Computer Processor Characteristics
Here are the important characteristics of processors:
Processor make and model
The primary defining characteristic of a processor is its make AMD or Intel and its model. Although competing models from the two companies have similar features and performance, you cannot install an AMD processor in an Intel-compatible motherboard or vice versa.
Another defining characteristic of a processor is the socket that it is designed to fit. If you are replacing the processor in a Socket 478 motherboard, for example, you must choose a replacement processor that is designed to fit that socket. Table 5-1 describes upgradability issues by processor socket.
Table 5-1: Upgradability by processor socket type
The clock speed of a processor, which is specified in megahertz (MHz) or gigahertz (GHz), determines its performance, but clock speeds are meaningless across processor lines. For example, a 3.2 GHz Prescott-core Pentium 4 is about 6.7% faster than a 3.0 GHz Prescott-core Pentium 4, as the relative clock speeds would suggest. However, a 3.0 GHz Celeron processor is slower than a 2.8 GHz Pentium 4, primarily because the Celeron has a smaller L2 cache and uses a slower host-bus speed. Similarly, when the Pentium 4 was introduced at 1.3 GHz, its performance was actually lower than that of the 1 GHz Pentium III processor that it was intended to replace. That was true because the Pentium 4 architecture is less efficient clock-for-clock than the earlier Pentium III architecture.
Clock speed is useless for comparing AMD and Intel processors. AMD processors run at much lower clock speeds than Intel processors, but do about 50% more work per clock tick. Broadly speaking, an AMD Athlon 64 running at 2.0 GHz has about the same overall performance as an Intel Pentium 4 running at 3.0 GHz.
'''MODEL NUMBERS VERSUS CLOCK SPEEDS''' Because AMD is always at a clock speed disadvantage versus Intel, AMD uses model numbers rather than clock speeds to designate their processors. For example, an AMD Athlon 64 processor that runs at 2.0 GHz may have the model number 3000+, which indicates that the processor has roughly the same performance as a 3.0 GHz Intel model. (AMD fiercely denies that their model numbers are intended to be compared to Intel clock speeds, but knowledgeable observers ignore those denials.) Intel formerly used letter designations to differentiate between processors running at the same speed, but with a different host-bus speed, core, or other characteristics. For example, 2.8 GHz Northwood-core Pentium 4 processors were made in three variants: the Pentium 4/2.8 used a 400 MHz FSB, the Pentium 4/2.8B the 533 MHz FSB, and the Pentium 4/2.8C the 800 MHz FSB. When Intel introduced a 2.8 GHz Pentium 4 based on their new Prescott-core, they designated it the Pentium 4/2.8E. Interestingly, Intel has also abandoned clock speed as a designator. With the exception of a few older models, all Intel processors are now designated by model number as well. Unlike AMD, whose model numbers retain a vestigial hint at clock speed, Intel model numbers are completely dissociated from clock speeds. For example, the Pentium 4 540 designates a particular processor model that happens to run at 3.2 GHz. The models of that processor that run at 3.4, 3.6, and 3.8 GHz are designated 550, 560, and 570 respectively.
The host-bus speed, also called the front-side bus speed, FSB speed, or simply FSB, specifies the data transfer rate between the processor and the chipset. A faster host-bus speed contributes to higher processor performance, even for processors running at the same clock speed. AMD and Intel implement the path between memory and cache differently, but essentially FSB is a number that reflects the maximum possible quantity of data block transfers per second. Given an actual host-bus clock rate of 100 MHz, if data can be transferred four times per clock cycle (thus "quad-pumped"), the effective FSB speed is 400 MHz.
For example, Intel has produced Pentium 4 processors that use host-bus speeds of 400, 533, 800, or 1066 MHz. A 2.8 GHz Pentium 4 with a host-bus speed of 800 MHz is marginally faster than a Pentium 4/2.8 with a 533 MHz host-bus speed, which in turn is marginally faster than a Pentium 4/2.8 with a 400 MHz host-bus speed. One measure that Intel uses to differentiate their lower-priced Celeron processors is a reduced host-bus speed relative to current Pentium 4 models. Celeron models use 400 MHz and 533 MHz host-bus speeds.
All Socket 754 and Socket 939 AMD processors use an 800 MHz host-bus speed. (Actually, like Intel, AMD runs the host bus at 200 MHz, but quad-pumps it to an effective 800 MHz.) Socket A Sempron processors use a 166 MHz host bus, double-pumped to an effective 333 MHz host-bus speed.
