Everyone loves a bargain. Who would pass up the chance to buy Porsche performance for a Yugo price? That is the great appeal that overclocking, i.e. deliberately running a microprocessor faster than its official maximum clock frequency, holds for a great many people. It is not hard to see the incentive either. In Figure 1 is a graph of the list price for various speed grades of Intel’s Pentium III and AMD’s Athlon microprocessor (1000 piece quantities, 9/12/99, source: MPR, Oct 6, 1999)
The graphs show that microprocessor price goes up dramatically with maximum rated clock frequency. This is due in some part to the nature of semiconductor manufacturing – the frequency yield distribution of MOS integrated circuits is approximately a bell curve and parts faster than the process center (the frequency at the peak of the distribution curve) are increasingly rare as the clock rate goes up. But mostly this is due to the fact that buyers placing a higher value on the faster running parts. After all, very slow parts are also rare but no chip vendor tries to price them higher than average speed parts! In the fact human psychology plays a great part in the pricing of MPUs for the PC market. Many people are willing to pay top dollar for the first or second fastest speed grade part despite the fact that the substantial price increment would probably speed up their system more if spent on a larger main memory or faster graphics card, hard disk drive etc.
There is an entire mythology that has been built up around overclocking. On one side are enthusiastic overclockers who encourage others to overclock their system and trivialize the risks to user’s systems and data. On the other side are hi-tech Cassandras warning all who will listen that overclockers risk dire fates like having their microprocessors “burn up” or “blow up”. In some ways it is almost more an ethical issue than technical. Some people think that overclocking is immoral or somehow cheats deserving chip companies out of hard earned dollars. I will attempt to address these issues a bit by explaining what factors determine how fast a microprocessor can be safely clocked. To do this I will have to delve into the basics of how the innards of microprocessors function.
Why Have a Clock at All?
If clock rates determine how fast or slow a microprocessor operates an obvious question is why have a clock at all? The role of the clock in a sequential digital system like a processor is similar to the joke about the role of time itself in our universe – it keeps everything from happening all at once. The clock provides order in the midst of chaos and allows chip designers to easily establish causal relationships in their designs and perform step by step operations. It is theoretically possible to design a microprocessor without a clock through the use of asynchronous design techniques. In fact, prior to SDRAMs, most memory chips were entirely composed of asynchronous and self-timed logic. However, the use of synchronous design techniques greatly simplifies the design process (from damn near impossible to just really, really hard) and reduces the time needed to create a working MPU design out of tens of millions of transistors.
There are many different clock signals in a personal computer. But I will refer only to the internal processor clock. This is the fastest clock in a PC and is generated and used only within the microprocessor device itself. The microprocessor communicates with the outside world using a divided down version of the internal processor clock. For example, an average PC today might have an MPU that operates internally at 500 MHz while all the signals running around on the motherboard would be clocked at 100 MHz or slower. The clock itself is a rather boring periodic signal that alternates between a “1” and a “0” and stays in each state for roughly equal time. The clock signal is very carefully buffered and distributed throughout the microprocessor die so that when it changes from a 0 to a 1 (a rising edge) it does so everywhere on the chip simultaneously to within 70 picoseconds or less.
Discuss (One comment)