a workout than my office machine does, but that's another topic entirely).

Despite Intel's efforts, the VX is often called the "Triton III", both to distinguish it from the HX, and as a marketing ploy to make it seem like it is "new and improved" or whatnot. This is very misleading, because the VX has in fact only one significant technical advantage over the HX, and is actually inferior in almost every other way to the HX set. In many ways it is more similar to the FX than the HX.

The VX chipset's advantages over the HX:

• Lower cost.
• Support for SDRAM.
• Improved burst memory reads when using SDRAM (though the initial read is slower). The timing with SDRAM is 7-1-1-1 vs. the HX's 5-2-2-2 when using EDO RAM.

The VX chipset's disadvantages compared to HX:

• No parity or ECC memory support.
• No dual processor support.
• Support for only 128 MB of system memory, like the FX.
• Support for only 64 MB of cached system memory, again, like the FX.
• Slower memory timings when using EDO memory (but faster than the FX).
• Fewer PCI I/O buffers (but more than the FX).
• No independent device timing for IDE/ATA devices.
• Fewer supported motherboard SIMM or DIMM slots.

The advantage of SDRAM over EDO RAM in Pentium systems is questionable; while the raw speed of SDRAM is much higher than EDO, only 5 to 10 percent of requests for memory reads actually go to the memory. The vast majority are satisfied by the level 2 cache; the end result is that SDRAM does not lead to nearly as large an increase in performance as its presence might imply. Choosing memory technologies is discussed here. In some ways, the only advantage of buying SDRAM now is in the hopes that it will be able to be used in future systems, since SDRAM seems to be the direction that memory technology is headed to. However, even this is debatable, since it seems that with new technologies such as RAMbus DRAM being pursued by Intel, the future even of SDRAM is not secure. In addition, SDRAM specifications seem to change surprisingly often.

In many ways, the biggest drawback of the VX chipset is the reduced memory cacheability, something that didn't seem like a big deal in 1995 when 64 MB of RAM cost over $2,000. It will seem like a big deal very soon when users, finding EDO memory dropping to $3 a megabyte or less, try putting 96 MB in their system and find that it dramatically slows down.

Warning: Beware the so-called "VX Pro" chipset, which is not the 430VX but a cheaper imitator trying to confuse the buying public by making it sound like an enhanced 430VX chipset.

The 430VX chipset can today be basically considered obsolete, since there is no technical reason to use it over the newer 430TX chipset.

With the 430HX chipset primarily the choice of those seeking high performance, Intel's announcement of a new chipset to be delivered to market in early 1997 had everyone hoping that the 430TX would be the "next step" forward, something to combine the benefits of the 430HX with new capabilities and position itself as the obvious winner for some time to come.

Instead, Intel delivered what seems to be much more a successor to the 430VX than the 430HX, which disappointed many people. It incorporates several new technologies, and improves performance over the VX chipset, but it leaves out several capabilities of the HX. The reason for this, in part, may be Intel's supposed desire to move high-end users from the Pentium family of processors to the Pentium Pro and Pentium II. Having no clear "do it all" chipset for the Pentium may be part of this strategy (it certainly makes sense to me from a business standpoint). One point in favor of this argument is Intel's decision not to support AGP with the 430TX, as had been originally anticipated.

Note: Some people like to call the 430TX "Triton IV" since it is the fourth in the "430" family of chipsets. I am quite sure that Intel has never called it that, so I do not.

The end result of all this is that instead of the 430TX being an obvious best of the Pentium chipsets, an "HX vs. TX" choice was set up. This is much more of a tradeoff than the HX vs. VX choice was (basically, the HX was clearly superior to the VX unless cost or SDRAM support were major concerns.) Here's how the 430TX chipset compares against its most recent predecessors.

The TX chipset's primary advantages over the VX:

• Improved memory timing for initial read from SDRAM; 5-1-1-1 instead of 7-1-1-1.
• Increased maximum memory, from 128 MB to 256 (but cacheable memory remains at only 64 MB).
• Support for more SIMM and DIMM slots on the motherboard than the VX provides.
• Support for DMA mode 3 transfers (a.k.a. DMA-33, Ultra-ATA, or ATA-33), allowing faster transfer rates on high-end drives.
• Independent device timing for IDE/ATA devices.
• Lower power consumption.
• Better performance overall.

