This writeup was inspired by the Jargon File's ignorant and inaccurate definition, no offense meant. Considering that many people are capable of getting 50% speed boosts out of their system with inexpensive, non-exotic cooling systems, some myths about overclocking may need dispelling here. As always, /msg me if you disagree about something.
What is overclocking?
Overclocking a piece of hardware
in a computer is making it run at faster speeds than it normally runs at. The term is most often applied to processor
s, as these are the part of the system that is most commonly overclocked, but many other parts of a modern PC can run above spec without endangering the system itself, including the video card
(by way of frontside bus
overclocks, more on this later), and more.
The answer to this question is different for different people. The most common one is saving money
; why buy a $200 CPU
when you can buy a $60 one and about $25-30 of cooling supplies and get the same thing? Others overclock just for the fun of it, as something to do when they're bored. Many like both.
But isn't overclocking dangerous?
If you know what you're doing, no. That overclocking can corrupt hard drives, melt systems, kick little puppies
, etc. is true in the same way that the statement "you could destroy your house with a single potato
" is true. In other words, it's certainly true provided that you are either a dedicated arson
ist or display an alarmingly low level of common sense
. If you take the proper precautions, overclocking is not remotely risky
. Back in the Days of Yore
, it was, but we're past the point where we need to play with crystals now. There are several things that one should always keep in mind when overclocking to ensure complete safety of the parts of the computer:
So wait, if I don't have adequate cooling or whatever, isn't it risky?
- The speed of a CPU is equal to the frontside bus (FSB) multiplied by a number called the multiplier. Raising the frontside bus speed effectively widens the "pipe" connecting the CPU to the rest of the system as well as increasing the CPU's clock speed, so it increases performance more; however, it can be dangerous on older motherboards without PCI/AGP bus clock locks.(*) You see, the AGP and PCI buses run at a set clock speed equal to the FSB divided by a number called the AGP divider (for the AGP bus) or the PCI divider (for the PCI bus). This number should always be 66MHz for the AGP bus and 33MHz for the PCI bus, but if the divider is too low (such as with 143/4 for the PCI bus), you will run the devices on that bus out of spec. Running the AGP bus out of spec is stupid and leads to little or no performance gain, as well as high risk. Running PCI devices out of spec is a terrible idea and highly dangerous, especially since the hard drives are linked to the PCI bus clock. You do not want to overclock a hard drive.
The best thing to do with these motherboards is to either raise the FSB to a multiple of 100/3 (33.3333...), provided that the motherboard is capable of handling a high FSB (if the manual says it's not, then the proper divider is missing, so don't do it), or to alter the multiplier. Raising the FSB by such a drastic amount works with only a few processors; Athlons and Athlon XPs, curiously enough, can take any bus speed up to 200MHz provided the motherboard itself can handle it. However, even if it fails, the system is in no danger; you simply reset the CMOS. If you can't raise the FSB by such a large amount, multiplier overclocking is your other safe alternative, as it has no effect on anything but the clock speed of the CPU. Many CPUs (including all Pentium 4s) are multiplier-locked, simply because the multiplier offers such an easy way to overclock the processor. Some of these, such as the AMD Athlon "Thunderbird," can be unlocked without too much trouble. (Instructions on unlocking Thunderbirds can be found elsewhere.)
- When you overclock something, it will put out more heat. Be prepared for this by ensuring that system cooling is adequate. Contrary to unpopular belief, "adequate" does not mean you need some sort of exotic freezing system or annoying banshee fans, it simply means making sure that your cooling system will not lead to stifling conditions inside the case. So long as you take care to ensure reasonably unobstructed airflow inside the case, you should be fine. The individual part that is being overclocked may need additional cooling too, but it may not. Most video cards are perfectly fine with the cooling systems they come with and don't need anything fancy to overclock. CPUs tend to be the most worrisome in terms of heat, but if you're willing to take half an hour or so to clear airflow in the system you should have no trouble achieving moderate overclocks with low-noise air coolers. You don't need to break the bank, either; a Speeze FalconRock II ($9) heatsink, a case fan to sit on top of it in case the normal near-silent fan that comes with the FalconRock II doesn't push enough air ($5), and maybe two case fans for the case ($10). That's $24 total-- hardly an enormous sum. Your average balanced low-noise case fan will offer about 32CFM of airflow and 30dBa in terms of noise.
