Processor
A processor is any electronic circuit that carries out computation by executing instructions or transforming streams of data, and the word is a family name rather than a single device. A central processing unit runs general programs; a graphics processing unit applies the same operation across thousands of data items at once; a neural processing unit or tensor processor accelerates the matrix arithmetic of machine learning; a digital signal processor is tuned for audio, radio and sensor streams; and a microcontroller bundles a small processor with memory and input-output on one chip to run an appliance. What they share is that each turns a stored description of work into electrical operations, and what separates them is which kind of work they are built to do quickly or cheaply.
The term "microprocessor" marks the moment a whole processor fit on one integrated circuit. Early computers filled rooms with vacuum tubes and later discrete transistors; in 1971 Intel's 4004 placed a complete four-bit processor on a single chip, followed by the 8008, 8080 and, in 1978, the 8086 that began the x86 line. Today a "processor" in a laptop or phone is usually a system on a chip: several CPU cores, a GPU, a neural engine, memory controllers, and radio and media blocks fabricated together, as in the Apple M-series chips introduced from 2020 or the Snapdragon and Tensor chips in phones. The processor in a desk-side server is more likely to be a CPU alone, surrounded by separate accelerator cards.
Different processors embody different bets about how to spend a limited budget of transistors and power. A CPU core spends most of its area on machinery that makes one instruction stream finish as soon as possible, such as large caches, branch prediction and out-of-order execution, so it has low latency on branching, unpredictable code but only a few cores. A GPU, a term NVIDIA popularised with the GeForce 256 in 1999, spends its area on thousands of simple arithmetic units, accepts slower individual operations and hides delay by switching among many threads, so it has high throughput on regular, data-parallel work such as graphics and neural-network training. Google announced its Tensor Processing Unit in 2016, a chip built around a large grid of multiply-accumulate units for neural networks, and phones now carry similar neural engines.
The instruction set architecture is the contract that decides which software a processor can run. A CPU understands one family of machine instructions: x86-64 from Intel and AMD, Arm, or the open RISC-V standard, and a program compiled for one will not run natively on another without translation or recompilation. Accelerators are controlled differently: a GPU or TPU does not boot an operating system but receives batches of work from a CPU through drivers and runtime libraries such as CUDA, ROCm or Metal, plus frameworks like PyTorch and TensorFlow that hide the details. Choosing among processor types is therefore a software decision as well as a hardware one.
Common misconceptions are that "processor" means CPU, that more cores or a bigger number always means faster, and that GPUs are only for graphics. The measure that matters is the match between workload and design: a GPU that runs a rendering or training job at enormous speed is slow at running an operating system, and most software gains little from extra cores unless it is written to use them in parallel. Efficiency is a second axis, since a phone's neural engine may use a small fraction of the energy a general CPU would need for the same inference. Microcontrollers are the most numerous processors by far, running cars, toys, sensors and keyboards, usually without an operating system, and their small cost and low power are their design goals.
Read a processor as a machine that turns instructions into action, and the many kinds of processor as different answers to one question: how much of the chip should be spent on speed for one task, and how much on doing many things at once.