Intel and AMD set up a single manufacturing technology that produces a class of products, and then they sell them at different prices for different speeds. Sometimes the “slow” product can actually run faster than its label indicates.

The mainboard generates a clock signal to the CPU measured in Megahertz (MHz, a million clock ticks per second). Internally the CPU needs a faster clock, so it applies a “multiplier”. If the multiplier is 10x, then the internal CPU clock ticks 10 time faster than the mainboard clock. A 200 MHz mainboard clock and a multiplier of 10 produce 2 GHz internal performance.  Every tenth internal clock tick is then synchronized with the mainboard tick so they do not drift apart.

An Intel CPU chip transfers 4 units of data to or from memory with every mainboard clock tick. An AMD CPU chip transfers 2 units of data per tick. Instead of quoting the mainboard clock speed that a chip is designed to us, Intel specification sheets often talk about the Front Side Bus (FSB) or memory speed, which is 4 times faster. Thus the 200 MHz mainboard clock would be called an 800 GHz FSB.

The 800 GHz FSB was introduced with the Pentium 4 and lasted for years. Then the generation of Core 2 Duo chips that came out in 2007 bumped the mainboard clock to 266 (FSB 1066) MHz, and the generation that will be widely used in 2008 bumped it again to 333 (FSB 1333) MHz. A few server chips run with a 400 (FSB 1600) MHz.

A typical mainboard is designed to support new and old CPU chips. If you just plug the chip into the socket, the BIOS on the mainboard will sense the type of chip and determine a default clock speed. However, the mainboard may support faster clock speeds (for other CPU models) and the user can manually override this default.

A CPU chip has millions of separate circuits that each perform a specific function. The clock tick tells all the circuits together when they can begin to perform the next operation. The clock signal should come after the very slowest circuit on the entire chip has completed its work. If you feed the CPU a faster clock, at some point the slowest circuit will not have completed its work when the tick occurs and then the CPU will make mistakes.

Consider a circuit like a drinking glass. If the glass is full of water, it is a 1. If empty, then it is a 0. You change its state by pouring water into the glass or pouring it out. Intel will rate its chip so that every circuit is 100%  or 0% full or very close to it. However, the circuit itself has some flexibility and can probably deliver precise results if a 0 is 0-25% full and 1 if the circuit is 75-100% full. It takes time to fill or empty the glass, and if you speed up the clock then the period will end before you complete the operation but as long as you are in the safe range there is no problem. Speed the clock up enough and you start to get measurements when the glass is either 40% or 60% full and now there is a chance for error. Intel will design server chips so they can never make an error, and desktop chips run at their rated speed might go decades without an error.

Most of the cost of a CPU chip is in the research, design, and manufacturing setup. The actual incremental cost of one chip is small. So to meet the demand for low cost, slower speed chips it is often simpler for Intel to artificially slow down a current generation chip than to design an inherently slower chip. While the 65 nanometer generation of chips sold throughout 2007 were designed for a 266 (FSB 1066) MHz clock, the least expensive chips in the family were sold to run at the older 200 MHz speed. When they first came out, sophisticated users could tell the mainboard BIOS to clock these cheap chips at the same 266 MHz clock speed as the rest of the family, and a chip that should nominally run at 2.4 GHz runs instead at 3 GHz. Today the 45nm chips designed to run with a 333 MHz clock can often be driven by a 400 MHz clock on a new mainboard.

After a small amout of overclocking that you may get “free”, you start getting to the “glass is half empty or half full” problem. If you run the clock at a rate where errors happen every two days instead of once a century, is that really a problem? If you are doing your income taxes, it is a problem (which is why Turbo Tax checks and rechecks every calculation to make sure the random error on an overclocked machine doesn’t cause a problem). If you are saving civilization from an army of zombies, then any random error will probably appear as a glitch on the screen that disappears so quickly you won’t notice it. You can get a little faster performance by increasing the voltage level (equivalent in the analogy to pouring the water faster) but that also generates more heat.

Before a new chip family comes out, Intel leaves a lot of slop factor in the design to make sure nothing unexpectedly goes wrong. After the chips have been available for 6 months, they may come out with a new revision that has been tweaked to defeat overclocking and force the CPU to run closer to its rated speed.

There is a paradox. The most successful overclocking occurs at the low end of the product line, where you can almost trivially make a $125 chip perform at a speed that Intel associates with a $180 chip. However, the sophisticated or enthusiast user doesn’t typically buy the very least expensive chips. As you go up in price and speed, you come closer to the actual design limits of the current generation of chips. Here overclocking becomes more difficult. You may have to increase the core voltage, and you may need a very powerful cooling tower to handle the extra heat. However, the teenager who is trying to get the very fastest gaming CPU is not the target audience of these articles. He will get his advice from other sources on the Web.

Bumping up the clock speed on an entry level CPU chip is one of the simplest things you can do. You don’t even have to open the covers. Simply press Del when you power on the system, go to the right BIOS page, and type in a new clock speed. Some vendors like Dell have a customized BIOS that will not allow this, but it will work on most mainboards.

However, if you don’t use your computer for gaming, then it probably doesn’t matter. There are very few things you can do that will tax even the slowest modern CPU. If you put a Blu-Ray disk on your computer, and you want to watch High Definition movies on your computer screen, and you don’t have or can’t use the decoding capability of your video card, then your 2.0 GHz Core Duo CPU chip will run 100% busy and the movie will stutter. Overclock it just a bit to 2.4  GHz and the playback will be smooth. It is not a lot of work and certainly cheaper than buying a new machine, or even a faster CPU chip.