The TX chipset's disadvantages compared to the VX:

• None really, except that it will be more expensive because it is newer and better.

So, the TX vs. VX question is pretty much a "no brainer": the TX is better, and the VX is cheaper, and that's the decision, performance vs. cost.

Against the HX we have a bit more of a contest. Here are the TX's main advantages over the HX chipset:

• Support for SDRAM.
• Superior memory timing when using SDRAM, with no trade-off for the initial read: 5-1-1-1 vs. 5-2-2-2 (for HX's EDO) as opposed to the VX's 7-1-1-1.
• Support for DMA mode 3 transfers (a.k.a. DMA-33, Ultra-ATA, or ATA-33), allowing faster transfer rates on high-end drives.
• Lower power consumption.

The TX's disadvantages compared to the HX are:

• No parity or ECC memory support.
• No dual processor support.
• Support for only 256 MB of system memory, instead of 512 MB.
• Support for only 64 MB of cached system memory, still!
• Fewer PCI I/O buffers (but more than the VX).
• Fewer motherboard SIMM/DIMM slots (but again, more than the VX).

This chipset obviously compares much better against the HX than the VX does, although it still has some glaring weaknesses, particularly the low DRAM cacheability, and no error correction support. (I'm in the minority but I do not use unprotected memory and that's that.)

For most of the first year since its introduction, the TX chipset was fairly successful, but really did not replace the HX chipset. The market was pretty much split between the two. Now, the TX is by far the most common new Intel chipset for the Pentium platform. Why? Simple: Intel discontinued the HX chipset. The TX is the last Intel chipset for fifth generation motherboards. The future appears to belong to alternative chipset makers such as VIA Technologies.

Warning: Beware the so-called "TX Pro" chipset, which is not the 430TX but a cheaper imitator trying to confuse the buying public by making it sound like it is superior to the 430TX. It is in fact not made by Intel but by another chipset vendor that doesn't seem to have enough faith in its product to use its own name.

Intel has been by far the most popular choice for Pentium motherboard chipsets; more than 90% of Pentium motherboards in existence use Intel chipsets. The 430FX, 430HX, 430VX and 430TX are similar in name and function. The table below contrasts these chipsets in order to demonstrate their most important features and performance factors. The topic headers are based on the topics in the Chipset Functions and Features section. These chipsets all support the Pentium and socket-compatible processors, and use PCI and ISA I/O buses.

Note: The TX chipset is listed before the HX instead of after it, to permit easier comparison to the HX and VX chipsets.

Note: System timings stated are "optimal" and depend on support from appropriately fast memory. Otherwise you may need to slow down the timing in your BIOS setup.

Fifth Generation (Pentium Class) Non-Intel Chipsets

From 1993 to 1997, Intel was "the" big chipset manufacturer for the Pentium platform. Intel totally dominated the market for fifth-generation chipsets, largely as a result of its large OEM deals with major manufacturers such as Dell Computer. However, there have always been alternatives to Intel available; they just haven't been very commonly seen in the general PC marketplace.

With Intel's decision to leave the fifth generation platform to concentrate on the Pentium II, a vacuum was created in the Pentium-compatible chipset market. The truly staggering numbers of existing Socket 7 fifth-generation motherboards, combined with AMD's and Cyrix's decisions to continue to develop for this platform, means that it is going to be around for a while. Since Intel discontinued the 430HX chipset, the most full-featured Pentium chipset it created, that has left the 430TX as Intel's only offering. Other companies have moved well beyond the TX, creating multiple new chipsets that surpass the 430TX in features and performance.

In 1998, three primary alternative chipset vendors have emerged. The best-known of the three is Via Technologies, which has been the biggest thorn in Intel's side for some time now. Via chipsets have always competed very well against Intel's technologically, but have suffered at the hands of Intel's established reputation and marketing muscle. The other two are Silicon Integrated Systems (SiS) http://www.sis.com.tw/  and Acer Labs Inc. (ALi) http://www.ali.com.tw/. These companies have in the past largely been crowded out of the market by Intel and are now poised to make a significant comeback--if Socket 7 remains a successful platform.

