As a PC enthusiast, my desire to overclock is almost compulsive. Pushing clock speeds must be hard-wired into my DNA, because I can’t actually remember the last time my personal workstation ran at stock speeds.
Overclocking isn’t for everyone, though; it can be time consuming and frustrating, and it will almost certainly void your warranty. But there’s a unique sense of satisfaction to be gained from pushing your hardware to its limits and achieving greater performance than you’ve actually paid for. For enthusiasts, many of whom have a borderline obsession with deriving the best bang for their buck, overclocking’s potential is simply too hard to resist.
To help the uninitiated get their feet wet with overclocking, we’ve whipped up a handy guide covering the basics. This is by no means an attempt to document every technique associated with turning up your system’s clock speeds, but it should be a good place to start for newbies looking for their first taste. We’ve chosen to focus our examples and advice on Intel’s Core 2 processors, since they’re a particularly popular choice right now, but many of the basic principles we’re exploring apply to any chip you might want to overclock, including AMD CPUs.
What is overclocking?
Simply put, overclocking refers to running a system component at higher clock speeds than are specified by the manufacturer. At first blush, the possibility of overclocking seems counter-intuitiveif a given chip were capable of running at higher speeds, wouldn’t the manufacturer sell it as a higher speed grade and reap additional revenue? The answer is a simple one, but it depends on a basic understanding of how chips are fabricated and sorted.
Chip fabrication produces large wafers containing hundreds if not thousands of individual chips. These wafers are sliced to separate individual dies, which are then tested to determine which of the manufacturer’s offered speed grades they can reach. Some chips are capable of higher speeds than others, and they’re sorted accordingly. This process is referred to as binning.
There’s considerably less demand for faster chips than for slower ones, though. The Core 2 Extreme QX9650 may be the fastest CPU Intel can produce, but with street prices hovering around $1200, it costs quite a bit more than most folks are willing to spend on a CPU. The Core 2 Quad Q6600, which sells for less than $300, is in much higher demand because it fits within the budget of a greater number of consumers. And demand for low-end chips is even greater still.
Chipmakers often find themselves in a position where the vast majority of the chips they produce are capable of running at higher clock speeds, since all chips of a particular vintage are produced in the same basic way. So chipmakers end up designating faster chips as lower speed grades in order to satisfy market demand. This practice is of particular interest to overclockers because it results in inexpensive chips with “free” overclocking headroom that’s easy to exploit. That’s the magic of binning: it’s often quite generous. A great many of the CPUs sold these days, especially the low-end and mid-range models, come with some built-in headroom.
Overclocking can do much more than exploit a chip’s inherent headroom, though. It’s also possible to push chips far beyond speeds offered by even the most expensive retail products. Such overclocking endeavors usually require more extreme measures, such as extravagant cooling solutions, so they’re a little beyond the scope of what most folks will want to tackle.
What you need
The most important ingredient in any overclocking endeavor is a good chip. If you’re looking to exploit the “free” overclocking headroom made possible by binning, you’re best off looking at lower speed grades. If you’re after the maximum overclock, you’ll probably want to pick the number of cores and the amount of cache that you want, and then select the lowest speed grade available with those characteristics. If you have a choice between chips with different front-side bus speeds, it’s probably best to pick the chip with the lower default bus speed. A slower front-side bus can make life easier for the motherboard, and you may even be able to overclock the processor without pushing the board beyond its specifications.
Overclocking forums are also rife with discussions of specific CPU steppings and batch numbers that have higher success rates than others. If you’re willing to do a little researchand if you can coax retailers into giving you more detailed information on chips they have in stockyou can increase your chances of success. Gathering stepping and batch information is particularly useful if you intend to push clock speeds well beyond any binning freebies.
Of course, success is never guaranteed with overclocking. Your mileage will amost certainly vary, and you might even end up with a complete dud incapable of running more than a few MHz faster than its stock speed.
