The need for speed--the phrase made popular in the movie "Top Gun"--applies to every phase of life for those living on the bleeding edge. Both Palm and Windows Mobile devices have software available to overclock them to a certain limit. Overclocking your PC isn't quite that simple, but it has become increasingly easier. For some of us, our toys can never be fast enough. For those who, like Tanker Bob, fall into this category, I dedicate this post.
Disclaimer - Overclocking you PC will void warranties, could physically damage or destroy expensive components therein, cause your sheep to miscarry, and result in your dog getting unsightly acne. Be warned!!!
If only overclocking your PC were as easy as installing some software. In the "old days" and probably on a few remaining motherboards, overclocking involved setting physical jumpers and working with clock multipliers. Fortunately, these tasks have been simplified in newer motherboards with BIOS setups that allow these changes in the BIOS software itself. No need to provide a blood sacrifice anymore.
Don't panic--this won't be an electrical engineering lesson. There are a few things that overclockers must understand, though, about their machines in order to be successful.
Don't buck the system!
First, overclocking must be approached from a system-wide perspective. About the most expensive and least effective way of getting more performance out of your PC is to buy a faster CPU. That sounds counter-intuitive, but the vast majority of the time, simply adding more RAM will produce a much greater benefit--especially on Windows operating systems. A better graphics card generally provides the next-best payoff in performance gain depending on your usage pattern. The reason has to do with what the CPU does and how it gets data in and out. Without going into gory details, the external communication pipes, memory bus architecture, and its support chips all conspire to limit the effectiveness of your CPU. If you consider your PC as a cohesive system, you can save a lot of money in pointless "upgrades." Remember that a system is only as fast as its slowest component. I'm not saying that you won't see benefits from upgrading a CPU (although you might not), just that the increase in performance won't necessarily be commensurate with the upgrade cost.
In practice, that means carefully considering your motherboard, CPU, system memory, and graphics card together when putting together or buying a new system as well as when upgrading components. Don't spend your money on a fast CPU and then "save" some bucks with slow memory or motherboards with narrow busses and slow chipsets. Generally speaking, a medium-priced system with well-matched components will perform faster on real-life tasks than a top-line CPU on a cheap system. So, if you see two PCs on a retail outlet shelf that are identical except that one has a faster CPU, you must run the same tasks on both to see if the difference in performance is worth the difference in price. If there's a large price difference, chances are you won't see a commensurate performance difference unless the slower-CPU is being under-challenged by its system. In that case, the lower-priced machine may be the worse value!
Which leads us to benchmarks. CPU manufacturers go to great lengths to master the current benchmark tests. Some have even been known to fudge a bit. But don't be fooled by clock ratings or CPU benchmarks. Reputable sites like Tom's Hardware and Anandtech post these "raw" benchmarks, but they test using identical hardware and include system-wide benchmarks as well. These results usually carry names like "Workstation" or "Graphics Workstation" to describe the task being tested. Look at the task-oriented tests that best simulate the work that you predominantly perform on your PC. These better predict your user experience.
This same advice applies to graphics cards and hard disks. For example, the newest Serial ATA (SATA) interfaces have a theoretical bandwidth limit of 300 MB/s. However, few hard drives are physically capable of accessing data on their platters at anywhere near that speed. System-wide benchmarks will show little difference in data interface performance for the most common 7200 rpm hard drives. In other words, it's never rush hour on the data bus so the big highway is underused. Don't let isolated raw numbers eat your wallet--there's often a yawning chasm between theoretical throughput and real-world results.
Casey Jones, watch your speed...
Second, speed ratings on CPUs (and many memory modules) are often conservative. The clock speed stamped on a CPUs isn't necessarily that chip's limiting top speed. The speeds assigned to otherwise identical chips, for example the Intel Core 2 Duos E6600 at 2.4 GHz and E6700 at 2.66 GHz, are just a guarantee based on quality control. So, the rating doesn't mean that a given E6600 CPU won't run at, say, 2.66 GHz, but simply that Intel doesn't guarantee that it will. Intel does guarantee that the E6700 will run at the higher 2.66 GHz speed. The same idea applies to memory modules, though less universally.
