Other meanings of Central processing unit
Computer architecture
The central processing unit (CPU) is the main processor of a computer that executes instructions. It reads encoded operations, performs arithmetic and logical work, and coordinates data movement among memory, storage, and peripheral devices. Modern CPUs commonly contain multiple cores, private and shared caches, and hardware for predicting and reordering instruction execution. 1
The CPU executes a program by repeatedly fetching, decoding, and executing instructions. A control unit interprets each instruction, while arithmetic-logic units perform operations such as addition, comparison, and bitwise manipulation. Registers provide very small, fast storage for operands, addresses, status flags, and intermediate results. This cycle is the practical expression of the stored-program concept, in which instructions and ordinary data are represented in memory and selected as execution proceeds. 1
Most general-purpose computers follow a form of von Neumann architecture, although the separation between instructions and data may be strengthened inside the processor. The CPU communicates with main memory through buses or point-to-point links, and operating-system software determines which processes receive execution time. A clock provides timing signals, but clock frequency alone does not determine performance: instruction count, work per instruction, memory latency, and parallelism also matter. 2
Modern CPU performance depends on a layered design rather than on clock speed alone. An instruction set architecture defines the programmer-visible operations, registers, data types, and memory rules; the microarchitecture is the particular hardware implementation that carries them out. Techniques such as pipelining divide instruction processing into stages, while superscalar execution allows several independent instructions to begin in one cycle. Branch prediction guesses the path of conditional code so that pipelines remain busy, and out-of-order execution can complete ready operations before earlier, stalled ones. 3
Cache memory reduces the delay of fetching frequently used instructions and data. L1 caches are usually smallest and fastest, followed by larger L2 and sometimes shared L3 caches. When software exposes enough independent work, a multicore processor can execute threads concurrently; otherwise, synchronization, serial code, memory traffic, and power limits restrict the gain. Performance comparisons therefore require a workload, not a single universal ranking. 2
CPU design includes several less visible forms of specialization. Some processors combine high-performance and energy-efficient cores, allowing a scheduler to match demanding work with economical hardware. Others add vector or matrix units for scientific computing, graphics, signal processing, or machine-learning operations. Embedded CPUs may omit features expected on desktop systems but emphasize predictable timing, low power consumption, or long service life. These differences explain why a small microcontroller, a smartphone system-on-chip, and a server processor can all be CPUs while serving very different purposes. 4
The boundary between CPU and other processors is also historically fluid. A system-on-chip may integrate CPU cores with graphics processors, memory controllers, radio circuitry, and accelerators. Open instruction-set projects such as RISC-V separate a freely usable instruction-set specification from particular commercial implementations. Security features have likewise become part of CPU architecture, including privilege levels, memory protection, trusted execution facilities, and defenses against speculative-execution side channels. 5
The modern CPU emerged from the stored-program computers developed in the mid-20th century, when electronic switching replaced slower electromechanical and relay mechanisms. Early processors occupied large systems; integrated circuits then placed increasing numbers of transistors on a chip. Moore's law described a long-running trend in transistor density, but shrinking transistors did not remove practical limits involving heat, leakage current, manufacturing variation, and memory access. 6
Since the mid-2000s, designers have generally gained more useful capacity through multiple cores, larger caches, specialized accelerators, and improved efficiency rather than through indefinitely increasing clock rates. CPUs also remain constrained by the memory wall: computation can be far faster than access to data outside nearby caches. Compilers, operating systems, and application developers consequently share responsibility for performance, using locality, parallelism, and appropriate instruction sets while preserving correctness and security. 3
Sources are numbered in order of appearance; citations identify the authoritative references supporting specific claims.
Help improve the encyclopedia. Reports go straight to the site manager.