Processors use two types of cache memory to improve performance by buffering transfers between the processor and relatively slow main memory. The size of Layer 1 cache (L1 cache, also called Level 1 cache), is a feature of the processor architecture that cannot be changed without redesigning the processor. Layer 2 cache (Level 2 cache or L2 cache), though, is external to the processor core, which means that processor makers can produce the same processor with different L2 cache sizes. For example, various models of Pentium 4 processors are available with 512 KB, 1 MB, or 2 MB of L2 cache, and various AMD Sempron models are available with 128 KB, 256 KB, or 512 KB of L2 cache.
For some applications particularly those that operate on small data sets a larger L2 cache noticeably increases processor performance, particularly for Intel models. (AMD processors have a built-in memory controller, which to some extent masks the benefits of a larger L2 cache.) For applications that operate on large data sets, a larger L2 cache provides only marginal benefit.
'''Prescott, the Sad Exception''' It came as a shock to everyone not the least, Intel to learn when it migrated its Pentium 4 processors from the older 130 nm Northwood core to the newer 90 nm Prescott-core that power consumption and heat production skyrocketed. This occurred because Prescott was not a simple die shrink of Northwood. Instead, Intel completely redesigned the Northwood core, adding features such as SSE3 and making huge changes to the basic architecture. (At the time, we thought those changes were sufficient to merit naming the Prescott-core processor Pentium 5, which Intel did not.) Unfortunately, those dramatic changes in architecture resulted in equally dramatic increases in power consumption and heat production, overwhelming the benefit expected from the reduction in process size.
Process size, also called fab(rication) size, is specified in nanometers (nm), and defines the size of the smallest individual elements on a processor die. AMD and Intel continually attempt to reduce process size (called a die shrink) to get more processors from each silicon wafer, thereby reducing their costs to produce each processor. Pentium II and early Athlon processors used a 350 or 250 nm process. Pentium III and some Athlon processors used a 180 nm process. Recent AMD and Intel processors use a 130 or 90 nm process, and forthcoming processors will use a 65 nm process.
Process size matters because, all other things being equal, a processor that uses a smaller process size can run faster, use lower voltage, consume less power, and produce less heat. Processors available at any given time often use different fab sizes. For example, at one time Intel sold Pentium 4 processors that used the 180, 130, and 90 nm process sizes, and AMD has simultaneously sold Athlon processors that used the 250, 180, and 130 nm fab sizes. When you choose an upgrade processor, give preference to a processor with a smaller fab size.
Different processor models support different feature sets, some of which may be important to you and others of no concern. Here are five potentially important features that are available with some, but not all, current processors. All of these features are supported by recent versions of Windows and Linux:
SSE3 (Streaming Single-Instruction-Multiple-Data (SIMD) Extensions 3), developed by Intel and now available on most Intel processors and some AMD processors, is an extended instruction set designed to expedite processing of certain types of data commonly encountered in video processing and other multimedia applications. An application that supports SSE3 can run from 10% or 15% to 100% faster on a processor that also supports SSE3 than on one that does not.
Until recently, PC processors all operated with 32-bit internal data paths. In 2004, AMD introduced 64-bit support with their Athlon 64 processors. Officially, AMD calls this feature x86-64, but most people call it AMD64. Critically, AMD64 processors are backward-compatible with 32-bit software, and run that software as efficiently as they run 64-bit software. Intel, who had been championing their own 64-bit architecture, which had only limited 32-bit compatibility, was forced to introduce its own version of x86-64, which it calls EM64T (Extended Memory 64-bit Technology). For now, 64-bit support is unimportant for most people. Microsoft offers a 64-bit version of Windows XP, and most Linux distributions support 64-bit processors, but until 64-bit applications become more common there is little real-world benefit to running a 64-bit processor on a desktop computer. That may change when Microsoft (finally) ships Windows Vista, which will take advantage of 64-bit support, and is likely to spawn many 64-bit applications.
With the Athlon 64, AMD introduced the NX (No eXecute) technology, and Intel soon followed with its XDB (eXecute Disable Bit) technology. NX and XDB serve the same purpose, allowing the processor to determine which memory address ranges are executable and which are non-executable. If code, such as a buffer-over-run exploit, attempts to run in non-executable memory space, the processor returns an error to the operating system. NX and XDB have great potential to reduce the damage caused by viruses, worms, Trojans, and similar exploits, but require an operating system that supports protected execution, such as Windows XP with Service Pack 2.
