Every program you have written, studied, or used runs on physical hardware — real components made of metal, silicon, and circuits that carry electrical signals at extraordinary speed. Understanding that hardware is not optional background knowledge for a computer scientist; it is the foundation that explains why software behaves the way it does. When a program runs slowly, crashes, or consumes too much battery, the explanation usually begins inside the machine.

This chapter introduces the physical layer of computing. You will learn what the major hardware components are, what each one does, and how they work together to execute programs. The central idea is simple: a computer is a machine that fetches instructions from memory, decodes what each instruction means, and executes the required operation — one instruction at a time, billions of times per second. Every other hardware concept in this chapter connects to that cycle.

By the end of this chapter, you will be able to identify the major components of a computing system, explain the role each plays, trace the fetch-decode-execute cycle conceptually, and reason about how hardware choices affect performance, cost, energy use, and software behavior. These ideas prepare you directly for Chapter 16, which covers the operating system — the software layer that sits between hardware and everything else.

The word hardware refers to the physical components of a computing system — everything you can touch, see, and measure in grams and centimeters. A processor chip, a memory module, a hard drive, a keyboard, a screen, and the circuit board they are all connected to: these are hardware. Hardware is distinct from software, which refers to the programs and data that run on the hardware, and from data, which is the information being processed and stored.

The relationship between hardware and software is one of the most important ideas in computer science. Hardware provides the physical capacity to compute — the circuits, the storage cells, the electrical pathways. Software gives those circuits meaning and direction — it tells the hardware what to do and in what order. Neither is useful without the other. A processor with no software is like a factory with no instructions; software with no processor is like a recipe with no kitchen.

Hardware can be divided into several broad categories that will structure this entire chapter. The processor performs the actual computations. Memory holds the instructions and data the processor is actively working with. Storage preserves information even when power is off. Input devices bring information into the system. Output devices send results out of the system. The motherboard connects all of these components and allows them to communicate. Understanding what each category does — and why it must be designed carefully — is the foundation of computer architecture.

The Central Processing Unit, or CPU, is the component that actually executes instructions. It is sometimes called the "brain" of the computer, though the analogy is limited — the CPU does not think or decide; it executes a sequence of precisely defined operations at enormous speed. What makes the CPU remarkable is not intelligence but relentless, high-speed execution of simple operations: add these two numbers, compare these values, move this data from one location to another, jump to a different instruction if a condition is true.

Inside the CPU, two functional units do most of the work. The Arithmetic Logic Unit, or ALU, performs mathematical operations (addition, subtraction, multiplication, division) and logical operations (AND, OR, NOT, comparison). Every computation your program performs — calculating a score, sorting a list, encrypting a message — passes through the ALU. The Control Unit manages the flow: it fetches the next instruction, decodes what it means, directs the ALU or other units to carry it out, and moves to the next instruction. The control unit is the coordinator; the ALU is the worker.

Between the ALU and main memory sit the registers. Registers are tiny, extremely fast storage locations built directly into the CPU chip. A modern CPU might have a few dozen registers. They hold the values the CPU is working with right now: the two numbers about to be added, the result of the last comparison, the address of the next instruction. Registers are faster than any other form of storage by a large margin, but they are also very small. A program cannot keep all of its data in registers, which is why the CPU must constantly exchange information with memory.

The clock controls how fast the CPU operates. A clock is an electronic oscillator that produces a regular signal — a beat — at a fixed frequency measured in gigahertz (GHz). One gigahertz means one billion cycles per second. The CPU performs one or more operations per clock cycle. A CPU running at 3.5 GHz executes up to 3.5 billion operations every second. Clock speed is one measure of CPU performance, but it is not the only one. Modern CPUs also use pipelining — overlapping the stages of successive instructions — and execute multiple instructions simultaneously using a technique called superscalar execution.

Most modern CPUs contain multiple cores. Each core is an independent processing unit capable of executing its own stream of instructions. A quad-core CPU contains four such units on a single chip. Programs designed to use multiple cores simultaneously — through a technique called parallel or concurrent programming — can run significantly faster than on a single-core CPU. Operating systems distribute tasks across cores automatically, which is why a modern computer can play music, download a file, and run a text editor all at the same time without any of them grinding to a halt.

