Overclocking:
Some Basic Principles
Cautionary Prelude
The process of overclocking a processor or a video card is fairly simple to do. However, before we plunge into the how-to portion of this article, there are a couple of significant drawbacks to overclocking that you should be aware of. First of all, when you overclock a component, you almost always are increasing the electronic wear and tear on it. Second, by running a component out of specification and at a point close to that at which it fails to operate properly, you greatly increase the risk of introducing instability into your system. This instability may not appear right away, but the likelihood of it developing increases over time.
One way to think about overclocking is that it is taking advantage of the extra headroom that is left between the operating range that the product is specified to operate within and the actual point at which the component begins to malfunction. Obviously, the manufacturer doesn't want a bunch of their devices returned for buggy behavior, so they design in a cushion between the specified operating range and the actual point at which the device will begin to fail. This unused operating potential also means that the product can absorb a certain amount of deterioration in its functioning over the course of its lifetime, without this having a noticeable impact on its operation. However, when you overclock a component, you use up this leeway in its operational specifications, and you increase the rate at which the component will begin to deteriorate. Eventually these two factors will come together and you'll have a component that is no longer working as it should. The detrimental results may not be immediately or dramatically noticeable, but you can be sure that your system will become increasingly unstable, as corrupted data is steadily generated.
This means that, if you use overclocked components in your system, you must always be wary that they are beginning to fail. You need to rigorously test your system for stability, not just when you first set up your system and determine what overclocked settings to use, but periodically after that, as well. Otherwise, you can never know whether any crashes, freezes or blue screens that occur are the result of buggy software or glitches caused by your failing hardware.
Selecting a Good Candidate
For most of this article, I'm going to focus on overclocking CPUs, since these are the most commonly overclocked system components. The how-to process of overclocking them also illustrates the more general principles of overclocking. Later, I'll say a few specific things about overclocking video cards.
When considering which processors make good candidates for overclocking, it is useful to understand how the manufacturer goes about producing their CPUs. The manufacturer, such as Intel or AMD, typically offers a line of processors, such as the Pentium 4 line or the Athlon XP line. Within these product lines there usually are sub-groups of similar chips that are based on the same "core". Although these chips may be offered to run at different clock speeds, and perhaps even make use of different front side bus speeds, if they use the same core, they are essentially the same chip. There is no real difference in how they are manufactured. There may be a certain amount of sorting, after their manufacture, into those that are considered better or worse in terms of their performance capabilities, but whether a particular chip will end up being designated a 2.4 GHz version or a 3.0 GHz version, for example, is done completely after it is produced. Usually, as the manufacturer gains more experience producing products based on a particular core, their yield of "good" chips increases, meaning that almost all the chips based on this core are capable of functioning at a level near to the core's full potential.
Some examples of various cores among current processors are the Thoroughbred and Barton cores for the Athlon XP and the Northwood core for the Pentium 4. The latest version of the Northwood is distinguished from the earlier Northwoods by it being designated to run on a 800 MHz front side bus and by it offering hyperthreading capabilities for all its versions.
So, the first principle of overclocking is to select a chip from further down the line of those based on a particular core with an aim of running it at a speed close to that of the top end version. These chips will obviously be less expensive than their higher speed kin, and they should have the most room for increasing their performance, hopefully up close to the limits determined by the design of the core itself. Of course, what is true in principle may not be born out by experience, so I recommend that you look into the experiences that others are having overclocking the same chip that you are interested in. You can find these kinds of discussions on various hardware forums, and there is a database of useful information at overclockers.com. When using the database, look at the results that are obtained with voltages kept at the default level (or very close to this) and focus on those that used conventional cooling, not water cooling or some form of refrigeration. These results will give you the best idea of what is a likely overclock for particular chip.
Another principle is that if the processor you are interested comes in versions that run on different speed front side buses, the lower front side bus speed model is the better choice for overclocking. The reason for this will be apparent as we go into how to overclock your processor, in the next section.
