Just imagine if the only difference between the two liter sports saloon and the 1.4 family hatchback was the badge on the back. Imagine a car manufacturer that fitted all its cars with the same engine but detuned those in the budget models to protect the price premiums at the top of the range. Imagine if the red line limit on the rev counter was so conservatively marked that you could comfortably exceed it by 50 per cent without damaging the engine.



It sounds implausible, but read ‘Intel' for ‘car manufacturer' and we've pretty much described the state of the processor industry for the last decade. For years now, power users have known how to squeeze more performance from their processors for free, by pushing them to their true limits instead of those artificially set by the chip makers. This black art is known as overclocking and if you ever wondered what it is all about, this may be your last chance to find out before Intel plugs the loophole forever.

The pace of all processors depends on their clock speed – the higher the clock speed, the faster your programs run. In the days of the 486, this clock speed was the same as the bus speed, and pioneering overclockers could only increase the processor speed by increasing the speed of the entire bus. The development of the 486DX2 and DX4, however, saw the introduction of the concept of a clock multiplier and with it a whole new way to overclock.

These chips ran on a fixed multiplier of X2 or X3 for the DX2 and DX4 respectively but when the Pentium was developed, CPU's could suddenly be configured to use any multipliers from x1.5 to x3 simply by changing a few jumpers on the motherboard. Thus a P150 ran on a 50MHz bus with a multiplier of x3 but a P166 used a 66MHz bus and a 2.5x multiplier.

Silicon Twins

The important thing to realize is that both systems use essentially the same piece of silicon in the processor. When the chips come off the assembly line, Intel does not know in advance what the speed rating of each one will be because the difference is not architectural. Production batches are graded for quality and a rating that is thought to be safely within the tolerances of that batch.

Intel's own grading criteria are probably the most stringent in the industry but there is still a degree of leeway that overclockers can exploit, but there are two other factors that increase this safety margin still further.

The first is that during the life of a particular chip family, manufacturing processes often improve. For example, the resolution of the etching process used to define the circuit lines of the processor may increase. This allows a smaller die size, which dissipates less heat and can therefore be run at a higher speed.

These manufacturing advances may occur ahead of marketing demands and, to satisfy the need for large volumes of chips in the budget market sector, Intel has often resorted to marking CPU's with speed ratings lower than the actual limits imposed by the quality of the silicon. This is why new processors usually debut with enormous heat sinks and chip fans but later steppings of the same CPU have smaller heat sinks and may even avoid the need for a fan altogether.

The second trick in the overclocker's armoury is that of supply voltages to the CPU. As already mentioned, the P166 uses the same silicon as the P150 but when Intel first developed the P166, the company was unable to get it to perform reliably using the P150 supply voltage of 3.3 volts. This is because the lower the voltage, the smaller the difference between a high and a low signal, and at high clock speeds this can result in signal noise.

The solution was to release the P166 at 3.45 volts. Modern motherboards allow the core and I/O voltages to be altered manually and often the only way to overclock successfully is to use a higher supply voltage. If the increase is too great, this can damage the processor permanently but a five percent voltage increase is normally within the listed operating range of the chip and can significantly improve signal quality at high clock speeds.

Fickle Electrons

Computer folklore has become filled over the years with tales of processors and other system components that have choked on this free lunch and suffered permanent damage. Your CPU can be destroyed instantaneously if you set the supply voltage too high but there is another side effect of overclocking that takes longer to manifest. This is the phenomenon of electromigration. This occurs at high temperatures over long periods of time as the electrochemical structure of the circuits on the silicon gradually becomes eroded. This shortens the lifespan of the chip and the higher the temperature, the more pronounced the effect.

The Intel marketing machine has tried hard to make users aware of the risk of electromigration and to link it to overclocking, but in reality, it is a problem that affects all processors to some extent. Overclocked CPU's are more prone to it only insofar as they are hotter, and this will depend on the degree of the overclocking and the effectiveness of any additional cooling measures installed.

In cases where the same silicon is used, there is no difference in the heat dissipation of an overclocked CPU and an official Intel CPU at the same clock speed, so the rate of electromigration is the same. In any case, Intel CPU's have a normal operating temperature range that extends up to 80 degrees Celsius and at this temperature they have a life expectancy of 10 years. Almost all processors – even overclocked ones – run significantly below this temperature and are obsolete long before their 10 years is up anyway.

More significant than the bogeyman of electromigration, however, is the issue of system stability. Very often, an overclocked system will boot up successfully but will then crash or hang more frequently than before. Often this is because the other components in the system are unable to adjust to the faster pace. If the bus speed has been increased, SCSI controllers may refuse to work and other interface cards, particularly graphics cards will also get hotter and suffer signal timing conflicts. Some EIDE interfaces may be driven to run too fast for their hard disks and CD-ROM's – this can be solved by changing to a lower mode for each device but only at the cost of decreasing overall performance.

To be sure that your system can operate at the new clock speed, it is important to increase the clock in small increments and to test it thoroughly at the new speed before attempting to move to the next one up. A long recorded game of Quake is a good way to put a processor through its paces.