CPU Packages

8088 CPU Family

One of the many features that make PCs attractive is the ability for users to replace one CPU with another. If you want a removable CPU, you need your CPUs to use a standardized package with matching standardized socket on the motherboard. CPUs have gone through many packages, with manufacturers changing designs like snakes shedding skins. The fragile little DIP package of the 8088 gave way to rugged slotted processors in the late 1990s, which have in turn given way to CPUs using the now prevalent grid array packaging.

The grid array package has been popular since the mid-1980s. The most common form of grid array is the pin grid array (PGA). PGA CPUs are distinguished by their square shape with many—usually hundreds—of tiny pins

Collectively, Intel and AMD have used close to 100 variations of the PGA package over the years for hundreds of different CPU models with names like staggered-PGA, micro-PGA, ball grid array (which uses tiny balls instead of pins) and land grid array (which uses flat pads instead of pins). There are also many different varieties of PGA CPUs based on the number of pins sticking out of the CPU. These CPUs snap into special sockets on the motherboard, with each socket designed to match the pins (or balls or pads) on the CPU. To make CPU insertion and removal easier, these sockets—officially called zero insertion force (ZIF) sockets—use a small arm on the side of the socket  or a cage that fits over the socket to hold the CPU in place. ZIF sockets are universal and easily identified by their squarish shape.

The first generations of sockets used a numbering system that started with Socket 1 and went through Socket 8. The hassle of trying to remember how many pins went with each type of socket made it clear after a while that the CPU makers might as well give all sockets a name based on the number of pins. Most sockets today have names like Socket 940 and Socket 775 to reflect the number of pins.

The Pentium CPU: The Early Years

Since the advent of the 8088 way back in the late 1970s, CPU makers have added a large number of improvements. As technology has progressed from the 8088 to the most current CPUs, the sizes of the external data bus, address bus, and registers have grown dramatically. In addition, the clock speeds at which CPUs run have kept pace, getting faster and faster with each successive generation of processor. The 1980s were an exciting time for CPU technology. The 8088 CPU was supplanted by a series of improved processors with names such as 80286, 80386, and 80486 .Each of these CPU families incorporated wider buses, increasingly higher clock speeds, and other improvements.

Original Pentium

The Pentium is not a new chip; it’s been around since 1990, and the last versions of the Pentium chip were discontinued in 1995. The original Pentium was the springboard for the Pentium Pro, however, probably the most important CPU ever made, and thus it makes sense to start there. In the rest of this chapter, we look at all the popular CPUs developed since the Pentium and see how they’ve built on this legacy CPU

Early Intel Pentiums

• External speed range: 50–66 MHz

• Internal speed range: 60–200 MHz

• Multiplier range: 1×–3×

• L1 cache: 16 KB

• Package: PGA

• Socket(s) used: Socket 4, Socket 5

AMD made a competitor to the Pentium called the AMD K5. The AMD K5 was pin-compatible with the Pentium, but to keep Intel from suing them, AMD made the K5 very different on the inside, using a totally different method of processing. The AMD K5 had some success, but was rather quickly upstaged by better AMD CPUs.

AMD Pentium Equivalents

• External speed range: 50–75 MHz

• Internal speed range: 60–150 MHz

• Multiplier range: 1.5×–2×

• L1 cache: 16 KB

• Package: PGA

• Socket(s) used: Socket 7

Pentium Pro

In 1995, Intel released the next generation of CPU, the Pentium Pro, often called the P6. The Pentium Pro was a huge CPU with a distinctive, rectangular PGA package. The P6 had the same bus and register sizes as the Pentium, but three new items made the P6 more powerful than its predecessor: quad pipelining, dynamic processing, and an on-chip L2 cache. These features carried oninto every CPU version that followed, so many people consider the Pentium Pro to be the true “Father of the Modern CPU.”

Intel Pentium Pro

• External speed range: 60–66 MHz

• Internal speed range: 166–200 MHz

• Multiplier range: 2.5×–3×

• L1 cache: 16 KB

• L2 cache: 256 KB, 512 KB, 1 MB

• Package: PGA

• Socket(s) used: Socket 8

Later Pentium-Class CPUs

Intel’s usual game plan in the rough-and-tumble business of chip making is to introduce a new CPU and simultaneously declare all previous CPUs obsolete. That did not happen with the Pentium Pro, however, because Intel never really developed the P6 for most users. It was to be the CPU for powerful, higher-end systems. This kept the Pentium as the CPU of choice for all but the most power-hungry systems. While the Pentium Pro languished on the high end for several years, Intel and AMD

developed new Pentium-class CPUs that incorporated a series of powerful improvements, some of which were taken from the Pentium Pro. These improvements required that they be regarded as a new family of CPUs, which I call the “later Pentium-class CPUs”. Although certainly some profound differences exist between these CPUs, they all have three groups of similar improvements: multimedia extensions (MMX), increased multipliers/clocks, and improved processing.

