Overclocking is making a computer component run at a higher clock speed than the manufacturer's specification. Although there are many different reasons for overclocking, the most popular reason is to increase hardware performance. Overclocking can result in system instablity and sometimes even hardware failure if done carelessly.
Overclocking is mostly practiced by PC enthusiasts in order to realize the full potential of their computers. Some hardware enthusiasts purchase a low-end computer and overclock it, thereby attaining performance of a higher-end system, while others will overclock high end components, attaining unprecedented performance. The fastest home computers in the world are overclocked and often a non-overclocked computer is slower than an overclocked one.
Considerations for overclocking
The main requirement of overclocking is an effective cooling system to remove the excess heat overclocked components emit. Because most coolers are designed for the heat a non-overclocked component produces, overclockers typically purchase high quality cooling systems. Copper heatsinks in combination with powerful fans provide better cooling than OEM coolers. Water cooling is used as well, which when properly implemented provides much more powerful cooling then heatsink/fan combinations.
System stablity is a major concern when overclocking. Without proper cooling, an overclocked component can overheat and crash. Stress tests (also known as burn-in) can be used to test system stablity. A stress test places a high load on the component being overclocked. Programs commonly used for stress testing are Super-Pi, Prime95, SiSoft Sandra.
Commonly overclocked components include: processors, video cards, motherboard chipsets, and RAM. Methods that have been used to cool overclocked components include: forced convection (a fan blowing onto a surface); liquid cooling (liquid carries waste heat to a radiator, similar to how automobile engines are cooled); liquid nitrogen (perhaps the most dangerous method); dry ice; phase change cooling (as used in refrigerators); and submersion (placing the entire computer in an inert fluid). Liquid nitrogen is a temporary cooling measure in most cases, since a sufficient supply of power to maintain the LN2 coolant at liquid state is uneconomical. Because of this, liquid nitrogen (or dry ice, for that matter) is used as an extreme measure to set a record in a one-off experiment rather than to cool a system for a normal period of use. One reason is the cost of these extreme cooling methods, or usually because the hardware exposed to such cooling is ironically destroyed in the process. Of the aforementioned methods, air cooling, liquid cooling, and phase cooling are the most popular, due to their efficiency, availability, and affordability.
Overclocking arises in part due to the economics of the manufacturing processes of CPUs. In most cases, CPUs with different rated clock speeds are manufactured via exactly the same process. A batch of CPUs may be tested and binned--that batch is set to operate at a specific frequency because all of the processors function at that clockspeed. The clock speed that the CPU is marketed under is the speed at which the CPU has been tested to operate consistently well, but often there is a distribution where on one end there are the CPUs which are near their physical limit at the specified clockspeed and on the other end there are CPUs which can operate at frequencies substantially higher than their specifications. With proper power and cooling, slower CPUs can be made to run at the same speed, or faster, than similar CPUs with higher stock clockspeeds. A commonly held view is that overclocking results in system instability. This is rarely the case when the system is properly tested and the temperature and voltages are monitored.
In addition, there have been situations in which a chip manufacturer will deliberately underrate a chip in response to market pressure. This results in an inexpensive component, which (with a little extra voltage) is easily overclocked to match the speed of a more expensive component. One example is the AMD Athlon XP 2500 (codename Barton) processor, which was easily made as fast as the AMD Athlon XP 3200 , which was four times as expensive at the time.
Recently computer experts experimented with a Pentium 4 3.4 GHz HT processor, cooling it using liquid nitrogen, and blowing cold air at high speeds past it. They managed to achieve over 3 GHz above the original frequency, which is a considerable amount. Of course, it is cheaper, more energy efficient and convenient to use a dual processor computer than to overclock a single processor to these speeds. Very few users would tolerate regularly topping off their computer with liquid nitrogen, but even the noise of such a system would make it unsuitable for many practical uses. These tests are interesting, however, as an illustration of what is possible when great amounts of heat can be removed from a system and are an indication of what could be achieved with better (but not as drastic) heat sinking.
Measuring effects of overclocking
For some overclockers, the increased clock statistics are a reward in themselves, while others take the more pragmatic view that perceptible improvements are necessary to justify the effort. Human judgment on the speed of a computer is inherently subjective and open to the placebo effect, therefore there are many de facto benchmarks used to evaluate performance. The benchmarks can themselves become a kind of 'sport', in which users compete for the highest scores.
Given only benchmark scores it may be difficult to judge the difference overclocking made to the computing experience. Some benchmarks test only one aspect of the system, such as Random Access Memory bandwidth, without taking into consideration how faster speeds in this aspect will improve the system as a whole; memory bandwidth is typically not a bottleneck, apart from 'serious tasks' like video encoding, high-demand databases and scientific computing. Other benchmarks, such as 3D Mark attempt to replicate game conditions, but because some tests involve non-deterministic physics, such as ragdoll motion, the scene is slightly different each time and small differences in test score are overcome by the noise floor. Most overclocking sites use recorded demos of recent games for benchmarks, which give a good indication of the performance in that game, but are still not completely realistic because the system does not have to perform physics calculations.
Overclocking by resellers
Commercial system builders or component resellers sometimes overclock to sell items at higher profit margins. By buying lower-value components, overclocking them, and selling on as higher value ones, the retailer makes more money. In some cases an overclocked component is functionally identical to a factory-clocked one, especially if it was deliberately underrated by the manufacturer; however, it is generally considered dishonest if the customer is not told they are buying overclocked equipment. It is felt that because of a risk of shortened component lifespan, the customer be allowed the informed choice to use overclocked components or not.
Overclocking is sometimes seen as a legitimate service, in which a company tests the 'overclockability' of individual items rather than the customer just buying and hoping theirs will overclock well. One example is the retailer who checks which GeForce 6800 cards work correctly with extra pixel shaders unlocked (effectively making it a 6800 ultra), and charges slightly above the retail price for the cards known to work. Many specific 'tricks of the trade' such as this are highly dangerous and can render hardware non-functional, though they are attractive since as with all overclocking, the user is getting a 'free lunch'. Simple feature unlocking such as this can often be done by simply joining two points on a circuit with a graphite pencil (known as the pencil tick)
Of course, manufacturers would like high performance seekers to pay extra for high-end products, but also fear that less reliable components and shortened life span would damage brand image. It is mainly fear of this kind of practice that motivates major manufacturers to design overclocking prevention mechanisms such as CPU locking. These measures are claimed to be customer protection, which often meets a mixed reception.