Between the CPU and main memory sits the cache. Cache is a small, very fast memory built directly into or very close to the CPU chip. Modern CPUs typically have multiple cache levels: L1 cache (the fastest and smallest, often 32–128 kilobytes per core), L2 cache (larger and slightly slower, often 256 kilobytes to 1 megabyte per core), and L3 cache (shared among cores, often 8–64 megabytes). When the CPU needs data, it checks L1 first. If the data is not there, it checks L2, then L3, then main RAM, then storage. Each level is larger but slower. Keeping frequently used data in cache dramatically reduces the time the CPU spends waiting.

Memory is where the CPU finds the instructions and data it needs to work with right now. The word "memory" in computer science refers specifically to Random Access Memory, or RAM — the fast, temporary storage that holds the operating system, open applications, and data currently in use. RAM is volatile: when power is removed, everything stored in it disappears. This is why your work is lost if the computer loses power before you save, and why the computer must reload the operating system every time it starts.

RAM is measured in gigabytes. A modern desktop might have 16 to 64 GB of RAM; a smartphone typically has 6 to 16 GB. More RAM allows more programs to run simultaneously without the system slowing down. When RAM fills up, the operating system uses storage as an overflow area called virtual memory or a swap file — but storage is dramatically slower than RAM, so this causes noticeable performance drops.

The relationship between RAM and the CPU is governed by memory bandwidth — how much data can move between them per second. Even if a CPU can compute very quickly, it can only go as fast as memory can feed it. This is sometimes called the memory bottleneck. CPU designers address it with cache, which stores copies of recently used data close to the processor so the CPU does not need to wait for the slower RAM on every operation.

It is critical to understand the difference between RAM and storage. Students frequently confuse the two because both hold data. The key distinction is persistence and speed. RAM is fast (measured in nanoseconds) but temporary. Storage is slow (measured in microseconds or milliseconds) but permanent. When you save a file, you are moving data from RAM into storage. When you open a file, you are moving data from storage into RAM. The CPU only works directly with data that is in RAM or in cache — never directly from storage.

Storage refers to the hardware that retains data even when the system is powered off. While RAM forgets everything when you shut down, storage keeps files, programs, databases, photographs, videos, and the operating system itself safely preserved for days, years, or decades. This persistence makes storage the foundation of practically everything users care about in a computer.

The two dominant storage technologies in modern computing are hard disk drives (HDDs) and solid-state drives (SSDs). A hard disk drive is a mechanical device: it contains one or more rigid magnetic platters that spin at high speed (typically 5,400 or 7,200 revolutions per minute), and a read/write head that moves across the platters to access data. Because data retrieval depends on physical motion — waiting for the right sector of the disk to spin under the head — HDDs are significantly slower than SSDs. However, HDDs offer very high capacity at low cost, which is why they remain common for large-scale backup and archival storage.

A solid-state drive uses flash memory — the same type of storage found in USB drives and smartphone storage — with no moving parts. Flash memory stores data as electrical charges trapped in memory cells. Because there is no mechanical motion, SSDs can access any location at nearly uniform speed, dramatically faster than HDDs. SSDs are also smaller, lighter, quieter, and more shock-resistant. Their primary disadvantage is cost per gigabyte, though SSD prices have dropped significantly in recent years. Most modern laptops and many desktop computers use SSDs as their primary drive.

Storage capacity is measured in standard units. One kilobyte (KB) is 1,024 bytes. One megabyte (MB) is approximately one million bytes. One gigabyte (GB) is approximately one billion bytes. One terabyte (TB) is approximately one trillion bytes. A typical SSD in a laptop today might hold 512 GB to 1 TB. An enterprise server might use multiple drives totaling dozens or hundreds of terabytes.

Beyond HDDs and SSDs, other storage media are worth understanding. USB flash drives use flash memory in a portable, removable form. Optical discs (CDs, DVDs, Blu-ray) store data as patterns of pits and lands read by a laser; they were once common for software distribution and backups but have largely been replaced by downloads and cloud storage. Cloud storage refers to storage provided by remote servers accessed over the internet — the data exists on SSDs and HDDs in data centers somewhere else, and the user accesses it through a network connection.

A computer that could not receive information from the outside world or send results back to users would be useless. Input devices bring data into the computing system. Output devices send processed results out of it. Together, they form the interface between the computing system and the humans — or other systems — that interact with it.