Power reduction technology
AMD and Intel both offer power reduction technology in some of their processor models. In both cases, technology used in mobile processors has been migrated to desktop processors, whose power consumption and heat production has become problematic. Essentially, these technologies work by reducing the processor speed (and thereby power consumption and heat production) when the processor is idle or lightly loaded. Intel refers to their power reduction technology as EIST (Enhanced Intel Speedstep Technology). The AMD version is called Cool'n'Quiet. Either can make minor but useful reductions in power consumption, heat production, and system noise level.
By 2005, AMD and Intel were both reaching the practical limits of what was possible with a single processor core. The obvious solution was to put two processor cores in one processor package. Again, AMD led the way with its elegant Athlon 64 X2 series processors, which feature two tightly integrated Athlon 64 cores on one chip. Once again forced to play catch-up, Intel gritted its teeth and slapped together a dual-core processor that it calls Pentium D. The engineered AMD solution has several benefits, including high performance and compatibility with nearly any older Socket 939 motherboard. The slapdash Intel solution, which basically amounted to sticking two Pentium 4 cores on one chip without integrating them, resulted in two compromises. First, Intel dual-core processors are not backward-compatible with earlier motherboards, and so require a new chipset and a new series of motherboards. Second, because Intel more or less simply glued two of their existing cores onto one processor package, power consumption and heat production are extremely high, which means that Intel had to reduce the clock speed of Pentium D processors relative to the fastest single-core Pentium 4 models.
All of that said, the Athlon 64 X2 is by no means a hands-down winner, because Intel was smart enough to price the Pentium D attractively. The least expensive Athlon X2 processors sell for more than twice as much as the least expensive Pentium D processors. Although prices will undoubtedly fall, we don't expect the pricing differential to change much. Intel has production capacity to spare, while AMD is quite limited in its ability to make processors, so it's likely that AMD dual-core processors will be premium priced for the foreseeable future. Unfortunately, that means that dual-core processors are not a reasonable upgrade option for most people. Intel dual-core processors are reasonably priced but require a motherboard replacement. AMD dual-core processors can use an existing Socket 939 motherboard, but the processors themselves are too expensive to be viable candidates for most upgraders.
'''HYPER-THREADING VERSUS DUAL CORE''' Some Intel processors support ''Hyper-Threading Technology (HTT)'', which allows those processors to execute two program threads simultaneously. Programs that are designed to use HTT may run 10% to 30% faster on an HTT-enabled processor than on a similar non-HTT model. (It's also true that some programs run slower with HTT enabled than with it disabled.) Don't confuse HTT with dual core. An HTT processor has one core that can sometimes run multiple threads; a dual-core processor has two cores, which can always run multiple threads.
Core names and core steppings
The processor core defines the basic processor architecture. A processor sold under a particular name may use any of several cores. For example, the first Intel Pentium 4 processors used the Willamette core. Later Pentium 4 variants have used the Northwood core, Prescott-core, Gallatin core, Prestonia core, and Prescott 2M core. Similarly, various Athlon 64 models have been produced using the Clawhammer core, Sledgehammer core, Newcastle core, Winchester core, Venice core, San Diego core, Manchester core, and Toledo core.
Using a core name is a convenient shorthand way to specify numerous processor characteristics briefly. For example, the Clawhammer core uses the 130 nm process, a 1,024 KB L2 cache, and supports the NX and X86-64 features, but not SSE3 or dual-core operation. Conversely, the Manchester core uses the 90 nm process, a 512 KB L2 cache, and supports the SSE3, X86-64, NX, and dual-core features.
You can think of the processor core name as being similar to a major version number of a software program. Just as software companies frequently release minor updates without changing the major version number, AMD and Intel frequently make minor updates to their cores without changing the core name. These minor changes are called core steppings. It's important to understand the basics of core names, because the core a processor uses may determine its backward compatibility with your motherboard. Steppings are usually less significant, although they're also worth paying attention to. For example, a particular core may be available in B2 and C0 steppings. The later C0 stepping may have bug fixes, run cooler, or provide other benefits relative to the earlier stepping. Core stepping is also critical if you install a second processor on a dual-processor motherboard. (That is, a motherboard with two processor sockets, as opposed to a dual-core processor on a single-socket motherboard.) Never, ever mix cores or steppings on a dual processor motherboard that way lies madness (or perhaps just disaster).