MMX

In 1996, Intel added a new enhancement to its Pentium CPU, called multimedia extensions (MMX), in response to the large number of programs with heavy graphic needs coming out at this time. MMX was designed to work with large graphics by calculating on large chunks of data and performing vector math (vector math is needed to handle graphical issues such as spinning a 3D object). MMX was not heavily supported by the folks who wrote graphics programs, but MMX did start the idea that CPUs should have special circuitry just for such programs. Over time, the graphics community began to work with Intel to improve MMX, eventually replacing it with better solutions.

Pentium II

Intel’s next major CPU was the Pentium II. Although highly touted as the next generation of CPU, the Pentium II was little more than a faster Pentium Pro with MMX and a refined instruction set. The Pentium II came in a distinctive single edge cartridge (SEC) that gave more space for the L2 cache and made CPU cooling easier while freeing up more room on the motherboard Aggressive advertising and pricing made the Pentium II extremely popular.

Pentium III

The Pentium III improved on the Pentium II by incorporating Streaming SIMD Extensions (SSE), Intel’s direct competitor to AMD’s 3DNow!; a number of internal processing/ pipelining improvements; full support for 100-MHz and 133-MHz motherboard speeds; and high-speed L2 cache. The Pentium III was first produced using an SEC package, but improvements in die technology enabled Intel to produce PGA versions later, ending the short reign of the SEC-package CPUs

Intel Pentium III CPUs

• External speed range: 100–133 MHz

• Internal speed range: 450 MHz–1.26 GHz

• Multiplier range: 4×–10×

• L1 cache: 32 KB

• L2 cache: 256 KB or 512 KB

• Package: SEC-2, PGA

• Socket(s) used: Slot 1, Socket 370

Just as the Pentium II had a Celeron, so did the Pentium III. Unfortunately, Intel makes no differentiation between classes of Celerons, making buying a challenge unless you ask. The Pentium III–based Celerons were PGA and used Socket 370

AMD Athlon Thunderbird

AMD’s first major improvement to the Athlon CPU was codenamed Thunderbird . The Thunderbird Athlon marked AMD’s return to a PGA package with the adoption of the proprietary 462-pin socket called Socket A. The change between the Classic and the Thunderbird wasn’t just cosmetic. Thunderbird had an interesting double-pumped frontside bus that doubled the data rate without increasing the clock speed. Athlon Thunderbird CPUs have a smaller but far more powerful L2 cache, as well as a number of other minor improvements.

AMD Thunderbird Athlon CPUs

• Process: 180 nm

• Watts: 38–75

• External speed range: 100–133 MHz (Double-pumped)

• Internal speed range: 650 MHz–1.4 GHz

• Multiplier range: 6.5×–14×

• L1 cache: 128 KB

• L2 cache: 256 KB

• Package: PGA

• Socket(s) used: Socket A

AMD Duron

Duron is the generic name given to all lower-end CPUs based on the Athlon processor. Basically an Athlon with a smaller cache, the Duron supported the same 200-MHz frontside bus as the Athlon, giving it a slight edge over the Celeron. The Duron connected to the same 462-pin Socket A as the later Athlon CPUs.

Intel Pentium 4 Willamette

While the Pentium II and III were little more than improvements on the Intel Pentium Pro, the Pentium 4 introduced a completely redesigned core, called NetBurst. NetBurst centered around a totally new 20-stage pipeline combined with other features to support this huge pipeline. Each stage of the pipeline performed fewer operations than typical pipeline stages in earlier processors, which enabled Intel to crank up the clock speed for the Pentium 4 CPUs. The first Pentium 4s included a new version of SSE called SSE2, and later versions introduced SSE3. The Pentium 4 achieved a 400-MHz frontside bus speed—twice the Athlon’s 200 MHz—by using four data transfers per clock cycle on a 100-MHz bus. Intel used this same quad-pumped frontside bus technology on a 133-MHz bus to achieve a 533-MHz

frontside bus. There were two packages of early Pentium 4 CPUs. The first Pentium 4 CPUs came in a 423-pin PGA package and had a 256-KB L2 cache. These were replaced by the 512- KB L2 cache Pentium 4 with a 478-pin PGA package. Even though the new package has more pins, it is considerably smaller than the earlier package.

