What do I do if my computer won't post (Display the BIOS screen when it is turned on)?
This varies based on the motherboard you have. The "fail-safe" solution (unless you killed something) is to reset the CMOS, usually by moving a jumper for a set amount of time. Check your motherboard manual for the specifics. Most recent enthusiast level boards have an option to post at reduced frequencies if the overclock is pushed too high but leave the BIOS settings intact, so you can go in and lower the clock speed to where it is stable. On some motherboards, this is done by holding the Insert key when you turn on the computer (usually has to be a PS/2 keyboard). The DFI LanParty PRO875 is one such board. Others automatically reduce the frequency if the computer didn't post on the previous attempt. The A7N8X usually does this. Sometimes a computer will not cold boot (post when the power button is pressed) but will work if it is left on for a while, then reset. On other occasions the computer will cold boot fine, but will fail to warm boot (reboot). Those are both indications of instability, but if you are happy with the stability and able to deal with the issues than it usually won't cause any huge problems.
What limits my overclock?
Generally, the RAM and CPU are the only significant limiting factors, especially in AMD systems because of the problems inherent in running the memory asynchronously (see the FSB section down below) The RAM has to run at the same speed as the FSB or at a fraction of it. Complex fractions are allowed, meaning the memory can be run at a higher rate than the FSB, not just a lower one. With the option to run looser timings/more voltage through memory, though, it is becoming less and less the limiting factor, especially since newer platforms (P4 and A64) suffer less of a performance hit from running async. (again, see below) The CPU has become the main limiting factor. The only way to deal with a CPU that doesn't want to run any faster is to pump more voltage through it, though exceeding the maximum core voltage shortens the life of the chip (though overclocking does this as well) but sufficient cooling stems this problem. Another problem with running too high of a core voltage manifested itself on the P4 platform in the form of SNDS, or Sudden Northwood Death Syndrome, wherein running any voltage over something like 1.7 (not sure of the exact number, no one is) would result in the quick and untimely death of the processor, even with phase change cooling. However, the newer 'C' core chips, the EE chips, and the Prescott chips have not had this problem, at least not to nearly the same extent. The cooling can also prevent a good overclock, as having temps that are too high can lead to instability. But if your system is stable, then the temps usually are not too high.
Now that I've overclocked a lot, what should I do?
Run some benchmarks if you want to. Run Prime95 (Or your stress test of choice - it is up to you) for a sufficient time period (Usually 24 hours straight is considered a stable system)
/Begin shameless F@H plug
Then install Folding@Home if you haven't already.
/End shameless F@H plug
That covers the basic aspects of overclocking. The questions from this point on are the more technically involved sections.
What is the FSB?
FSB (or the Front Side Bus) is one of the easiest and most common ways to overclock. The FSB is the speed at which the CPU interfaces with the rest of the system. It also affects the memory clock, which is the speed the memory runs at. Generally speaking, higher=better for both the FSB and the memory clock. However, there are certain cases where this is not true. For example, running the memory clock faster than the FSB does not really help at all. Also, on Athlon XP systems, running the FSB at a higher rate, but forcing the memory to run out of sync with the FSB (using memory dividers which will be discussed later on) will hamper performance far more than running at a lower FSB with the memory in sync.
The FSB is referred to in different ways on Athlon and P4 systems. On the Athlon side, it is a DDR bus, meaning that if the actual clock is 200mhz, it is said to be running at 400mhz. On a P4, it is "Quad-Pumped" so if the actual clock is the same 200mhz, it is said to be 800mhz. This makes for a good marketing strategy for Intel, because as every average Joe knows, higher=better. Intel's "Quad-Pumped" FSB actually has a real-world advantage, as it is what allows P4 chips to run the memory out of sync with less performance loss. The higher rate of cycles per clock gives it a better chance of having the memory cycles line up with the CPU cycles, which equates to better performance.
Why does running the PCI/AGP bus out of spec cause instability?
Running the PCI bus out of spec causes instability mainly because it forces components with very strict tolerances to run at a different frequency then they are intended to. The PCI spec is usually stated at 33mhz. Sometimes it is stated at 33.3mhz, which I believe is closer to the real spec. The main victim of high PCI speeds is the hard drive controller. Certain controller cards have a higher tolerance than others, and so are able to run at increased speeds without noticeable corruption. However, the onboard controllers on most motherboards (especially SATA controllers) are extremely sensitive to high PCI speeds, and can have corruption and data loss if the PCI bus is running at even 35mhz. Most are able to do 34mhz, as it is really less then 1mhz out of spec (depending on where the motherboard stops rounding to 34mhz... for example, most motherboards will probably report any FSB from 134mhz-137 as being a 34mhz PCI speed. The actual range is from 33.5mhz to 34.25mhz, and may vary even more based on variations in the clock frequency of the motherboard. At higher FSBs and higher dividers, the range can be even more). Audio and other integrated peripherals also suffer when the PCI bus is run out of spec. ATI video cards are a lot less tolerant to high AGP speeds (directly related to PCI speed) than nVidia cards. With that in mind, most Realtek lan cards (the PCI based ones that occupy an expansion slot) are rated for safe operation at anywhere from 30-40mhz.
What is the multiplier?
The multiplier acts in conjunction with the FSB to determine the clock speed of the chip. For example, a multiplier of 12 with a FSB of 200 will give a clock speed of 2400mhz. As explained in the overclocking section above, some chips are locked and some are unlocked, meaning only certain chips allow adjustment of the multiplier. If you have multiplier adjustment, it can be used to either get a higher clock speed if the FSB is limited on your motherboard or to allow a higher FSB if the chip is limited.
What is a memory divider?
A memory divider determines the ratio of the memory clock speed to the FSB. A 2:1 FSB:RAM divider would net a 100mhz ram clock with an FSB of 200mhz. The most common use of a divider is to allow a P4C system to run 250FSB with PC3200 ram, with a 5:4 divider. There are also 4:3 dividers and 3:2 dividers on most Intel (DDR1) systems. Athlon systems can't use the memory as efficiently as P4 systems when a divider is used, as explained above in the FSB section. The memory dividers should only be used to obtain stability, not just at a whim, because even on a P4 it hurts performance somewhat. If your system is stable without resorting to a memory divider (or if a memory voltage bump can fix the problem) then don't use the dividers.
What do the different memory timings mean?
CAS Latency , sometimes referred to as CL or CAS, is the minimum number of cycles the ram must wait until it can read or write again. Obviously, the lower the number (time), the better.
tRCD is the delay before the data on a particular row in memory is read/written. Again, the lower the number the better.
tRP is basically the precharge time of the row. tRP is the time the system waits after writing something to a row before another row can be active. Once again, lower is better.
tRAS is the minimum value for how long a row can be active. So basically, tRAS is how long the row has to be turned on. This number varies quite a bit with RAM settings.
What do the various memory ratings refer to? (PC2100/PC2700/PC3200 etc.)
The rating refers directly to the maximum bandwidth obtainable and indirectly to the memory clock rate. PC2100, for example, has a 2.1GB/S maximum transfer rate, and a clock rate of 133mhz. PC4000, as another example, has a 4GB/S ideal transfer rate and a 250mhz clock. To obtain the clock rate from the PCXXXX rating, divide the rating by 16. Multiply the mhz rating by 16 to obtain the bandwidth rating.
How does the DDRXXX refer to the actual clock speed of the memory?
The DDRXXX is just two times the actual clock speed; i.e. DDR400 is clocked at 200mhz.
if you want know the pc-XXXX speed of the DDRXXX speed, times it by 8.