Overclocking For Performance

Pages: 1 2 3 4 5 6 7 8

Overclocking Issues

Although there can be some very tangible benefits from overclocking, there are also some risks. Strange as it may seem to some, it is actually possible to overclock, yet gain no real performance benefits. In fact, it is entirely possible to end up with a decrease in any or all of the factors related to performance.

The drawbacks of overclocking may be:
  • Damage to motherboard and/or processor
  • Voided warranty on motherboard and/or processor
  • System instability (crashes and lockups)
  • System unreliability (possible data corruption)

One of the primary issues to deal with when overclocking is heat. The faster the CPU ‘spins’, the more power it dissipates, which in turn generates more heat. Heat flows from points of higher temperature to points of lower temperature. Therefore, before heat can move away from the center of a conductor toward it’s surface, the internal temperature must be higher than the surface temperature. Even with a very good heatsink/fan it is still possible for the internal chip temperature to be significantly higher than what it should be, thereby potentially reducing the life of the chip and potentially introducing problems with stability and reliability.

As stated, a faster CPU means increased power requirements, which may result in some motherboard components exceeding their specs as well. Generally, this is not a major issue however it should not be overlooked. The current draw by the processor is determined by the speed. The processor data sheets will usually indicate the number of milliamp (mA) for each 1MHz of speed, so the current (and therefore power dissipated) can be fairly easily calculated (P=I*E). One common practice is to bump the voltage to the processor up to improve stability, which dramatically increases the power being dissipated by the processor (P=E^2/R). The reason that higher voltage can improve stability is that transistors require power to operate, and increasing the power may make them operate a bit faster (up to a point!). On the other hand, transistors of a given size (for example .35 micron) have an optimum voltage (in this case 2.5v), and increasing that voltage too much can damage the transistors more quickly.

An example of this would be the K6-233 running at 3.2V, which draws about 8.9A for a power draw of 28.5W, with a peak of 9.5A (30.4W) under heavy load. This equates to about 38mA to 41mA per MHz, so increasing the processor speed to 266MHz adds between 1.25A and 1.35A of current, or 4W+ of power. Bumping up the voltage to 3.4V to increase stability would add another 2-3W, bringing the total power being dissipated to between 35W and 39W, depending upon the processor load. If the manufacturer had designed the board to run the K6-233 as it’s most powerful processor, assumed the maximum power requirements (30.4W) plus 25% for safety, this would mean that the maximum ‘safe’ power draw would be 38W. If the CPU happened to be working very hard for a period of time, and any of the motherboard components was marginal, chances are pretty good that something will fail prematurely. Note that this example uses one of the ‘worst case’ processors in the K6 233MHz (see the related CPU power requirements chart for more info)

Another consideration is the phenomenon known as ‘electromigration’, which occurs more rapidly as the current increases, similar in some ways to the erosion of a riverbank. Electromigration causes the circuits to ‘erode’ away, resulting in tiny ‘pits’ that increase the resistance, which in turn increases the heat that is generated. This occurs in all circuits over time, the extent of which is a factor of both time and the current that has been running through it. (see This Page for additional information

Based upon these facts, it is best to run a processor within the limits specified by the manufacturer. While it is true that manufacturers will sometimes ‘down bin’ parts to fulfill market requirements, this is generally not the case, and should not be relied upon when deciding how or if you will overclock. It is said that a 25% tolerance is built into most processors from Intel, and somewhat less from the other manufacturers, however this number is probably too high. For maximum safety, 10% is more likely to be realistic, while 20% should be the absolute maximum increase in speed to ensure any kind of reliability, stability and longevity. While there are exceptions to this rule, unless you know for certain that the tolerance is much greater, exceeding these limits will almost certainly result in increased system problems and perhaps increased RMAs.

In some cases, the I/O bus may also be overclocked. Intel chipsets have only a synchronous bus option, VIA and Ali chipsets generally have a pseudo-synchronous bus, and the SiS chipsets normally have an asynchronous bus, though it may vary somewhat with specific chipsets. This may be a big consideration in deciding how to get the best throughput without introducing device problems. The PCI spec requires only that a device operate at 33MHz, so anything beyond that is no guarantee. Most devices will fall into the 10% tolerance discussed for processors, but some will not. This means that you should not allow the PCI bus speed to exceed approximately 37MHz if you want a stable and reliable system. Again, there are exceptions but timings are very critical and many of the true high-performance devices experience timing problems at higher bus speeds.

When trying to decide exactly how to overclock, and how much to overclock, it is necessary to identify the requirements of the user. There are some people for whom occasional crashes and shortened lifespan are small prices to pay for increased throughput (these people usually measure this in video frame rates). There are others, however, who would like better throughput, but cannot afford any loss of stability, reliability or longevity who will most likely be businesses or individuals with limited disposable income. The results from these tests very clearly show that there are situations where overclocking provides zero benefits, but may reduce the reliablity and/or stability of the system. There is obviously a difference between simply overclocking and intelligent overclocking.


Pages: « Prev   1 2 3 4 5 6 7 8   Next »

Be the first to discuss this article!