How to Do It
The clock speed of a CPU is a function of the speed of the front side bus that it is running on and a "multiplier", which has been set by the manufacturer. For example, a PIII 700 is specified to run on a 100 MHz bus and it has a multiplier of 7 (7 x 100 MHz = 700 MHz). These days almost all processors have a multiplier that is hardwired into the chip, so it cannot be changed by the user (through motherboard settings, for example). An exception to this has been the Athlon XP processors, which could have their multipliers changed through the BIOS, if the motherboard used an Nvidia chipset. Recently, AMD began locking the multipliers on the Athlon XPs with the Barton core. Chances are, if you buy an Barton Athlon now, that it will have a locked multiplier. I don't know if they have starting locking the multipliers on their Thoroughbred cores, as well, and I believe that the new Athlon 64 processors can only have a lower multiplier set, but not a higher one, compared to what was set by the factory. So, it is best not to plan on using a processor's multiplier to overclock, these days.
Assuming that the multiplier cannot be changed, the only way to overclock a processor is by manipulating the other variable in the equation, the front side bus. This is why, if given the choice, a processor that uses the slower bus speed will make for a better overclocking candidate than a similar model running on a faster front side bus. Thus, a kind of ideal overclocking candidate will be a processor with a low overall clock speed, compared to similar processors using the same core, and it will be running on a slower speed bus, if such a version available.
Some processors that I have successfully overclocked are good examples of these two factors. The Celeron 300A was meant to be run on a 66 MHz bus, but it did very well running on a 100 MHz bus, yielding a nice 50% overclock to 450 MHz. (Although there were no Celerons running on a 100 MHz bus, at that time, the Pentium II ran on a 100 MHz bus; so, this setting was readily available.) The PIII 700 and the PIII 750 were both designated to run on a 100 MHz bus, but they could tolerate running on a 133 MHz bus, yielding a 33% overclock to 933 MHz and 1000 MHz, respectively. Both the 100 MHz and the 133 MHz bus speeds were used by Pentium IIIs, at that time. The Celeron 1.2 GHz is also meant to run a 100 MHz bus, but it took to a 133 MHz speed bus with no apparent difficulty, yielding a new clock speed of 1.6 GHz. Again, both of these bus speeds are readily available on PIII motherboards.
Among the current processor that are popular for overclocking are the Pentium 4 2.4 GHz and the Athlon XP 2500. Both of these are lower end processor, compared to top of the line versions, and they come in flavors that run on lower speed buses, compared to the fastest bus speeds that other processors are using.
Just a side note on the front side bus speeds that you'll see listed for Pentium 4s and Athlon XPs, these speeds are usually given as their "effective" clock speeds, not their actual clock speeds. The Pentium 4 makes use of a "quad pumped" front side bus, so when you see a processor that is listed as running on a 800 MHz front side bus, the actual clock speed of this bus is only 200 MHz. This actual clock speed is the number that is multiplied 12 times, in the case of a 2.4 GHz processor, to give you the core's clock speed, 2400 MHz or 2.4 GHz. Athlons make use of a double data rate bus, so if you see references to a 400 MHz bus or a 333 MHz bus, these are actually 200 MHz and 166 MHz speed buses that are being talked about. Again, those are the numbers that are being "multiplied" to give you processor's clock speed.
What Else Should You Look For?
Ok, lets say that you've picked out a processor and are confident it has lots of potential. Is that all there is to it? Well, it's going to be a lot easier if you have the proper supporting components. Let's start with how you are going to cool your processor.
We've already talked about how overclocking your processor is going to increase the electrical wear and tear on it, and in part, this is because running a processor at a faster clock speed will produce more heat. And, since overheated electrical components don't work well, we need a cooling solution that will let us achieve and maintain high clock speeds. The stock heatsink that typically comes with a retail processor probably will not be adequate for the overclocked speeds that you expect to push your processor to. So, it is a good idea to pick up an aftermarket cooling solution. Typically, we are looking for a big chunk of aluminum or copper (or some combination of the two) along with a big enough fan to move plenty of air across this heatsink. Just make sure that this heatsink will be compatible with your motherboard and case. Since these tend to be oversized components, they may not fit properly with all motherboards and cases.
