A processor is a hardware component or integrated circuit that performs computational operations by executing instructions or otherwise processing data. The word is broader than CPU because processor can refer to many kinds of computational units, including central processing units, graphics processing units, digital signal processors, neural-processing units, network processors and specialised accelerators.
In everyday computer terminology, processor is often used as a synonym for CPU, particularly when discussing desktop or laptop specifications, but technically the terms are not always interchangeable. A processor can be general-purpose or specialised, and it can contain one or many processing cores. Its fundamental characteristic is that it performs defined computational operations according to a hardware architecture. Modern processors are usually semiconductor devices containing large numbers of transistors arranged into logic circuits that implement arithmetic, control, storage and communication functions.
The meaning of processor depends strongly on context. In a desktop specification, saying that a computer has a particular processor generally means its main CPU. In a graphics system, a graphics processor refers to a GPU whose architecture is optimised for highly parallel graphics and numerical operations. A digital signal processor is designed for efficient processing of streams of sampled signals such as audio. A neural-processing unit is specialised for machine-learning operations. A network processor may accelerate packet processing.
These components can coexist in the same system-on-chip. The term processing unit is sometimes used for specialised functional blocks that are not independently general-purpose processors. Therefore, interpreting processor as automatically meaning CPU can be acceptable in ordinary consumer discussion but can be technically incomplete when describing heterogeneous computing systems.
Processors execute operations according to an instruction set or another defined computational interface. A CPU may implement an instruction set such as x86-64 or ARM's AArch64, while a GPU may expose its own programming and execution model. An instruction set architecture specifies the programmer-visible operations, registers, data types, addressing rules and memory semantics available to software. The processor's microarchitecture determines how those operations are implemented internally.
Different processors can implement the same architecture while using very different pipelines, caches, execution units and prediction mechanisms. This distinction allows software compatibility to be separated from hardware implementation. A program compiled for a particular instruction set can potentially run on processors from different manufacturers if those processors implement compatible architectural requirements, even though their internal circuitry and performance characteristics differ considerably.
Processors can be classified according to their intended workload and degree of specialisation. General-purpose CPUs are designed to execute diverse software and support broad programming models. GPUs contain large numbers of arithmetic resources optimised for data-parallel workloads. DSPs commonly provide specialised instructions for filtering, transforms and signal-processing operations. NPUs and other AI accelerators provide hardware designed around matrix multiplication, tensor operations or other machine-learning workloads. Microcontrollers may integrate a relatively simple processor with memory and peripherals for embedded control.
Application-specific integrated circuits can contain fixed or highly specialised processing logic for particular tasks. Field-programmable gate arrays can implement custom digital processing structures after manufacture. These categories overlap in modern systems, and many contemporary chips contain several different kinds of processing resources rather than relying on a single processor type.
The physical implementation of a processor is normally based on semiconductor transistors that form logic gates, registers, multiplexers, arithmetic units and control circuitry. Billions of transistors can be integrated into modern high-performance processor designs. Semiconductor manufacturing involves processes such as photolithography, deposition, etching and implantation to construct microscopic structures on silicon wafers. Processor design must balance performance, power consumption, area, manufacturing cost and reliability.
Increasing transistor density allows more cache memory, execution units and specialised accelerators to be integrated, but physical constraints such as heat dissipation, interconnect delay and leakage current limit how efficiently additional transistors can be used. Advanced processors may therefore use multiple dies or chiplets rather than placing every function on a single monolithic die.
Processor performance depends on architecture and workload rather than on a single specification. Clock frequency indicates how rapidly a processor's timing reference cycles, but it does not directly state how much useful work is completed per second. A processor with fewer but more capable cores can outperform one with more slower cores for a single-threaded workload, while a highly parallel workload may favour greater core or accelerator capacity.
Cache sizes, memory bandwidth, branch prediction, execution width, vector instructions, accelerator availability and software optimisation can all affect performance. Thermal limits and power-management policies also determine how long a processor can sustain peak performance. Benchmarks are therefore normally used to compare processors under specified workloads rather than assuming that one headline specification fully describes computational capability.
The historical meaning of processor has expanded as computing systems have become heterogeneous. Early computers used separate circuits and boards for different computational functions, whereas integrated circuits progressively combined more functionality. Microprocessors placed a CPU's principal processing circuitry on a single chip, enabling personal computers and embedded systems. Modern system-on-chip designs can integrate CPUs, GPUs, media engines, image processors, security components, memory controllers and communication interfaces.
The result is that a single physical chip may contain several distinct processors or processing engines. A smartphone SoC, for example, may contain CPU cores for general-purpose software, GPU cores for graphics, an image signal processor for cameras and an NPU for machine-learning tasks. Calling the entire chip a processor can therefore be common in marketing while technically referring to a system containing multiple processing units.
The distinction between processor, CPU and core is important. A CPU is a particular kind of general-purpose processor responsible for executing the system's main instruction streams. A CPU core is an individual processing engine within a multicore CPU capable of independently executing an instruction stream. A processor package can contain one or multiple dies, and each die can contain several cores. Meanwhile, a thread is a software or architectural execution context rather than necessarily a physical processor. A GPU core or execution unit is also not equivalent to a CPU core because its architecture and intended workload can be substantially different. These distinctions prevent specifications such as processor count, core count and thread count from being interpreted as interchangeable quantities.