- One common trick to get more out of a part is to put more voltage into it. However, too much voltage will drastically decrease the component's lifespan, and increased voltage results in increased heat. As a rule, you should not give a part more than ~112% of the voltage that it is meant to handle. A 12% increase in voltage will still decrease lifespan, but not by any remotely significant amount. Often higher-end CPUs that use the same core as the lower-ends have increased voltage; in this case, there's a little less headroom at the high end. The sole exception is the Thoroughbred-B core, which was designed for the higher-end Athlon XPs and was actually ramped DOWN for the low-end ones, meaning that you could put about 115% of the stock voltage into a low-end Thoroughbred-B and be safe.
No. If you don't have adequate cooling or the part isn't up to it, the overclock will merely fail. If your system doesn't kill itself running at stock speeds, it won't kill itself overclocked either. If your cooling, power supply, or other parts aren't up to the job, the system will either be unstable or fail to boot. Either way, nothing will occur that can't be solved by just putting things back the way they were, and you certainly won't fry anything.
Aren't overclocked systems unstable?
This is a common misconception. Very few overclocked systems are unstable; as Benjamin Franklin said, "those who sacrifice stability for performance deserve neither.
" Most people who overclock their systems run "burn in" tests such as Prime95 (CPU), 3DMark2001SE and 2003 (video card), and Memtest86 (memory) for as long as 24-48 hours to ensure that the system isn't unstable at its new speeds. Amusingly, this means that most overclocked systems are more stable than your Average Joe
's new Dell
. Be aware that if you have a flaky
power supply your system could well be unstable when overclocking. For higher voltages (again, up to ~112% of normal voltage), a quality power supply manufactured by a company such as Antec
is a good idea. In fairness, though, cheap power supplies often die in plain old stock-setting systems too. With power supplies, the extra ~$20US
is worth the peace of mind. If you don't think so, you haven't seen an exploding generic power supply before.(**)
Is there any real-world performance gain associated with overclocking?
Yes, definitely. Yet another popular misconception about overclocking is that you get small increases in performance that don't ever matter in the real world. The fact is, if you choose your components carefully, you can get CPU overclocks that will give you enormous real-world performance increases. (Video card
s tend to be less overclockable, and usually "only" provide performance gains of 10-25%, but when you consider this is a free performance gain since they already have coolers attached, you can see why it's so popular.) In case you're skeptical, there are a few CPUs worth mentioning. Note that I didn't include the 50% overclocks of the old Duron
s (the Athlon
s were better anyway, so why bother?), the Athlon XP 2500+ and its given no-voltage-adjustment 20% overclock, the respectable 20%-30% overclocks non-2.4GHz P4C
s achieve, the 35% overclocks of many AIUHB Thoroughbred-Bs, and so on. This is just a small list with some of the most notable
How can parts run so far above spec?
- Celeron 300A - This now-famous processor always overclocked its FSB from 66MHz to 100MHz, resulting in a clock speed of 4.5x100=450MHz. What's more, unlike the Pentium II (supposedly the much faster of the two), its L2 cache was on-die, meaning it was actually as fast as a 450MHz Pentium II when overclocked despite having a quarter of the L2 cache.(***)
- Athlon 1.0GHz AXIA - This is a rather famous processor too. The AXIA stepping(`) was usually seen in 1.4GHz Athlons; if a CPU was an AXIA, it was guaranteed to run at 1.4GHz. However, increased demand for the 1.0GHz chips and the conversion of production to newer steppings resulted in a whole bunch of AXIA stepping 1.0GHz Athlons, and nearly all ran perfectly at 1.4GHz with no voltage increases at all.
- Pentium 4 1.6A/1.8A - These two Pentium 4s had an FSB of 100MHz (400MHz effective, the Northwood core has a quad-pumped frontside bus), but were basically underclocked 533MHz FSB Pentium 4s. Both of the two had no trouble whatsoever running at 133x16=2.133GHz and 133x18=2.4GHz respectively, and like the Athlon 1.0 AXIA and Celeron 300A, they didn't even need any additional juice to get there.
- Athlon XP 1700+ JIUHB - The JIUHB-stepping 1700+ was famously overclockable, with overclocks of about 2.2GHz considered average even for early JIUHBs. This was a 1.466GHz processor, mind you, meaning that these people were achieving 50% overclocks. Reports of Week 12 JIUHBs doing 2.4GHz and above without putting a dangerous amount of voltage into the processor were quite common. This was, by the way, not a very expensive processor. You could usually get them for $50US including shipping. What's more, the Thoroughbred-Bs at this time were unlocked, meaning that you could play with the multipliers.
- Pentium 4 2.4C - This chip, when combined with the Intel 865PE chipset or something comparable, is capable of overclocking to about 3.2GHz on average with the stock cooler. The 33% overclock is entirely frontside bus, though performance scaling actually indicates multiplier overclocking, which is odd. While not as remarkable as, say, the 1700+ JIUHB, it's considerably faster. The 2.6C was also regularly capable of overclocks on this scale.
