CPU Fundamentals
CPUs also have settings in the BIOS that need to be set to overclock. They are FSB, multiplier and vCORE. Similar to RAM, the FSB setting will control the speed of the CPU. In addition, some Intel CPUs can have their multipliers adjusted (more later on this), although most Intel CPUs have their multipliers locked, and thus they cannot be multiplier adjusted. Almost all AMD CPUs can have their multipliers lowered, with some models having upwards adjustable multipliers as well.
The final setting common to both Intel and AMD systems is vCORE, or the voltage supplied by the motherboard to the CPU. Unlike vDIMM, most modern motherboards can easily supply more than enough vCORE to not only run your system at any reasonable, or unreasonable, overclock,;but also to fry your CPU quite nicely. Nothing smells failure better then burnt CPU, so be careful with vCORE. Before playing around with vCORE, be absolutely sure that you have consulted either the Intel or AMD sites' CPU specifications pages. Also, please be very aware that some information on those pages is incorrect. For example, last time I looked, the published vCORE specifications for the 3.4 P4C Northwood (Intel CPUID SL793) were incorrect. There are far too many types of CPUs out there for me to give you proper guidelines on how much vCORE is right for your system. Just remember, too much vCORE and you will probably need to buy a new CPU and motherboard.
Having said all that, why adjust the vCORE at all? Appropriate question. You adjust it up because as you increase your overclock, the CPU needs to draw more and more current to stay stable at higher speeds. The goal is to set the vCORE just high enough to keep the system stable, but not so high as to add too much heat. The higher the voltage, the more heat your CPU produces - the more heat it produces, the greater the potential for it to fail. That's why extreme overclockers use extreme methods for keeping their CPUs cool. Phase Change and Peltier cooling are common with extreme overclockers, and a few on the very outer fringe even use liquid nitrogen for cooling. On the other end of the spectrum, there are many ways to improve air cooling beyond what the "stock" heatsink/fans ("HSFs") supplied by CPU manufacturers provide. Aftermarket air cooling HSFs are plentiful, with heatpipe technology being the most effective air cooling method at the time this Guide was written. Many more casual overclockers have moved to water cooling (including chilled water), and several very well made kits, as well as individual components for more experienced water cooling overclockers to build from, are readily available.
So, what is a multiplier? I did promise that I would come back to this question. A CPU's multiplier specifies its' internal speed relative to its' base FSB. A CPU designed to work on a 200 mHz motherboard with a 15x multiplier will work internally at 3.0 gHz; i.e., the base FSB times the multiplier gives you the CPU's specified internal speed. If you increase the FSB settings with a 15x CPU from 200 to 220, then the CPU's internal operating speed will increase to 3.3 gHz. Similarly, decrease the FSBs to 150 and a 15x CPU will operate at 2.25 gHz. On the other hand, for CPUs with adjustable multipliers, if you take a 15x CPU operating at 3 gHz (FSB = 200), and operate it at 12x, it will be operating at 2.4 gHz. As to why you would want to do this given that you want more performance; well, I will discuss that a bit later in this Guide.
Goal of Overclocking
The fundamental goal of overclocking is improved system performance. Improved system performance is obtained by increasing data throughput through your system. Ultimately this becomes a compromise between two factors: CPU throughput, and RAM bandwidth. CPU throughput is easy to understand: the faster the CPU runs, the more instructions it can execute per second, and the faster it will process data.
RAM bandwidth is more complex - it too can be measured by data throughput, but the two factors of latency and FSBs can tend to operate together in a fairly complex manner. And, bandwidth will also be affected to a large extent by whether the system is Intel or AMD based, as well as how the particular CPU has been designed.
Without getting overly complex, every CPU contains built-in RAM, commonly at multiple levels. These CPU caches, as they are called, are used to pre-fetch instructions and data for the CPU to work on. As the CPU works, it calls for instructions and data from the various caches. The internal caches of memory built into each CPU run much more quickly, and have much lower latency than system RAM. Their purpose is to always try to keep the CPU core operating at maximum core capacity. How well they accomplish this goal depends on several factors, the most important of which are their size and how well they can predict what instructions and data the CPU is going to need to work on next. The logic used is called "predictive caching", and depending on how good that logic is, it will have a significant effect on how fast your overall system will operate and what your overall memory bandwidth will be.
An example: Intel decided to replace the Socket 478 P4C Northwood with the P4E Prescott. The Prescott had twice the Level 2 built-in cache as that of the Northwood design. Sadly, despite this apparent advantage, the predictive cache logic built into the Socket 478 Prescott was poorly designed, and failed to work very well. This meant that the larger size L2 cache actually became a liability, and the Prescott ran slower on almost all applications other than video encoding. Why? Well, when the predictive cache logic fails, the CPU will force the cache to flush, and go out looking for new instructions and data to process. The better the predictive logic, the fewer times a CPU needs to refresh the cache, and therefore the greater the effective CPU throughput. Northwoods, despite their smaller L2 cache, had better predictive logic and could easily keep the central core processor working utilized more often in most applications. Bottom line, the older, "less capable" Northwood not only ran significantly cooler than the poor "Preshott" as it came to be known, but also much faster.
As I have tried to indicate, many factors have a significant effect on RAM bandwidth, and the only way to really optimize it is to experiment with various latency and FSB settings. And, that is where the importance of the CPU multiplier becomes apparent. Your CPU can only run so fast, no matter what you do realistically. For the P4Cs and Es discussed earlier, 4 gHz was the "Holy Grail" test for CPU speed. Very few Socket 478 CPUs could break, much less reach the 4 gHz barrier. Sure, some could and did, and with extreme cooling solutions the probability of doing so improved considerably. Now, if a CPU has an unlocked multiplier, and you lower the multiplier, you can use higher speed rated RAM and push the FSBs up, thus increasing bandwidth without pushing the CPU beyond its' innate maximum speed.
A numerical example might help you understand this better. Suppose you have a 3.2 gHz processor. With a 200 mHz motherboard, that processor has a 16x multiplier. Let us suppose that the maximum speed the CPU can reach with its' 16x multiplier is FSB 250 or 4.0 gHz. If we can lower the multiplier to 15x, then FSB 250 would equal 3.75 gHz, and in theory we could increase the FSBs to 267, thereby increasing the CPU speed back to 4 gHz, and also increasing the RAM speeds at the same time. Thus, the same CPU speed with a lower multiplier, would produce greater RAM bandwidth, assuming, of course, that the RAM latencies are unchanged as a result of increasing the FSBs. Having an unlocked multiplier gives additional options for increasing overall throughput.
All right, so why then would you ever want to increase the multiplier? Well, once you reach the effective RAM bandwidth limits, and find the "sweet spot" compromise between latencies and FSBs for your RAM, you still may be able to increase overall system performance by increasing the speed of the CPU by increasing the multiplier. That's why!