The most familiar input devices are the keyboard and the mouse. A keyboard converts physical key presses into electrical signals that the computer interprets as characters, commands, or shortcuts. A mouse tracks physical movement and translates it into cursor position on screen, sending click events when buttons are pressed. Touchscreens combine input and output on a single surface: a display that also detects the position and pressure of finger touches. Most smartphones, tablets, and many laptop screens use capacitive touchscreens that sense the electrical properties of a fingertip.

Other input devices capture different types of information from the physical world. A camera converts light into digital image data — a grid of pixels with color values. A microphone converts sound pressure waves into electrical signals that are then digitized. A scanner converts printed or physical images into digital files. Sensors in embedded systems capture temperature, pressure, acceleration, location (GPS), and dozens of other environmental measurements. Every sensor is ultimately an input device: it samples something from the physical world and converts it into numbers the computer can process.

The primary output device is the monitor or display. A display presents visual information by controlling the color and brightness of millions of tiny pixels arranged in a grid. Resolution describes how many pixels the display contains: a 1920×1080 display has 1,920 columns and 1,080 rows of pixels, totaling about 2 million pixels. Higher resolution displays show more detail. Refresh rate, measured in hertz (Hz), describes how many times per second the display updates its image — 60 Hz means 60 full images per second, which appears smooth to human eyes. Speakers convert digital audio signals back into sound waves. Printers convert digital documents into physical marks on paper or other materials.

Some devices serve as both input and output. A network interface card both sends and receives data over a network. A hard drive or SSD reads data (input to the CPU) and writes data (output from the CPU). A touchscreen displays information and receives touch input simultaneously. In computer architecture, these are called I/O devices collectively — the acronym I/O standing for input/output.

The individual components described so far — CPU, RAM, storage, and I/O devices — are all separate pieces of hardware. Something must connect them into a functioning system. That something is the motherboard, a large printed circuit board that physically and electronically connects every major component of the computer.

The motherboard contains sockets and slots where components attach. The CPU installs into the CPU socket, which holds it securely and makes electrical contact with hundreds or thousands of tiny pins. RAM modules plug into memory slots, usually two or four slots on a desktop motherboard. The motherboard also contains slots for expansion cards — graphics cards, network cards, and other peripherals — and connectors for storage drives, fans, power, and front-panel controls.

Data travels between components through buses. A bus is a set of electrical conductors — wires or traces on a circuit board — that carry signals between components. The data bus carries the actual data being transferred. The address bus carries information about where in memory to read from or write to. The control bus carries command signals that coordinate operations. The width of a bus (how many bits it can carry simultaneously) and its speed (how fast signals travel) determine bandwidth — how much data can move per second.

Modern computers use several important bus standards. PCIe (Peripheral Component Interconnect Express) is the high-speed bus used to connect graphics cards, NVMe SSDs, and other high-bandwidth expansion cards directly to the CPU. USB (Universal Serial Bus) is the standard for external device connections — keyboards, mice, flash drives, phones, cameras, and hundreds of other devices all connect via USB in its various versions (USB 2.0, USB 3.x, USB-C). SATA (Serial ATA) was the standard interface for connecting hard drives and SSDs to the motherboard, though NVMe over PCIe is now preferred for high-performance drives.

The motherboard also contains the chipset — additional silicon that manages communication between the CPU and other components including storage controllers, USB hubs, and audio systems. Older systems divided this into a northbridge (handling fast connections to RAM and GPU) and a southbridge (handling slower connections to storage and USB), though modern CPUs integrate much of this functionality directly onto the processor die.

The CPU is a general-purpose processor: it is designed to handle any type of computation, but it does so with a relatively small number of highly capable cores optimized for sequential, complex tasks. The Graphics Processing Unit, or GPU, is fundamentally different in design philosophy. A GPU contains thousands of smaller, simpler cores all designed to execute the same simple operation on many pieces of data simultaneously. This architecture is called Single Instruction, Multiple Data (SIMD), and it is ideally suited to the work of rendering graphics.

