From the late 1940's through the present the von Neumann architecture, named after the mathematician John von Neumann, has been employed in practical general-purpose computers. The hallmark of a von Neumann machine is a large random-access memory. Each cell in the memory has a unique numerical address, which can be used to access or replace the contents of that cell in a single step. In addition to its ability to address memory locations directly, a von Neumann machine also has a central processing unit (the CPU ) that possesses a special working memory (register memory ) for holding data that are being operated on and a set of built-in operations... The exact design of the central processor varies considerably, but typically includes operations such as adding two binary integers, or branching to another part of a program if the binary integer in some register is equal to zero (so-called conditional branching). The CPU can interpret information retrieved from memory either as instructions to perform particular operations or as data to apply the current operation to. Thus, a portion of memory can contain a sequence of instructions, called a program, and another portion of memory can contain the data to be operated on by the program. The CPU repeatedly goes through a fetch-execute cycle, in which the next operation and its arguments...  A von Neumann machine runs efficiently because of its random-access memory and because its architecture can be implemented in electronic circuitry that makes its basic operations extremely fast (currently from millions to tens of millions of instructions per second in desktop computers).
--From Cognitive Science: an Introduction, David W. Green (editor)