Just because we’re focused on processor overclocking doesn’t mean that other system components aren’t important. A system’s motherboard, cooling system, power supply, and even memory can affect the success of an overclocking attempt. These don’t necessarily need to be expensive high-end partsthat would defeat much of the value proposition behind overclockingbut you’ll be better off with quality components from reputable manufacturers.
On the motherboard front, you want to ensure that the BIOS has ample overclocking options, including the ability to manipulate bus speeds and system voltages. The more control we have over system variables, the more freedom we’ll have to tweak settings carefully in pursuit of higher clock speeds. Motherboard cooling becomes more important when you turn up clock speeds, as well. You don’t need a mess of heatpipes snaking every which way on the board, but try to stay away from boards with tiny chipset coolers that don’t offer much surface area to dissipate heat.
Depending on how far you intend to push clock speeds, you may also want to consider beefing up your system’s CPU cooling. Overclocked chips tend to run hotter than those at stock speeds, particularly when you start increasing the CPU voltage, and you don’t want a stock cooler holding your system back. Aftermarket coolers designed for overclocking feature significantly more surface area than the stock coolers AMD and Intel bundle with their processors. Aftermarket coolers also tend to have much larger fans to generate more airflow, often while making less noise. Decent coolers can be had for as little as $30, so they won’t put a big dent in your budget.
We always recommend that users spend a little extra to get a quality power supply for their systems, and this goes double if you want to overclock. Our concern here isn’t getting gobs of extra wattage, but ensuring that the PSU delivers clean power to the system.
Fancy memory isn’t always necessary if you’re looking to overclock a processor, but DIMMs rated for operation at higher frequencies can give you a little more freedom when playing with clock speeds. Memory module manufacturers often guarantee their products to run at higher clock frequencies, even if those speeds aren’t officially endorsed by the JEDEC standards body that governs system memory.
Even more important than individual component choices is having a completely stable system before you dive into overclocking. If you’re building a new system from scratch, stress test it at stock speeds to ensure that everything is working properly. The last thing you want is to burn an afternoon trying in vain to overclock a system hampered by a faulty component that isn’t even stable at stock speeds.
The obligatory warning
Overclocking will probably void your warranty, and it has the potential to damage not only the hardware being overclocked, but other system components, as well. This is where The Tech Report absolves itself of any responsibility for damaged hardware, voided warranties, puffs of magic smoke, core meltdowns, and bruised egos that may result from unsuccessful overclocking attempts. Or, heck, even successful ones.
We should also warn you that this guide covers overclocking through the motherboard BIOS. If you’re not comfortable poking around in the BIOS, you probably shouldn’t be overclocking in the first place.
Finally, before you begin overclocking your CPU, you should start by making a backup of any important data on your system. You may even want to consider using a disk imaging program like Symantec Ghost to make a complete image of your boot partition. We’ve seen more than one OS installation rendered unbootable by file corruption caused by an unstable processor in the midst of an overclocking attempt. The trial-and-error process of seeking a stable overclocked configuration necessarily involves some risk on this front, so make provisions ahead of time.
Basic clock theory
Overclocking starts and ends with clock speeds, and you’ll need to know about several clocks in a system in order to understand how things work. For Core 2 processors, the first clock of note belongs to the front-side bus. The front-side bus, or FSB, links the CPU to the rest of the system, including main memory, storage, graphics, and peripherals. Like several of the key elements in a modern PC, the FSB is a little tricky, because its effective data rate differs from its base clock speed. For example, the Core 2 Duo E6750 is advertised as having a 1333MHz front-side bus. That’s the effective data rate, but the base clock is 333MHz. You’ve got to multiply by four to get the effective bus speed from the base FSB clock. Intel uses the term “quad-pumped” to describe the FSB’s nature. Overclockers tend to refer to the base FSB clock rather than the effective speed, because the base clock frequency is usually the value shown in the system BIOS.
