Table of Contents
Part of the series:Computers
A computer is a machine that follows instructions astonishingly fast. That is the entire idea. Everything else — gigahertz, gigabytes, operating systems, apps — is detail piled on top of that one concept. This post stays at the panoramic level: what the main parts are, what each one does, and how they cooperate to run a program. Each part deserves its own deep dive; here you get the map of the territory.
The stored-program idea
Before computers, machines did one fixed job: a loom wove one pattern, a calculator added. The breakthrough behind the modern computer1 is the stored-program concept2: instructions live in memory as data, exactly like the numbers they operate on. Because a program is just data, the same machine can run a spreadsheet, a game, or a web browser — you only have to load different instructions.
Most machines still follow the von Neumann architecture3: one shared memory holding both programs and data, and a processor that fetches instructions from it one at a time. It is simple, general, and — despite decades of refinement — still the blueprint of your laptop and phone.
The processor (CPU)
The central processing unit4 is the part that executes instructions. At a conceptual level, its work can be described as fetch, decode, execute: grab the next instruction from memory, figure out what it says, do it, move to the next one — billions of times per second. Real modern cores blur this simple picture with pipelines, superscalar and out-of-order execution, and layers of cache, but every one of those optimizations preserves the fetch-decode-execute contract.
Inside, a CPU has a small set of ultra-fast slots called registers for the values it is working on right now, an arithmetic-logic unit (ALU) that does the actual math and comparisons, and a control unit conducting the whole dance. Modern CPUs pack multiple cores (independent workers) and layers of cache (tiny, blazing-fast memory sitting next to the cores) to keep that loop fed. Clock speed, measured in gigahertz, is how many clock cycles occur per second — one factor among many in determining performance.
Memory (RAM)
Random-access memory5 is the computer’s workbench: fast, temporary storage where running programs and their data live. Every byte has an address, and the CPU can jump directly to any of them — hence “random access.” Direct does not mean instantaneous, though: every access takes time (latency), and that delay is one reason machines surround the CPU with caches.
Two traits define RAM: it is fast (orders of magnitude quicker than storage) and volatile (it forgets everything when the power goes out). That is why unsaved work vanishes in a blackout and why machines with more RAM juggle more apps at once — there is simply a bigger workbench.
Storage
Storage6 is the filing cabinet: slower than RAM but persistent — it remembers without power. Hard drives keep bits as magnetic patterns on spinning platters; SSDs7 store data in NAND flash memory, using electrical charge to represent information. With no moving parts, they are faster and silent.
The classic tradeoff of computing lives here: registers are among the fastest and tiniest, followed by cache, RAM, and persistent storage such as SSDs and hard drives — each step slower, bigger, and cheaper per byte. Remote storage adds another dimension: network latency. The whole art of performance is keeping the data the CPU needs right now as high up that ladder as possible.
Input and output
A computer with no senses and no voice would be useless. Input devices8 (keyboard, mouse, touchscreen, microphone, camera, network card) bring information into the computer; output devices9 (screen, speakers, printer, network card again) carry information back out. The network card is both at once, which is why a machine can be useful with nothing plugged into it except Wi-Fi.
Software: the machine’s soul
Hardware without instructions is an expensive paperweight. System software10 — above all the operating system11 (Linux, Windows, macOS, Android) — manages the hardware on everyone else’s behalf: deciding which program gets CPU time, handing out memory, talking to disks and screens through drivers. Application software — your browser, editor, games — runs on top, asking the OS for everything instead of touching hardware directly.
Specialized processors (GPU)
Not every job suits the CPU. Graphics, machine learning, and scientific simulation all share a shape: the same simple operation applied to enormous amounts of data at once. The graphics processing unit12 is built for exactly that — thousands of small, simple cores optimized for throughput, where a CPU core is optimized for single-thread speed and branching agility.
GPUs were born to draw triangles and shade pixels, work that is embarrassingly parallel. Programmers then realized the same hardware accelerates anything with that shape, a practice called general-purpose GPU computing13. Today the CPU routinely offloads suitable work to the GPU (and increasingly to NPUs for AI), an arrangement called heterogeneous computing: each processor doing what it does best.
From power button to program
Press power and the machine wakes up without an operating system running — RAM holds no persistent copy of it. So a tiny program baked into the motherboard (firmware) runs first, checks the hardware, finds the disk, and loads the operating system’s kernel into memory. The OS starts its services, shows you a login screen, and waits.
Launch an app and the OS copies it from storage into RAM, points the CPU at its first instruction, and the fetch-decode-execute loop takes over. The app asks the OS to draw windows and read files; the OS talks to the hardware; results flow back as pixels and sound. Billions of instruction cycles per second, all to move a cursor or play a song — and it feels instant.
The big picture
CPU executes, memory hosts the work in progress, storage remembers, input and output connect to the world, specialized processors accelerate parallel work, and software tells everyone what to do. Change any one of them and the machine changes character — but the division of labor stays the same, from a smartwatch to a supercomputer. If you want to know what the bits flowing between these parts actually mean, start with Information Representation Explained.