The clock on a CPU controls when the logic changes state. There are a number of steps involved in executing an instruction...the basic ones are instruction fetch, instruction decode / register fetch, execute, memory access, write back. Modern CPUs employ a number of other performance techniques like register renaming and out-of-order execution, so that's why the basic pipelines of the Athlon and P4 are 12 and 20 stages, respectively. Therefore, it takes 12 or more clock cycles for the Athlon to execute an instruction. By forcing the clock signal to go faster, you are forcing the CPU to change states more often...thus, it executes instructions faster. The reason that overclocking can introduce instability (besides heat issues) is that each stage of the pipeline takes a certain amount of time to complete its task. Say, for example, you have a CPU with a 10ns clock period (100 MHz), and it's arithmetic logic unit (used during the execute stage) has an 8ns delay for an add instruction. Therefore, if you force the clock period to be shorter than 8ns (125 MHz), the CPU might change state before the add is completed. The result is that the CPU will latch an incorrect result into the register file, and errors will ensue.