The FSB is notable because processor clock speeds are determined by the product of the base FSB clock and the CPU multiplier. For example, the Core 2 Duo E6750 uses an 8X multiplier with a 333MHz base clock, yielding a processor speed of 2.67GHz on an effective 1333MHz front-side bus. (For AMD chips, the CPU multiplier is applied to the HyperTransport base clock.)
The easiest way to overclock a processor would be to increase the value of the CPU multiplier. However, with Core 2 processors, doing so is only possible with Extreme Edition chips that are far too expensive to be reasonable overclocking candidates for most enthusiasts. Intel prevents users from adjusting the CPU multiplier upward on its other chips. AMD does the same thing with its processors; only its Black Edition and FX chips allow the CPU multiplier to be increased.
Interestingly, it’s actually possible to decrease the CPU multiplier for Core 2 processors. That won’t help with processor overclocking, though, leaving us with no choice but to tackle the front-side bus. This is where it helps to have a CPU whose native front-side bus speed is lower than the maximum FSB supported by the motherboard. Combine a Core 2 that has a native 1066MHz front-side bus with a P35 or X38 Express-based mobo that supports FSB speeds up to 1333MHz, and you have 266MHz of quad-pumped overclocking headroom right out of the box.
For the sake of illustration, we’re going to be overclocking a Core 2 Duo E6750 processor on a Gigabyte GA-X38-DQ6 motherboard as we step through the key system clocks. The menu above is typical of many high-end motherboards. Gigabyte labels the base FSB clock as the CPU Host Frequency. The default value for this CPU is 333MHz, but we can change it by setting CPU Host Clock Control to “Enabled” and simply keying in the value we wish to use.
Oh, and let’s hope you’re using a home-built PC with a decent motherboard. Don’t expect to find these menus in the BIOS of your average Dell, folks.
I generally like to overclock in small increments, increasing the front-side bus base clock in 10MHz steps. Overclocking is usually a very iterative process: set the speed a little higher, test, and repeat. However, for this example, we’ve skipped ahead a little bit, because we know this chip has lots of headroom. If you do the math, an 8X multiplier on a 390MHz bus will get us a CPU clock frequency of 3120MHz or 3.12GHz. That’s… quite healthy. We’ll stick with that for the time being.
Of course, the CPU isn’t the only system element that bases its clock speed on the front-side bus. The PCI Express, PCI, and memory subsystems also typically derive their clock frequencies from the FSB clock. We’ll want to ensure that overclocking the front-side bus doesn’t inadvertently increase the clock speeds of any of these other subsystems, lest they hold back our overclocking attempt. We can manage these auxiliary clock speeds in one of two ways.
The easiest means of keeping these clocks in check is locking them down at a given clock speed, regardless of the FSB clock. Any motherboard with a good suite of overclocking tools should include such an option for the PCI and PCIe subsystems. Sometimes, these options are even enabled by default. In the example below, we’re locking down the PCI Express clock at 100MHz, which is its proper default speed.
A much bigger concern is the memory clock, which most often cannot be locked. Instead, the memory clock’s relationship with the front-side bus is usually governed by a single valuea multiplier, divider, or ratio, depending on the BIOS. By manipulating this value, we can keep the memory speed from exceeding the rated speed of our system’s memory modules, even as we increase the front-side bus speed. That’s exactly what we’ll want to do during our initial CPU overclocking attempts: adjust the memory ratio in order to keep the memory clock at or below our DIMMs’ stock speeds. Later on, if you wish, this memory ratio can also be used to overclock your memory, if fiddling with CPU clock speeds isn’t enough to satiate your appetite.
In our example case, we’re using 800MHz DDR2 memory along with our 390MHz FSB, so we’ve set the system memory multiplier to 2X the FSB clock. That should give us a memory clock of 780MHza little slower than stock for these DIMMs, so we’re sure overclocked RAM won’t be a source of instability. There’s no real harm in running our memory at this somewhat odd speed, either.