- Athlon XP-M 2500+ - "Wait, XP-M?" you ask? Or maybe you don't. Probably you don't, but I'll explain anyway. The XP-M was actually a mobile processor, and as such, it ran at lower voltages than the desktop 2500+ (1.45v instead of 1.65v) even though it ran at roughly the same clock speed. Now, how did AMD manage that? The answer is cherry-picking, or simply sorting out the chips that could handle this. Now, CPUs that can use significantly less power than normal while remaining at normal operating speed are bound to be good overclockers; they're the cream of the crop. Realizing that the XP-Ms were pretty much identical to the desktop XPs in every other way (they go just fine in desktop motherboards, they look the same), some intrepid person decided to see how well XP-Ms did in desktops. The answer was very, very well, and the people involved only kept it under wraps for a couple weeks before it exploded into the (relative) mainstream. These chips usually made it to 2.5~2.6GHz on air cooling. (Stock speed for these mobile 2500+es is 1.866GHz, so that's about 33 to 40%.)
- Socket 939 144/146/148 - That's right, Opterons. AMD's enterprise-grade chips. Normally they're Socket 940, but AMD decided to start selling some budget-range (relatively speaking) 1-series Opterons in Socket 939 so that they could be used on cheaper, more widely available Socket 939 boards. These were basically Athlon 64s, really, only they were cherry-picked units. Why are enterprise-grade cores always chosen from nearer the middle of the wafer? . But these were some rather nice chips, since they all had 1MB of L2 cache (many desktop Athlon 64s had 512KB of L2 cache), sold for about the same price as equivalently priced A64s (well, until, y'know, everyone ran out of stock and prices went through the roof), and went very, very far (40-50% boosts were fairly commonplace, even on air cooling). There were dual-core equivalents too (165/170/175, Socket 939), but they weren't quite as spectacular (remember, a dual core processor can not overclock faster than whichever core runs slowest) with regards to overclocking.
Usually, processors are speed binned
, which means that they're tested to see how fast they can run at and are then sold at that speed. So why all the big exceptions? Well, some, like the Celeron 300A and Pentium 4 1.6A/1.8A, were effectively underclocked to become budget processors to begin with. Others, like the Thunderbird 1.0 AXIA, simply weren't sold as their faster, higher-priced brothers because of a lack of demand for the high-end chips. The low to midrange makes up the vast majority of the market, meaning that many slower processors perfectly capable of being sold as higher-end ones aren't, simply because they wouldn't get sold then.
With video cards, RAM, and the like, the story is a little different. While there are usually many clock speed "steps" with CPUs (e.g., Athlon 1.0GHz, Athlon 1.1GHz, Athlon 1.2GHz, etc.), video cards rarely have more than two or three of these steps, and most memory manufacturers only have two versions of memory at a given speed. Also, whereas there are only two big CPU companies right now (AMD and Intel), there are many companies that produce video cards (ASUS, Sapphire, FIC, Gigabyte, Tyan, and so on) and memory (Corsair, Crucial, Kingston, OCZ, GeIL, et cetera). As a result, reviews of these products usually also rate overclockability, and the manufacturers often tweak their designs to make sure that the reviewers will get large overclocks out of the items. By offering highly overclockable parts, manufacturers can distinguish themselves. Also, in video cards these days, a failure of the core
to overclock more than 10% indicates that the design is being pushed to its limit, and that the card should probably be avoided. Clearly, this is the last thing a manufacturer wants.
* Motherboards based on the nForce2, Intel 865, and Intel 875 chipsets have the AGP and PCI bus clocks locked by default.
** Yes, this happens, with overclocked and non-overclocked systems alike. It's not terribly common; usually they just fail in slightly less spectacular ways, like going "piff" and shooting out a little cloud of smoke. Some of the nastier ones can fry the motherboard. Remember, there's nothing between the power supply and the system for protection.
*** I'm sure 150MHz doesn't seem like much to you people these days with your fancypants multiple-GHz processors, but this was a 50% overclock. By way of comparison, imagine if an inexpensive "budget" 2GHz processor existed that always overclocked to 3GHz with its retail cooler with no trouble whatsoever. Now imagine it was just as fast as the far more expensive one at the same speeds. See my point now? Ever since the 300A, the 50% overclock has been used as a milestone.
` Wondering what a stepping is? As a given process for manufacturing a processor becomes more refined, manufacturers use "steppings" to denote more recent processors capable of achieving higher speeds. Think of them as very, very minor processor revisions.