Pipelining

Another possibility to increase a processor's speed is the instruction level parallelism (ILP), also known as superscalar computation.

The processing of a CPU instruction can internally be split into several stages, such as instruction fetch, decode, execute, and write-back. Before the Intel 486 processor, each instruction has to be finished before the next can be started. With pipelining, when the first stage of an instruction is ready, that instruction can be forwarded to the next stage, and the next instruction's processing can begin with its first stage. In that manner, several instructions can be in flight in parallel, keeping a processor's resources optimally utilized. The next screenshot illustrates this principle, graphically, using a hypothetical four-stages pipeline:

The original Intel 486 pipeline was five stages long, but on modern processors, it can be much longer. For example, the current Intel Atom processors command a pipeline of 16 stages. 

That's all well and good, but, unfortunately, there are some problems lurking in the corners.