Overclocking Noob Guide

Overclock in small steps.
- Test at each step.
- Repeat if stable.
- Reduce clock speed if not stable.
- Consider increasing voltage and cooling *slightly* if unstable, as you may be able to ramp up the clock speed significantly with small voltage increases... so long as cooling is adequate.

There's an algorithm for the amount of power required to run at a given clock speed, relative to the known good stable voltage/current and clock speed. However it may not be practical to use this with <= 45nm fabricated processors. (We're already well below 45nm now).

Just because a system can POST, boot, load Windows, load the Desktop, run Explorer, etc. does not mean that it is stable. (The processor and/or RAM could be defective, overclocking too far 'simulates' this somewhat).

There is a point at which circuits just cannot switch any faster for a given voltage, and given amount of energy output as heat, without destroying themselves.


Generally higher voltages let transistors switch faster, but you can't just run a < 22nm processor die at 5 volts (for example) to achieve high clock speeds as the chip would fail.

Processors are generally safe 'up to' 90 Celsius, however most people are not comfortable running near the edge and so many suggest limiting the peak temperature to 72C to 80C. This also helps with headroom as the ambient temperature in most rooms is not a constant year to year.

Power Supply Units lose efficiency over time, and if you're drawing more power (current is amperage, volts is voltage, wattage is the two multiplied for namesake) then you may 'one day' find the PSU is not outputting enough power (internally, DC) to be stable... whereas a year ago it was all OK.


In the 'old days' people used MemTest86, MemTest86+, Prime95, etc. to test that their processor and/or FSB (or BCLK) overclock was stable.

There are tools that check for artifacts when rendering with an overclocked GPU.

Generally running a part at max load for over 21 minutes is going to give a good indication if it's stable or not. That said your system should be stable when ALL parts are running at their 'new' max loads 'at the same time'.

Going past 21 minutes of testing, excluding RAM, doesn't help significantly, you could run for 72 hours and still not be 100% certain an overclock is 100% stable. However if it's >= 99.999,999,999,975% stable then 'for the most part' it's an acceptable overclock that is highly unlikely to cause problems.

Yes, all those 9's count, otherwise you'll notice issues at load within minutes.


Different parts of a system may require voltage tweaks to get processors to overclock.

Reading up on RAM timings and how they relate to clock speed is also a good start.
- A 16 tick wait at 1600 MHz is the same amount of time as a 4 tick wait at 400 MHz.
- Once that 1600 MHz 'memory' starts transferring some useful data it's performance would be 'up to' four times better though.

Overclocking is more about the relationship between TIME, VOLTAGE and HEAT than CLOCK-SPEED, VOLTAGE and HEAT.

A linear increase in clock speed generally requires more than a linear increase in voltage to achieve, as such power efficiency drops as you scale a given component to higher clock speeds.

i.e. You may need to 'up to' double power just to get a +41.42% overclock. Assuming the part doesn't melt as it was heavily under 'powered' to begin with.

If you need more than double the power to gain less than or equal to +41.42% more clock speed you're going to melt your component (probably), with <= 45nm parts this is far more likely at a much lower spec.

Most people never go past a +16.667% increase in voltage (relative to stock or 'reference' voltages).

Don't be greedy or stupid, it's far easier than people think and the typical worse case is just going to require using the CLEAR CMOS jumper on your motherboard to get BIOS defaults back.


I think that's about it, there should be some old guides in the forums somewhere with more specific info.