Pentium 4 Extreme Edition

The Pentium 4 Extreme Edition was designed to place Intel at the top of the CPU performance curve. The Extreme Edition CPUs used a Socket 478 or LGA 775 package, making them identical to other Pentium 4s, but packed some powerful features. Most interesting was the 2-MB L3 cache—the only non-server CPU to carry an L3 cache. The Pentium 4 Extreme Edition also had some of the highest wattages ever recorded on any Intel desktop CPU—over 110 watts! Extreme Edition CPUs ran incredibly fast, but their high price kept them from making any significant impact on the market.

• Process: 130 and 90 nm

• Watts: 85–115

• External speed range: 200 MHz (Quad-pumped), 266 MHz (Quad-pumped)

• Internal speed range: 3.2 GHz–3.7 GHz

• Multiplier range: 14×–17×

• L1 cache: 128 KB

• L2 cache: 512 KB

• L3 cache 2 MB

• Package: 478-pin PGA, 775-pin LGA

• Socket(s) used: Socket 478, Socket LGA-775

Intel Xeon Processors

Just as the term Celeron describes a series of lower-end processors built around the Pentium II, Pentium III, and Pentium 4, the term Xeon (pronounced “Zee-on”) defines a series of high-end processors built around the Pentium II, Pentium III, and Pentium 4. Xeon CPUs built on the Pentium II and III core processors via the addition of massive L2 caches, but their strength comes from strong multiprocessor support. Both the Pentium II Xeon and the Pentium III Xeon used a unique SEC package that snapped into a Xeon-only slot called Slot 2. In general, people buy Xeons because they

want to run a system with more than one processor. Most modern CPUs can run with one other identical CPU, but putting together two CPUs that were never designed to work together requires an incredibly complex MCC. Xeon processors, on the other hand, are carefully designed to work together in sets of two, four, or even eight CPUs. Although very expensive, their immense power lets them enjoy broad popularity in the high-horsepower world of server systems.

The Pentium 4 Xeon is quite a different beast from the previous Xeon types. First, the Pentium 4 Xeon’s caches are smaller than other Xeons; advancements in pipelining make anything larger less valuable. Second, Intel sells two lines of Pentium 4 Xeons. One line, simply called the Pentium 4 Xeon, is for single or dual processor systems; and the second line, called the Pentium 4 Xeon MP, is for four or eight multiprocessor systems. Last, Intel went back to the PGA package with the Pentium 4 Xeons, a Xeon-only 603-pin package.

Dual-Core CPUs

CPU clock speeds hit a practical limit of roughly 4 GHz around 2002–2003, motivating the CPU makers to find new ways to get more processing power for CPUs. Although Intel and AMD had different opinions about 64-bit CPUs, both decided at virtually the same time to combine two CPUs into a single chip, creating a dual-core architecture. Dual-core isn’t just two CPUs on the same chip. A dual-core CPU has two execution units—two sets of pipelines—but the two sets of pipelines share caches (how they share caches differs between Intel and AMD) and RAM.

Pentium D

Intel won the race for first dual-core processor with the Pentium D line of processors (Figure 3-61). The Pentium D is simply two late-generation Pentium 4s molded onto the same chip with each CPU using its own cache—although they do share the same frontside bus. One very interesting aspect to the Pentium D is the licensing of AMD’s AMD64 extensions—the “smarts” inside AMD CPUs that enables AMD CPUs to run either 64- or 32-bit code. Intel named their version EM64T. Even though the Pentium D is technically a 32-bit processor, it has extra address wires and 64-bit registers to accommodate 64-bit code. There are two codenames for Pentium D processors: the “Smithfield” (model numbers 8xx), using a 90-nm process, and the “Presler” (model numbers 9xx), using a 65-nm process. Pentium Ds use the same LGA 775 package as seen on the later Pentium 4s.

Intel Pentium D

• Process: 90 nm and 65nm

• Watts: 95–130

• External speed range: 166 MHz, 200 MHz (Quad-pumped)

• Internal speed range: 2.6 GHz–3.6 GHz

• Multiplier range: 14×–20×

• L1 cache: Two 128-KB caches

• L2 cache: Two 1-MB caches or two 2-MB caches

• Package: 775-pin LGA

• Socket(s) used: Socket LGA-775