What does this mean to you? Well, CPUs develop reputations for their ability or inability to be pushed beyond their rating. Some companies stamp their chips with a rating that sits pretty much at that chip's ability. This doesn't leave much room for variation in the consumer's motherboard clock chips, and can lead to problems in some machines without overclocking them. The fastest chips can also be poor overclockers as they approach their design limits. Other companies provide some headroom for their CPUs that allow for more variations in end-user motherboard oscillators. These latter chips attract the demand of overclockers, and it's what PDA overclockers live for.
So this section comes down to risk management. Should you buy the slower-rated CPU version and hope that you can overclock it to the level you desire? Or should you play it safe and go with the faster-rated chip? Tanker Bob cannot answer that question for you, but reading product reviews on places like NewEgg will give you a good feel for others' experiences with particular CPUs. Be warned that just like investments in the stock market, past history is no guarantee of future performance. Go in with your eyes open.
Gotta pay to play
The first rule of physics for dummies is that you never get something for nothing. (That works for economics as well.) It takes power to run everything--your dishwasher, your refrigerator, your car, and your CPU. Every CPU clock-speed rating carries a commensurate power level to attain that operation. Laptop and PDA users live this in a very personal way. The faster chips tend to eat power compared to slower chips of the same design. You can see this clearly in your car--it won't go as far on a tank of gas at 100 mph (167 kph) as it will at 60 mph (100 kph).
While battery life is an issue for laptops and PDAs, how does this affect your desktop PC? That question can be too hot to handle--literally. More power means more generated heat to dissipate. That's just basic physics. The only ways around it are to either develop a more efficient CPU that doesn't use as much power to generate the same speed (think Celeron), or to get the heat away from the CPU. That's why overclockers spend money on sophisticated heat sinks, heat pipes, fans, and liquid-based cooling systems. If you don't get the heat away from the CPU, it will literally cook like popcorn (sounds like it, too). That would be an expensive mistake.
Although retail CPUs usually come with stock cooling heat sinks and fans, these cooling devices may not be effective enough for overclocking. Well-established physics leads overclockers away from aluminum heat sink/fan combinations to more expensive copper ones that conduct heat far better. You also need a good thermal compound to put in between the CPU and the heat sink to efficiently draw the heat from the chip to the heat sink. Experienced overclockers live by Arctic Silver 5, which contains real micronized silver, sub-micron zinc oxide, aluminum oxide and boron nitride particles. Don't skimp here.
Motherboards designed to run a particular CPU can identify it and provide the correct power to run it at its rated speed. The CPU might actually run a bit faster with the same power input, but probably not much. Setting the CPU clock or multiplier faster will try to run the chip faster, but the CPU will lock up and quit if it doesn't have enough power to operate at the higher speed. Fortunately, overclockable motherboards provide an answer.
Physics and Tim the Toolman both tell us that if we want more power, we should crank up the juice. For PC systems, that means increasing the voltage input to the component we wish to overclock. All other things being equal, that will provide more power. BIOS' that support overclocking provide the ability to vary the voltage to individual components. On older motherboards, this can be done through jumpers on the board. Seems simple, eh?
Not so fast. Remeber that power = heat. So, more volts into the CPU or memory module will result in more--sometimes a lot more--heat generated that must be quickly and efficiently removed from the component. In addition to the cooling mentioned above, this requires some balance. Component life will be degraded if run too hot for too long. The idea is to choose just the right (i.e., lowest) voltage to run at the desired speed, which is a trial-and-error process. Running everything at maximum voltage will cause components to fail much sooner than normal operation, so don't go nuts.
