We've finally arrived at the meat of this guide: the actual overclocking. Before you start, it's a good idea to download and install a good stress-tester, such as Prime95, as well as CPU-z for obtaining information about your overclock. If you followed the second page of this guide, you'll already have the CPU-z .EXE file somewhere.

It's time to get down and dirty with your BIOS. Reboot your operating system, and get into your motherboard's BIOS. Please note that the screenshots I've provided here are from the DFI LanParty UT nF4 Ultra-D motherboard. All BIOSes differ slightly in vague ways, though the general layout and label names are very similar. If you have trouble understanding some of the options, write their names down or take a photo, and ask on our forum.

At this point, you now have to decide whether you want to attempt to overclock your RAM as well as your CPU, or just your CPU. As I talked about in my Athlon 64 Overclocking Techniques guide, overclocking the HT bus alone but not the CPU/RAM yields no performance benefits.

I'll start by outlining the rough overclocking procedure for a CPU-oriented overclock:

1. Raise HT bus by 10-20 MHz
2. If needed: Lower LDT bus multiplier
3. If needed: Increase memory divider so as to keep memory as close to highest rated speed as possible without going over
4. Save BIOS settings and reboot
5. If you successfully get into Windows, shut down, go into the BIOS, and increment the HT bus some more, keeping in mind the LDT bus multiplier and memory divider
6. If you fail to get into Windows, reboot, go into the BIOS, and increase the CPU voltage a bit (1.4V to 1.45V for example, though up to 1.6V is safe for a Winchester Athlon 64)
7. If you still fail to get into Windows with the voltage bumped up, reset the voltage to its default value, and decrement the HT bus by 5 MHz
8. Repeat the above procedure until you find the highest HT bus setting that you are capable of getting into Windows at
9. Attempt to run the first test in Prime95's Blend torture mode
10. If the first test fails, go back into the BIOS and drop the HT bus by 5 MHz, and repeat the Prime95 test
11. When you find an HT bus setting that can pass Prime95's first test, continue testing. You should generally aim for at least two hours of Prime95 without failures, though if you plan on doing any computationally-intensive tasks where the final result should be as accurate as possible (SETI@Home for instance), then you should aim for at least 8 hours of stable Prime95

That is the general flow of how an overclocking session goes. As you can see, it's very hit-and-miss-oriented: you try, and try, and try, until you fail, and then you turn it down a notch. I will now delve into the specifics of the actual procedure and decision-making involved.

The HT bus setting refers to the reference speed from which the processor and various buses derive their true speed (what you know as the "GHz rating"). The processor takes the HT bus setting, and "multiplies" it using a multiplier (on an Athlon 64 3000+ S939, this is 9x), to give the internal clock speed. Raising the HT bus setting but keeping the multiplier constant yields an increase in processor clock speed.

It seems simple enough, doesn't it? It isn't. If overclocking were as simple as raising the HT bus speed alone, there would be no need for this guide. It's the adverse effects of raising the HT bus speed that snags users. These three effects are the following:

• Raising the HT bus speed also raises the LDT bus speed, which is derived in a similar way to that of the internal processor clock speed
• Raising the HT bus speed raises the internal processor clock speed, from which the memory speed is derived through the use of a memory divider
• On certain chipsets/motherboards, the PCI/AGP and SATA/IDE busses are not locked at their operating frequencies, and instead scale as you raise the HT bus speed

The "LDT bus", which stands for "Lightning Data Transport bus", is the bus through which the processor communicates with the chipset. The chipset controls things such as hard drives, the Ethernet interface, and USB, and any time the processor needs to send or receive information from any of those sources, it travels over the LDT bus.

Fortunately, the first two items are addressed through the use of the LDT bus multiplier, and an adjustable memory divider. Unfortunately for some of you, if you own a motherboard that does not lock the PCI/AGP/SATA/IDE busses, you will not be able to raise your HT bus very much. There is nothing you can do to improve that, short of buying a new motherboard. However, most Athlon 64 owners will probably be using an nForce 3/4-based motherboard, in which case they have nothing to worry about. Some nForce 3 motherboards have two unlocked SATA ports (usually the ones located at the bottom of the board).

As I mentioned before, the processor derives its internal clock speed by multiplying the HT bus speed with the processor multiplier. The LDT bus works in a similar fashion. The LDT bus multiplier multiplies the HT bus speed to achieve its final clock speed. By default, the HT bus speed is 200 MHz, and the multiplier (depending on the chipset you have) is either 4x or 5x, giving 800 MHz and 1000 MHz, respectively. The problem with the LDT bus is that it doesn't take too well to overclocking. On some motherboards, you may be able to get 10% out of it, and that's a maximum. On nForce 3 250Gb and nForce 4 chipsets, the highest I tend to see is about 1100 MHz. For this reason, it is important to turn down the multiplier when overclocking the HT bus.