Water cooling and even refrigeration, are becoming increasingly popular as ways to get the most out of an overclocked processor, but I'll leave aside these approaches for this article, since these solutions tend to be expensive and perhaps a bit too experimental.
We've also already talked about how most processors will be overclocked by raising the speed of the front side bus. You do this through settings made available by the motherboard's BIOS, but some motherboards provide more flexibility and options for this doing this than others. One of the most important overclocking features to look for is the ability to lock the PCI bus and the AGP bus speeds to 33 MHz and 66 MHz, respectively. This is a relatively new feature, but it really makes an overclocker's work much easier. Without this feature, the PCI and AGP bus speeds will increase at the same time that you are increasing the speed of the front side bus. This can cause problems, since some components, such as the hard drive controller, which is connected to the PCI bus, don't tolerate being overclocked. Usually, a new "divider" kicks in after you have increased the front side bus speed up to the next speed that processors are officially designated to run on (such as 66, 100, 133, 166, or 200 MHz), but for those in between speeds, the PCI and AGP buses still uses the divider appropriate for a slower front side bus. By divider, I mean that the PCI and the AGP bus speeds are calculated as fractions of the front side bus speed. This is why the processors that I previously mentioned as examples of those that I have overclocked all went from one typically used front side bus speed to another typical speed. The motherboards that I was using did not have the ability to lock the PCI and AGP bus speeds, and this was the only way to make sure that these speeds remained within spec, while increasing the speed of the front side bus.
Another feature to look for is the ability to increase the speed of the front side bus in relatively small increments, which is pretty much always the case now with motherboards, except perhaps for those designed specifically for use by a large system integrator, such as Compaq. It also is helpful to be able to increase the voltage to the processor by some small increments. It is not unusual for an overclocked processor to be unstable at the original, default voltage setting, but a small increase will allow it to function at the faster speed. In fact, you sometimes see processors that the manufacturer has designated as their fastest models are specified to use a bit more voltage than the lower clocked versions. The key is to not get carried away with this, because increasing the voltage means that your processor is going to be subjected to even more strain and more heat. Ideally, I try to stay within the range of what the fastest versions of this processor use as their default voltage. If one or two tiny voltage bumps won't bring about stable operation, than you probably are asking too much of this particular processor.
Since you will be running your processor on a front side bus that is faster than it was originally meant to be run on, you'll want to have memory that can run on a memory bus clocked at the same speed as the new front side bus speed. This is referred to as running the memory synchronously with the processor, which simply means that the memory bus and the front side bus speeds are the same. Although many motherboards will let you run these two buses asynchronously, you generally get the best memory performance when these two buses are in step with each other.
If you are starting out with a processor that originally is meant for a low speed front side bus, finding memory that is specified to run at the same speed as the overclocked bus isn't much of a problem. However, if you are starting out with a processor that is already running on a fast front side bus, finding memory (and a motherboard) that can manage these very high bus speeds is going to require a some careful research. For example, those who would like to overclock one of the newer Pentium 4s are starting from default front side bus and memory bus speeds of 200 MHz. This means that to overclock a Pentium 4 2.4C (12 x 200 MHz) up to the speed of a 3.2 GHz processor they'll need a motherboard that can provide a 267 MHz front side bus (12 x 267 MHz = 3200 MHz) and memory that can run at DDR 534 speeds (2 x 267 MHz = 534 MHz, roughly the equivalent of PC 4200 Memory). Although there are indeed motherboards and memory modules that are capable of tolerating such high speeds, you'll have to looking for these capabilities from the start, when selecting your components. (The reason someone might still want to try to overclock on of these 800 MHz processors, despite the handicap of starting from such a high default bus speed, is that that they all have hyperthreading capabilities. Most of the P4's running on the 533 MHz bus don't have hyperthreading capabilities, except for the highest speed model.)