Nor will a given CPU or memory module necessarily run faster with more voltage. Manufacturing flaws and design factors will limit its speed even if you give it enough power to heat your house (think Prescott). Overclocking is a delicate balancing act that requires patience, persistence, and caution. If you cook your chip, your system speed will be zero--not a desirable outcome.
One final note. Pretty much everything I said above also applies to overclocking graphics cards.
Turning the crank
As usual, Tanker Bob didn't just sit down to write about theory. Recall that Tanker Bob just built a system with an Intel Core 2 Duo E6600 2.4 GHz CPU, OCZ Dual Channel, 800MHz MHz DDR2 SDRAM, EVGA e-GeForce 8800GTS graphics card with 640 MB GDDR3 SDRAM, and an MSI P6N SLI Platinum motherboard. You should note that none of these components is top of the line, though some come close. However, these items match each other well on a high performance level. The MSI uses AMI BIOS and includes detailed overclocking capability for the brave of heart.
An important rule in experimentation says to only vary one parameter at a time. If you change two things and the system fails, how would you know which change caused the failure? That's a great theory, but often left on the shelf in the heat of the moment. A disciplined approach can save hours of pointless troubleshooting.
Tanker Bob started with the 2.4 GHz-stamped CPU. The AMI BIOS controls the CPU's speed through the front-side bus (FSB) setting. I increased the FSB from 1066 MHz to 1288 MHz, generating 2.91 GHz in the E6600. That's about a 20% overclock with no voltage increase. It also runs my E6600 faster than the $200 more expensive (on the street) E6700. Tanker Bob considered that gamble a prudent risk given the consumer overclocking feedback on the E6600.
One catch was that the BIOS automatically decreased my memory bus speed from 800 MHz to around 533 MHz. I had to set it separately back to 800 MHz. Hmmm.
Kununtu Linux ran perfectly for almost two weeks at that speed. So, being greedy, Tanker Bob bumped the FSB up to 1500 MHz which yielded 3.4 GHz in the CPU. This didn't work without bumping up the voltage, so the PC locked up. I didn't want to increase the CPU core voltage at this point, so I backed off to a 1400 MHz FSB yeiding a 3.17 GHz CPU. This worked fine for a while without increasing the CPU voltage. That's just over a 30% overclock. CPU temperature remained stable at about 29C with the Arctic Silver 5 and Zalman 9700 copper heat sink/fan setup. However, later I started to have random lockups after hours of running. Although I think that the RAM is the real issue with the lockups, I backed off to 1300 MHz FSB, 2.94 GHz CPU as a precaution until I find the real issue. That's still slightly over a 20% performance increase.
I didn't think that an 800 MHz memory bus did justice to a 3.17 GHz CPU, so overclocking the system memory came next. Through trial and error (lots of error!), I got the memory bus up to 900 MHz--about a 13% boost. However, although OCZ has a good reputation for overclocking, the SDRAM wouldn't overclock at all without increasing its input voltage. The OCZ comes rated for between 1.8 and 2.8 volts. A 900 MHz memory setting required 2.3 volts to the module. This leaves the possiblity of moving up to maybe 1000 MHz at max voltage, but I'm tired of restting my PC for now and will give the current settings a week or so to stabilize in the system. Overclocking the memory took much more trial-and-error effort than the CPU. In fact, I had another lockup several hours after I initially typed this. As a result, I backed the CPU off to 2.94 GHz and the memory back to 800 MHz. I will revisit when I have more time to play.
Speed on tap
To recap, Tanker Bob took a 2.4 GHz, 1066 FSB system and has them running at 2.94 GHz and 1300 MHz respectively. That's a pretty significant performance boost which is easily apparent during normal use. In fact, my E6600 is running as fast as a stock E6800 Core 2 Extreme which costs about $700 more than the E6600. Tanker Bob can hardly contain himself. YMMV.
Overclocking isn't for the faint of heart, nor for the foolhearty. Even with the new motherboards and BIOS designed for overclocking, it still remains as much art as science--therein lies the beauty and the challenge.
Powered by UBB.threads™ 6.5.5