Setting this board’s memory ratio involves a multiplier much like the CPU’s, but it seems like every motherboard maker does it a little differently. Each chipset has its own quirks, too, like the MCH strapping on this Intel chipset. The ratios can be confusing and may not always behave like you’d first expect, but a good mobo will help. The X38-DQ6’s BIOS even shows us the resulting memory frequency of 780MHz just below the multiplier option.
Like Intel’s front-side bus, DDR memory transfers data multiple times per clock cycle. In this case, DDR’s double data rate means that we multiply by a factor of two. A 400MHz memory clock corresponds to an effective memory data rate of 800MHz. (This 2:1 ratio between the base memory clock and the data rate holds true for all DDR memory types, including DDR2 and DDR3 modules.) Unlike the FSB, however, motherboard BIOSes usually express all values using the effective memory clock, so you shouldn’t have to worry about doing the math.
Putting the theory into practice
Now that we have explored the basics of modifying your system’s clock speeds, let’s talk about putting that theory into practice. Like we said earlier, overclocking is generally an iterative process. You’ll want to start with a relatively modest overclock, save the BIOS settings, and then reboot the system to see if it POSTs and makes it all the way to your operating system. If it does, you’ll want to confirm that the BIOS changes you made have produced the intended result. Fire up CPU-Z to confirm that the bus speeds you’ve set in the BIOS are in effect and yielding the correct processor clock speed. CPU-Z can also be used to check on the CPU voltage, which will become more important in a moment.
However, we’ve found that CPU-Z’s voltage readouts aren’t always quite correct. To get more precise insight into your system’s health, you’ll want to install your mobo maker’s system monitoring application, which is usually included on your motherboard’s driver CD. These apps often have flashy or frustrating interfaces, but since they come from the board maker, they should give you correct readings for system voltages, clock speeds, and temperatures. Here’s a look at Gigabyte’s Easy Tune application on our test system.
If you poke around a little in Easy Tune, you’ll find a single page that shows voltages, fan speeds, and temperatures all at once.
We’ll want to leave this window open to keep an eye on processor temperatures. Core 2 chips start throttling around 72° Celsius, and I wouldn’t want to exceed the mid 60s under load.
Speaking of load, it’s time to fire up a stability test to ensure that everything is working smoothly. We like to combine a couple of different applicationssomething like Prime95 to hit the processor and memory in addition to a graphics demo to stress the video cardto ensure that a system is stable. Prime95 will toss out computational errors or even hang your system if the CPU is overclocked too far.
If you’re running a multi-core processor, be sure your stress testing produces a full load on all your processor cores. You can run extra instances of Prime95 by copying the program executable to another directory and launching it from there. The processor affinity for each Prime95 instance can then be set through the app’s advanced options.
If Prime95 runs stable for five minutes or so without triggering uncomfortable CPU temperatures, it’s probably time to move on to higher clock speeds. If Prime95 throws an error, the system locks or reboots, or you’re treated to Windows’ dreaded Blue Screen of Death, well, you’re going to need some additional help.
You’ll also need some additional magic if you can’t even get that far. As your processor’s clock speed rises with the front-side bus speed, you’ll eventually hit a wall where the system will either refuse to POST, refuse to boot your operating system, or otherwise become unstable. You’ve now exceeded the capabilities of your system, at least with stock voltages.
If you can’t get the system to boot far enough to allow you to modify the BIOS, you’ll probably start to sweat a little, like I do, as the worry begins. That’s never fun. But worry not, because salvation is but a simple step away. You can clear out your overclocked BIOS settings easily. Just find the “clear CMOS” header on your motherboard and slide a jumper onto its two prongs for five seconds or so in order to reset the BIOS to its fail-safe default values. This header is usually clearly marked and located somewhere near the CMOS battery.
Our example board has just two prongs on its header, but many boards have three prongs, with a jumper pre-installed on pins two and three. Moving the jumper to pins one and two will usually do the trick in arrangements like that. However, you’ll want to consult your motherboard’s manual for advice on your particular board’s exact CMOS reset procedures.
After the CMOS has been cleared, the system should happily POST again. You can then go back to your last stable BIOS settings and keep tweaking, if you wish. Here you have the option of calling it a day or pushing on with a little extra juice.