Overclocking Video Cards
When thinking about overclocking a video card, you can make use of many of the same principles that were discussed with regard to CPUs. This means looking at a graphic chipmaker's product line to see if there aren't lower end products that use essentially the same GPU as the higher end products, just one that is clocked at a lower speed. Be aware that some of the lower end GPUs have been crippled in ways other than just lowering their clock speed. They may have less "pipelines" for processing video data or other features may have been disabled. While these chips still can make good overclocking candidates, they will never be the equal of a chip that has the full set of video features enabled. In some ways, this is similar to how the Celeron and the Pentium lines are related to each other or how the Athlon and the Duron lines are related. Although in many respects they are the same chip, the value version of the chip has been hobbled with disabled features and less capabilities than the full version of the chip.
The proof of the concept of that some lower end graphics chips make good overclocking candidates can be found in the fact that some video card manufacturers offer their own overclocked versions. The Geforce Ti4200 was a popular one for this, as a number of manufacturers offered Ti4200s clocked at speeds essentially the same as the Ti4400. To my way of thinking, when a big name manufacturer, like Asus, feels comfortable overclocking their Ti4200s, even a modest amount, this speaks volumes about how much confidence they have what this this chip's real capabilities are.
One thing to be mindful of when overclocking a video card is that overclocking just the GPU may not result in much of a performance benefit, if you can't also overclock the video memory. Typically, the default memory and GPU speeds are in a good balance with each other. In other words, the memory is not a bottleneck that interferes with the performance of the video card, except perhaps at extremely high resolutions. However, If you successfully overclock the GPU, it's going to need more memory bandwidth. Unless you can also speed up the memory, this will hold back the card's overall performance.
Unfortunately, it is hard to tell which memory is capable of being overclocked. More often than not, the speed rating that the memory chip carries is pretty representative of it's actual abilities. So, your best bet will be to look for video cards that make use of memory chips that actually are rated faster than the speeds that they are being run at. Otherwise, you'll just have to read the reviews of a particular make and model and see what sort of experiences others are having when it comes to overclocking particular video cards, paying attention not just to the GPU speeds that they are achieving but to the memory speeds, as well.
Keep in mind what was said previously about heat; you very well may find it necessary to replace the stock video card heatsink with something more substantial, as you increase the GPU's clock speed. If the memory chips don't laready have heatsinks, you can attach some, which may help a bit with the memory speeds that you can reach.
Testing for Stability
Finally, before you can declare your overclocking experiment a success, you need to test your setup for stability. Better to find out earlier, instead of later, whether things are really as good as they appear at first glance. For stressing the processor and the memory system (which includes both the system memory and the cache memory integrated into the processor) I like to use Prime95 and MemTest86. Prime95 is a distributed computing project that searches for extremely large prime numbers. The client software can be used to run some tests ("torture test") that put your system to work crunching numbers for hours at a time. It monitors the results for miscalculations. Prime95 primarily is a processor test, but it secondarily will pick up on some memory problems, as well. MemTest86 is primarily a system memory and processor cache test. It can also be set to run for hours at a time, which increases the likelihood of it picking up on transient errors. Between the two, these tests can put your system through a pretty good work out, and after running each overnight, you can have some confidence that your system is behaving as it should.
For testing your video card, you'll need something with a lot of 3D graphics in it. A benchmark such as 3DMark2001SE has the ability to be put into a looping mode, which will give you a better chance of finding problems than the usual few minutes that it takes to run the benchmark for a score. Of course, playing your favorite 3D graphics game for a few hours is also good a test of system stability, especially if the game demands a lot from your hardware. Battlezone II is still my favorite for this purpose. It is a 3D, first person perspective game that is both CPU and memory intensive. You can also "record" a game and play it back. This process adds another level of difficulty to running the game, since during playback it actually recreates the original session. Any missteps in this process will keep the recording from playing back properly.
Well, this brief accounting of some basic ideas about overclocking has already turned out longer than I expected, so I'll leave off at this point. Hopefully, it will help you to decide whether overclocking components is something that you want to try, and it will increase your chances of a good outcome, if you do decide to overclock.
Back to Sequoyah Computer.
January 24, 2004