As a result of these three companies all trying to outdo each other, new non-Intel Pentium-class chipsets have been coming out at an astounding rate, sometimes a new one every few weeks.

Sixth Generation (Pentium Pro / Pentium II Class) Chipsets

If Intel dominated the Pentium chipset world, it does so even more for Pentium Pro / Pentium II chipsets. Intel built upon its success in the Pentium market, and given its superior knowledge of the Pentium Pro, was able to deliver the chip and the chipset at the same time. The Pentium Pro has been geared from the start to be a chip to be focused (in part) on servers and multiprocessing, and Intel has patented the type of multiprocessing (SMP) that the Pentium Pro does.

Now that Intel has decided to go to the Slot 1 interface for the Pentium II, while AMD and Cyrix are sticking with Socket 7 on Pentium-class boards, chances are high that Intel will continue to dominate sixth generation chipsets for some time to come. Virtually all Pentium Pro / Pentium II motherboards use Intel chipsets; their market share here is even higher than it is in the Pentium world (and that's saying something).

Intel 450GX/KX ("Orion")

Intel's first Pentium Pro chipset was actually a pair of chipsets, both to my knowledge codenamed "Orion". These are high-performance chipsets that are also priced rather high. The 450GX is the so-called "server" version (even though the 450KX is often more than enough for many servers!). It supports unique features not found on other chipsets, including support for 8GB of 4-way interleaved memory, quad Pentium Pro processors, and two separate PCI buses! The 450KX is the "workstation" version of Orion and supports dual processors, and "only" up to 1GB of 2-way interleaved memory. Both chipsets support parity/ECC memory.

Note: There were problems with the earliest steppings of the Orion chipset; older machines using the 450GX/KX should be checked carefully for their revision numbers. These bugs caused some of the chipset features to have to be turned off to avoid problems, which had an impact on performance.

Due to their high cost, you are quite unlikely to see these chipsets used in PCs. They are, however, the de facto standard in the server world due to their advanced features and high performance. For PCs the 440FX "Natoma" chipset became the standard pretty much as soon as it was introduced.

Intel 440FX ("Natoma")

The vast majority of current Pentium Pro motherboards use the 440FX chipset, which was offered as more of a "mainstream" replacement for the 450KX Orion. Feature-wise it is similar to the Pentium 430HX "Triton II" chipset, offering improved performance over the Orion at a lower cost. It does not, however, match the special capabilities of the 450GX chipset, making the latter still preferred for very-high-end applications.

Intel has really taken its time in coming up with a replacement for the 440FX chipset, which is starting to get a bit long in the tooth. (I guess having no viable competition will let that sort of thing happen...) Still, it is an adequate performer, and it formed the basis for the first Pentium II motherboards. The biggest weakness of the 440FX is that it does not incorporate the latest technologies--the 440FX has been around a while! This means in particular no support for Ultra DMA or SDRAM memory (which isn't really as big of a disadvantage as many people seem to think.) With the newer, Pentium-II-specific 440LX chipset now mainstream, few new Pentium II machines are being produced using the 440FX.

Intel 440LX

The 440LX is Intel's latest and current top-end chipset, designed specifically for their latest and current top-end processor, the Pentium II. The Pentium II was released several months before the 440LX was ready, and so early PII motherboards used the older 440FX chipset. This chipset was never designed with the Pentium II in mind of course, and the 440LX was, optimized specifically to take advantage of its architecture.

Compared to the 440FX, the 440LX chipset offers several improvements. The key ones are:

• It is the first Intel chipset supposedly to support the new Advanced Graphics Port (AGP) standard for increasing video performance, particularly 3D video.
• Support for SDRAM memory, the first time this memory has been supported for the Pentium Pro or Pentium II platform.
• Support for the Ultra DMA hard disk interface standard.
• Support for the universal serial bus (USB) interface.
• Improved performance, especially for the Pentium II.

Of course, the 440LX is used in the latest and best Pentium II motherboards, and supports the SEC packaging used by that processor.