Rendering a 3D scene for a video game requires computing the color and brightness of millions of pixels every frame — 60 times per second or more in modern games. Each pixel's calculation is largely independent of the others and follows the same pipeline of operations: transform 3D geometry, apply lighting equations, sample textures, blend colors. Because all these pixel calculations are the same type of operation on different data, a GPU's massively parallel architecture handles them far more efficiently than a CPU ever could. The GPU renders frames and outputs them to the display through a video port (HDMI, DisplayPort, or similar).

GPUs have their own dedicated memory called VRAM (Video RAM), distinct from the system's main RAM. VRAM is optimized for the high-bandwidth access patterns of graphics rendering — a GPU might need to read or write to every pixel in a 4K frame in milliseconds. High-end gaming GPUs contain 8 to 24 GB of VRAM.

Beyond graphics, GPUs have become essential for two other domains: scientific simulation and machine learning. Training a neural network requires performing billions of matrix multiplications — exactly the type of massively parallel operation GPUs excel at. This is why data centers training large AI models use thousands of GPUs rather than CPUs. Researchers have coined the term GPGPU (General-Purpose computing on Graphics Processing Units) for this use of GPU hardware for non-graphics tasks.

Computers use either integrated graphics — a GPU built directly into the CPU chip or the same package, sharing system RAM — or a discrete GPU — a separate card with its own processor, VRAM, and cooling. Integrated graphics consume less power and are suitable for everyday computing, video playback, and light creative work. Discrete GPUs are necessary for 3D gaming, video editing, and machine learning workloads.

Every program a computer runs — from the simplest script to a complex operating system — consists of a sequence of instructions stored in memory. The CPU executes these instructions one by one (or several at a time using pipelining) by following a simple three-phase cycle repeated continuously: fetch, decode, and execute. This cycle is the heartbeat of every computing device, and understanding it gives you a precise mental model of what "running a program" actually means at the hardware level.

Before the cycle begins, the CPU uses a special register called the Program Counter (PC) — sometimes called the Instruction Pointer — to track which instruction to execute next. The program counter holds the memory address of the current instruction. After each instruction is processed, the program counter is incremented to point to the next instruction, unless a jump or branch instruction changes it deliberately.

In the fetch phase, the CPU sends the address stored in the program counter to main memory (or cache, if available) and retrieves the instruction stored at that address. The instruction is loaded into a special register called the Instruction Register (IR). The program counter is then incremented to point to the next instruction, ready for the next cycle.

In the decode phase, the control unit examines the instruction now held in the instruction register and determines what operation it represents. Instructions are stored in binary as machine code — patterns of bits that the CPU's circuitry is designed to recognize. Decoding means identifying which operation this instruction requests (add, subtract, load, store, jump, compare, etc.) and which operands — registers, memory addresses, or immediate values — it involves.

In the execute phase, the control unit routes the instruction to the appropriate functional unit. If it is an arithmetic operation, the ALU performs the calculation. If it is a memory operation, data is loaded from or stored to the specified address. If it is a jump instruction, the program counter is set to a new address, redirecting execution. After execution, results may be written back to a register or to memory — sometimes called the write-back phase.

Modern CPUs accelerate this cycle through pipelining — processing different stages of consecutive instructions simultaneously. While one instruction is in the execute phase, the next instruction is already in the decode phase, and the one after that is being fetched. This overlapping means a new instruction can complete on every clock cycle in ideal conditions, greatly improving throughput over strictly sequential processing.

The computers most people think of — desktops, laptops, smartphones, tablets — are general-purpose computers: they can run any software you install and perform an enormous variety of tasks. But these represent only a fraction of the computing devices in the world. The vast majority of processors in existence are embedded systems: computing devices designed for a specific, dedicated purpose, integrated into a larger machine or product.

An embedded system is a computer built into another device to perform a fixed set of functions. It typically consists of a microcontroller — a small chip that integrates a processor, memory, and I/O peripherals on a single piece of silicon — along with sensors, actuators, and specialized software called firmware. Unlike a general-purpose computer, an embedded system usually runs a single program continuously and is not intended to be reprogrammed by end users.

Embedded systems are everywhere. The anti-lock braking system (ABS) in a car uses an embedded processor to monitor wheel speed sensors and modulate brake pressure hundreds of times per second — too fast for a human driver and with real-time reliability requirements no general-purpose OS could guarantee. A medical pacemaker uses an embedded system to monitor heart rhythm and deliver precisely timed electrical pulses; its software must never crash or delay. A microwave oven uses an embedded system to read keypad input, control magnetron power levels, operate the turntable motor, and manage the timer — all from a processor that costs a fraction of a dollar.