Better living through higher voltage
That’s right, after you’ve hit a wall, the next step in your overclocking odyssey probably involves voltage.
Increasing the voltage applied to a given system component usually helps it run at higher speeds. This is particularly true of CPUs, but be careful not to get carried away. Raising processor voltages will make a chip run hotter, requiring additional cooling. More voltage isn’t always better, either. Unless you’re using extreme liquid or sub-zero cooling, most chips tend not to benefit from more than a couple of extra tenths of a volt. Start out with small voltage increase increments and initially confine your voltage fiddling to the processor. Once you’ve applied a little extra voltage through the BIOS, shoot for the last front-side bus speed that failed to see if it’s stable. If system stability gets worse when higher voltages are applied, it’s time to beef up your cooling or back off to lower voltages.
When you reach a point where increasing the CPU voltage doesn’t result in higher stable clock speeds, turn your attention to chipset and front-side bus voltages. These can be increased, as well, and they tend to be most helpful at higher front-side bus speeds where you’re approaching the limits of the motherboard’s capabilities. In my experience, you’ll run out of CPU headroom long before your motherboard gives up, but it’s worth tweaking the mobo a bit just in case.
For the record, our example CPU eventually turned out to be quite happy at an astounding 3.64GHz at 1.3875V, using nothing more than a stock Intel air cooler. We stepped through at least 10 to 15 different clock speed and voltage increments in order to reach this speed, recording info as we went.
Be sure to document all the settings you used, your system temperatures, and the results of your stability testing along the way. This information will come in handy when you’ve determined the limits of your processor and motherboard. For some, the point of overclocking is to achieve the highest speed possible, at any cost. However, for others, it’s about reaching an optimal clock speed with the best blend of performance, power consumption, and temperaturethe sweet spot, if you will. This sweet spot may give you 90% of the highest stable clock speed your processor can achieve, but do so without the need to increase voltages or significantly impact temperatures. I’ll take that over bragging rights any day.
When you settle on a final configuration for your system, you should conduct a longer stability test, otherwise known as a burn-in. This test should peg your system at full utilization for several hours at the very least to ensure that everything is perfectly stable. It’s a good idea to log temperatures during this test to ensure that prolonged periods of heavy load don’t overwhelm your system’s cooling solution.
Successfully exploiting the “free” overclocking headroom available with most budget processors can act like a gateway drug, pulling you deeper into the obsessive underbelly of the overclocking world. And that’s not necessarily a bad thing. While most folks will probably want to limit their overclocking exploits, the more adventurous will find no shortage of options for further advancement.
We’ve only dealt with basic processor overclocking today, but if you’re willing to dole out some cash for a fancy water or sub-zero cooling system, you should be able to push your processor’s clock speed even higher. Other system components can be overclocked, as well. Overclocking your memory is probably the easiest to tackle next, since it can be done by manipulating bus dividers through the motherboard BIOS. For this, you’ll want some fancy memory modules rated for operation at higher speeds; budget or generic DIMMs tend not to fare well when pushed beyond their specifications.
Graphics cards are also prime candidates for overclocking, and you can push them to their limits without leaving the comfort of Windows. Nvidia’s nTune system utility, for example, allows users to set GPU core and memory clock speeds. nTune also has a built-in stress test that can be used to validate the stability of a given configuration. There’s no shortage of aftermarket cooling solutions for graphics cards, either.
Overclocking is almost a rite of passage for enthusiasts. Squeezing extra MHz from otherwise inexpensive processors speaks loud and clear to our drive to maximize the bang for our buck. If we can take advantage of market dynamics that force chip makers to bin processors fully capable of running at higher speeds to meet demand for low-end chips, then maybe we’ve stuck it to the man, as well. What’s not to like about that?
If we’ve whetted your appetite for more, be sure to check the Overclocking, Tweaking, & Cooling section of our forums for more discussion of all things overclocking.