For example, if you raise your HT bus to 250 MHz, the LDT bus would be running at 1250 MHz with the default multiplier of 5x. If you drop the multiplier to 4x, you are once again at the optimal 1000 MHz speed. Do not forget to double-check the LDT bus multiplier every time you make an adjustment to the HT bus. Having a calculator with you is handy, as it allows you to do quick calculations while you're in the BIOS.

Next up is the issue of memory. Most of you will probably have generic PC3200 memory, and you probably won't be able to get more than 10 MHz out of it. For this reason it's best to use a memory divider if you want to actually get some kind of overclock out of the system. A memory divider on the Athlon 64 platform is like a multiplier, except it goes in the opposite direction. In the BIOS, the memory divider is usually called "Memory speed", or "Max memory speed", and the options given are usually "200 MHz", "166 MHz", "133 MHz", and "100 MHz", though some motherboards may offer finer granularity.

For all intents and purposes, the memory divider is merely a ratio of HT bus speed to memory speed. The 200 MHz option is a 1:1 ratio, meaning that an HT bus speed of 200 MHz will yield a memory speed of 200 MHz. The next most common option is 166 MHz, which is a 6:5 ratio. If your HT bus speed is 200 MHz, and you use this ratio, you will get a memory speed of 166 MHz. It really is quite simple, but you will most likely need a calculator to do some calculations once you're tweaking things. For instance, using the 166 MHz divider (6:5), you have to run your HT bus at 240 MHz in order to achieve the standard 200 MHz memory clock again. This actually, isn't too bad of a situation, because most lower-end 90nm Athlon 64 parts (I have Winchesters in mind) can comfortably do 240 MHz on the HT bus with their stock multipliers. Using the Athlon 64 3000+ as an example, 240 MHz * 9 gives us 2160 MHz, which is already a healthy bump over the stock 1800 MHz.

As with the LDT bus multiplier, make sure you keep an eye on what you'll be running your memory at with every adjustment of the HT bus, trying to keep it as close to its rated speed as possible through the use of the memory dividers.

The moment that you come across an HT bus speed that doesn't boot into Windows successfully, double-check what your memory and LDT bus are running at, and if they seem to be within specification, give your Athlon 64 a little voltage bump. The stock voltage of Athlon 64 Winchester parts is 1.4V, and anything up to 1.60V is probably pretty safe, though you might need to keep an eye on CPU temperatures if you do end up using 1.60V. At this point, give 1.5V a try and see if you successfully boot into Windows. If it still doesn't work, give 1.6V a try. If that doesn't work, you will either have to turn down the HT bus speed, or attempt bumping the LDT bus speed voltage a bit. Most motherboards don't give you the option of adjusting this, but some higher-end motherboards do. Give the voltage a 0.1V bump and try again.

After going through this process, you should arrive at the highest HT bus speed that your motherboard/processor are capable of booting into Windows at. However, your adventure is not over yet. If there's anything that kills an overclocker's ego, it's the stability-testing portion of the crusade.

Fire up Prime95 (which you downloaded earlier), and click "Stress-testing Only" if given the option. Click Options, and then Torture Test. Make sure the "Blend" mode is selected, and then click OK. You will then see something like the following:

What this mean is that Test 1/1024K is running. If you are lucky, you should see a second (similar) line will appear, starting with the text "Test 2", after 3-4 minutes. Most of us are not so lucky, though. The highest bootable speed is rarely ever the highest stable speed, and Prime95 will error out with a message reporting hardware failure. At this point, reboot your computer, go into the BIOS, and turn down the HT speed a bit, and repeat the Prime95 procedure.

Eventually you will reach a speed at which the first test passes. Continue to let Prime95 run. You might as well go occupy yourself with something else at this point, as Prime95 might error out within 10 minutes again, or after two hours, or perhaps never. This is where the argument of what is stable comes in.

In general, if your hardware can sustain at least two hours of intense Prime95 action, it is "stable enough". What I mean by stable enough is that you should be able to use your computer regularly, and generally without trouble. The chances of something crashing are quite low, but they are there. Playing a game for a few hours might cause an error to creep out, in which case either the game or operating system will crash. If you reach 8 hours of stable Prime95, you can more or less be assured that your computer will never crash.

Technically, an overclock that crashes after 8 hours is not 100% stable, but stability really is relative to the observer. If you depend on your computer to always pump out 100% accurate results in whatever you do (scientific simulations, for instance), you shouldn't be overclocking in the first place, but if all you ever do is play games and listen to music, 100% accuracy is not required. Personally, I generally aim for at least 20 hours of stable Prime95'ing. If my processor can survive 20 hours of extreme torture, it should be able to handle anything I ever throw at it. I'm a perfectionist in this sense, and I'm sure many people would argue that 6-10 hours is enough.

If you so wish, you can, at this point, start turning down your voltage, and running tests after each decrease. The effect that this will have is that you will end up with the lowest voltage required to operate the CPU stably at your new overclocked speed, hence lowering heat output. This is, of course, quite optional, but something to consider nonetheless.