AMD K8 Overclocking Guide

Introduction

Overclocking, the practice of pushing your hardware to the limits is a widely practiced ritual by computer enthusiasts for as long as it has been possible to. Today, you can overclock the CPU, RAM and video card with ease. But if you don't know what you're doing, you can damage your precious hardware permanently. This guide's purpose is to get you familiar with overclocking on the AMD K8 platform, its basics, and even some advanced topics.

The AMD K8 Series

The AMD K8 series is well known for its overclocking capabilities. The K8 series includes all Socket 754 and 939 CPUs. There are many factors that make a CPU a good overclocker.
First we have the core. The core is the technology codename of your CPU. There are numerous cores for the K8 series. For Socket 754 we have Clawhammer and Newcastle for the Athlon 64, and the Paris and Palermo for Semprons. For Socket 939, we have the Clawhammer, Newcastle, Winchester, Venice, San Diego, Manchester and Toledo. This may seem like many, but there are very few differences between the cores.
Some are major, while others are less so. The difference between Sockets 754 and 939 is dual channel memory support. In real world applications the difference is about 4-6%, but you should always get the newest platform, in this case, Socket 939.

One major difference between cores is the process size and heat output. The process is measured in nanometers. The K8 series comes in 130nm and 90nm CPU flavors. The major differences are heat output and overclocking potential. The 130nm cores didn't overclock much, with an average max of about 2.5-2.6GHz using air/water cooling. They also put out significantly more heat. The 90nm cores can reach up to 3.0GHz with low voltages on air/water. The 90nm cores also default at 0.1V less, taking up to 10 degrees Celsius off of the maximum temperature. 130nm cores are no longer produced for the Socket 939 platform, so you don't have to worry too much.
There is no current reason to buy a Socket 754 platform unless you buy a budget Sempron, which are all 90nm. Overclocking is not guaranteed. Some CPUs overclock great, while the next one of the same model, doesn't overclock at all. These are extremes, so in general, most CPUs of the same Core overclock to the same maxes.

Let's examine the differences between the current cores. They are all 90nm. A complete list of all modern CPUs together with their specs can be found in our CPU Database.

Core	Socket	L2 Cache	Single/Dual Core
Palermo	754	128kb-256kb	Single
Venice	939	512kb	Single
San Diego	939	1024kb	Single
Manchester	939	512kb x2	Dual
Toledo	939	1024kb x2	Dual

As you can see, the 939 CPU cores can be paired together. They only differ in cache size.

It's All About Numbers

Now let's get into the speed numbers behind the K8 series. Let's use my Sempron 2600+ as an example.
Its stock speed is 1600MHz. How do they get this frequency? There are two factors that multiply to form 1600MHz. The first is the Frontside Bus (FSB). This is 200MHz for all K8 CPUs, regardless of the core. The second is the CPU Multiplier. For this CPU it is a maximum of 8x. This results in 200MHz x 8=1600MHz. for a 2000MHz CPU, the multiplier would be 10x. The multiplier can not be raised, only lowered. The CPU Multiplier on the K8 series CPUs can be lowered to 4x (though there is no need for that). The only exceptions to this rule are the FX series. Their multipliers are "unlocked" and can be changed in both directions. However things get a bit more complicated.

There is also a HyperTransport (HT) speed. This is calculated by multiplying the FSB by the HT multiplier. The HT Multiplier is a maximum of 5x for Socket 939 CPUs, and 4x for 754 CPUs. So the default HT Frequency for 939 CPUs is 1GHz, and for 754 CPUs it is 800MHz. The significance of this I'll discuss later. Unfortunately this gets even more complicated. The next part to get involved in this overclocking mess is the RAM. By default, the RAM speed is in sync (parallel) with the FSB. PC3200 or DDR400 memory is the default memory speed rating for the K8 series. This same memory runs at 200MHz (400MHz effective). This is the in sync, 1:1 ratio.

Overclocking example

Ok, let's look at a sample overclock. I've overclocked my Sempron to 2200MHz. The multiplier hasn't been changed. The FSB has been raised to 275MHz. Logically, you'd thing the HT multiplier is still 4x. However it is a rule that having HT Frequency higher than default brings instability, so I lowered the multiplier to 2x, just to be safe. Lowering the HT multiplier even at 200FSB doesn't affect performance one bit.
By default, then my memory would be at 275MHz as well. However, I'm using Corsair Value Select PC3200. As some of you may know, it doesn't overclock well, if at all. So I set a divider. A divider is a ratio between the memory speed and FSB. As we already know, by default the memory to FSB ratio is 1:1. Since my memory can't handle 275MHz, I had to set an async divider. Here are some dividers:




Those are the most popular dividers in all motherboards. I set the DDR266 divider. My RAM is now at 183MHz since 275MHz x 2/3 = 183MHz. There isn't any performance hit with my RAM under even default because of the high clock speed. PC3200 has a true frequency of 200 MHz. But it is doubled to 400 MHz due to DDR. The standard is PC3200. This means it has a peak rate of 3200 MB/s and runs at 200 MHz.

If your RAM does overclock, you just have to adjust its timings and voltage in order to reach the FSB on a 1:1 ratio. Tweaking RAM will be an upcoming article. This isn't the last thing I had to do to end up with my nice overclock. Voltages are critical to maintain stability.

Voltages

The default voltage for the 90nm CPUs is 1.4V. To maintain an overclock you often need to raise the voltage. They key is to set as low a voltage as possible, yet maintain absolute stability. Raising voltage increases temperature. You must make sure your 90nm CPU doesn't go far beyond 50°C at maximum load. There may be a need for additional/ aftermarket cooling, but for moderate overclocks, the stock cooling will do. There is usually a temperature monitoring program included with your motherboard. I only raised my voltage 0.04V, so my temperature didn't go up too much.

The temperature readings are usually off a few degrees compared to a diode attached to the CPU. Most newer motherboards can keep the difference of error within 2-3 degrees Celsius.

Stability Testing

Now you have to test the stability of your overclock. I use several programs to test my overclock. The first is 3DMark. I run both the 05 and 06 versions. If I pass without any artifacts (errors on the screen), I move on to SuperPi. There I run the 32M test which usually takes 30 minutes on average. If you have a dual core CPU, you must run two instances of SuperPi simultaneously. Once I pass that, I move on to Stress Prime 2004, a version of Prime95. This ferociously stresses both CPU and RAM. Again you have to run two instances if you own a dual core CPU. This program should run for at least 10 hours to guarantee "100%" stability.

Conclusion

So those are the basics of overclocking. Remember if you don't overclock you are missing free performance. Overclocking is a great method to get a free performance boost. If you don't overclock, you're missing out on free fps.