Other examples of embedded systems span nearly every domain of modern life. Traffic light controllers manage signal timing and respond to sensor inputs. Washing machines monitor water levels and motor speeds while executing programmed wash cycles. Airplane fly-by-wire systems translate pilot inputs into hydraulic and motor commands through embedded processors with multiple redundant systems for safety. Implantable glucose monitors measure blood sugar continuously and wirelessly transmit data to smartphones. Industrial robots execute precisely timed sequences of motor commands coordinated by embedded controllers.

Embedded systems face constraints that general-purpose computers do not. They often have very limited memory and processing power, because the chip must be inexpensive and energy-efficient for production at scale. They may need to operate on battery power for years — a tire pressure sensor must last the life of the tire without a battery change. Many embedded systems have real-time requirements: they must respond to events within a guaranteed time window. An airbag controller must detect a crash and trigger the bag within milliseconds; a delay of even a few extra milliseconds could be fatal. Meeting these constraints requires careful hardware selection, efficient firmware, and often a real-time operating system (RTOS) designed for predictable, bounded response times.

Designing or choosing a computer system always involves tradeoffs. No single configuration is best for every situation. Performance, power consumption, cost, size, reliability, and thermal management all pull in different directions, and real engineering decisions require balancing them deliberately based on the intended use case. Understanding these tradeoffs is one of the most practically important ideas in computer architecture.

Clock speed and core count are the most visible CPU performance metrics. Increasing clock speed — making the CPU tick faster — increases performance for single-threaded workloads (tasks that cannot be parallelized). But faster clocks require more power and generate more heat. There is a physical limit to how fast silicon circuits can switch at a given voltage, which is why clock speeds have largely plateaued around 4–6 GHz in recent years despite continued performance improvements through other means. Increasing core count helps with multi-threaded workloads, but only if the software is written to use multiple threads — a single-threaded program runs no faster on 16 cores than on 4.

Thermal design power (TDP) is the maximum sustained heat a CPU or GPU generates under full load, measured in watts. A processor with a 65W TDP requires cooling capable of dissipating 65 watts of heat. High-performance desktop CPUs may have TDPs of 125–250 watts, requiring large heatsinks and fans or liquid cooling systems. Laptop processors use mobile variants with lower TDPs — often 15–45 watts — to remain cool and preserve battery life, at the cost of some peak performance. Understanding TDP helps explain why high-performance laptops are thicker and heavier, why data centers require enormous cooling infrastructure, and why battery life and performance are often in tension in mobile devices.

Memory bandwidth determines how fast data can flow between RAM and the CPU. A CPU that can compute very quickly but must wait for slow memory is said to be memory-bound — the memory is the bottleneck. This is why cache exists: by keeping frequently used data close to the processor, cache reduces the frequency of slow trips to main RAM. Adding more RAM increases capacity but not necessarily bandwidth; faster RAM (higher frequency or wider bus) increases bandwidth.

Storage speed is frequently the bottleneck for loading applications, opening large files, and running databases. Replacing an HDD with an SSD in an older computer often produces a dramatic improvement in perceived performance because the system spends less time waiting for storage. NVMe SSDs offer further improvements for workloads that involve large sequential reads or frequent random reads, though the gains are less noticeable for simple web browsing or document editing.

Cost is always a factor. More RAM, faster SSDs, higher-core-count CPUs, and discrete GPUs all increase price. Budget systems make deliberate compromises — less RAM, slower storage, integrated graphics — that are perfectly adequate for many users. Workstation and server configurations spend significantly more for the performance that demanding tasks require. A well-informed buyer or system designer understands which components are the actual bottleneck for the intended workload and invests accordingly.

The tension between performance and power is most visible in mobile devices. A smartphone must run for a full day on a battery the size of a credit card while also handling communications, navigation, photography, and streaming. Mobile chips achieve this through aggressive power management: cores that can be clock-gated (switched off when idle), dynamic voltage and frequency scaling (running slower and at lower voltage when full speed is not needed), and heterogeneous architectures that combine high-performance cores for demanding tasks with efficiency cores that handle background work with minimal power drain.