Additional Hardware Components
Beyond the CPU: Additional Hardware
The CPU is the heart of any computer, but it does not work alone. Several additional hardware components are essential to how a modern computer operates. This lesson covers three named components from the specification: the GPU (Graphics Processing Unit), the sound card, and the motherboard. Understanding what each one does — and how they connect to the rest of the system — is tested directly in the theory paper.
The GPU — Graphics Processing Unit
A GPU is a specialised processor built to handle graphics. Displaying a single frame on a modern screen can require millions of calculations — working out the colour, brightness and position of each pixel. The CPU could theoretically do this, but it would be extremely slow because CPUs have relatively few, powerful cores optimised for serial tasks.
A GPU contains thousands of smaller, simpler cores that work in parallel. This makes GPUs far better than CPUs at performing the same simple calculation (such as “what colour is this pixel?”) on many pieces of data simultaneously. As a result, a GPU can render complex 3D scenes, play 4K video and run visual effects smoothly in real time.
GPUs also have applications beyond graphics. Because of their parallel architecture, they are now used for machine learning, scientific simulation, cryptocurrency mining and video encoding — any workload that involves doing the same calculation on thousands of data points at once.
Integrated vs Dedicated GPUs
There are two configurations you must know for your exam:
An integrated GPU is built directly onto the same chip (or die) as the CPU. It shares the computer’s main RAM as its video memory. Integrated GPUs consume less power, generate less heat and keep costs down, making them ideal for laptops, tablets and everyday computing tasks like web browsing, word processing and streaming video. However, because they share RAM, they perform less well when running graphically demanding applications.
A dedicated GPU (also called a discrete GPU) is a separate expansion card that plugs into the motherboard. It has its own dedicated video memory (VRAM) — separate from the system’s RAM — which gives it much higher bandwidth for graphics data. Dedicated GPUs are used in gaming computers, video editing workstations and servers running artificial intelligence. They are significantly more powerful than integrated GPUs but also more expensive, larger and more power-hungry.
An integrated GPU is built into the CPU chip and shares main RAM. A dedicated GPU is a separate card with its own VRAM. Integrated GPUs suit everyday use; dedicated GPUs are needed for gaming, video editing and machine learning.
The Sound Card
A sound card handles all audio input and output for a computer. It performs two conversions:
- Digital to Analogue Conversion (DAC): When you play music or a game, the audio is stored as digital data (binary numbers). The sound card converts this to an analogue electrical signal that a speaker can turn into sound waves.
- Analogue to Digital Conversion (ADC): When you speak into a microphone or record an instrument, the sound card converts the analogue electrical signal from the microphone into digital data that the computer can process and store.
Like the GPU, sound cards come in two forms. Most modern computers have an integrated sound card built into the motherboard, which is sufficient for everyday use (music, video calls, gaming). Dedicated (discrete) sound cards are used by musicians, audio engineers and broadcasters who need higher quality audio, more inputs/outputs, or lower latency than the integrated solution provides.
A sound card performs Digital-to-Analogue Conversion (DAC) for audio playback (turning binary data into sound waves via a speaker) and Analogue-to-Digital Conversion (ADC) for audio input (turning a microphone’s electrical signal into digital data the CPU can process).
The Motherboard
The motherboard is the central circuit board that everything else connects to. It does not process data itself; instead, it provides the physical connections and communication pathways that allow every component to work together as a system.
The key features of a motherboard include:
- CPU socket: the slot that holds the CPU. The socket type must match the CPU model (e.g. Intel LGA and AMD AM5 sockets are not interchangeable).
- RAM slots: connectors for main memory modules. A typical motherboard has 2–8 RAM slots.
- Expansion slots: long connectors (usually PCIe — Peripheral Component Interconnect Express) into which dedicated GPUs, sound cards, network cards and other expansion cards plug.
- Storage connectors: SATA and M.2 connectors for HDDs, SSDs and optical drives.
- Buses: the data pathways printed directly onto the motherboard. The data bus carries data between components; the address bus carries memory addresses; the control bus carries control signals from the CPU. Bus width (measured in bits) and bus speed (measured in MHz or GHz) both affect system performance.
- ROM chip (BIOS/UEFI): the motherboard holds the system’s firmware, which runs when the computer starts up before the operating system loads.
- External ports: USB, HDMI, audio jacks and network sockets are mounted directly on the motherboard, connecting peripherals to the system.
A bus is a set of electrical pathways (wires or printed traces) that carries signals between components. The motherboard contains three buses connecting the CPU to everything else:
- Data bus — carries the actual data being transferred (e.g. the contents of the MDR moving to RAM). It is bidirectional: data can travel either way.
- Address bus — carries memory addresses (e.g. the contents of the MAR). It is unidirectional: the CPU sends addresses, but addresses do not flow back.
- Control bus — carries control signals from the CU (e.g. “read” or “write” signals that tell RAM whether to accept or send data). Also bidirectional (some signals go from CPU to memory; interrupt signals go from devices to CPU).
A wider data bus (more parallel wires) means more bits can be transferred in one go. A 64-bit data bus can transfer 8 bytes per clock cycle, whereas a 32-bit bus transfers only 4 bytes. Bus speed (measured in MHz) multiplied by bus width determines bandwidth — the maximum data transfer rate.
Training a machine learning model requires applying the same mathematical operation to millions of numbers simultaneously — exactly the type of work GPUs were designed for. A high-end GPU can contain over 10,000 cores, each handling a small piece of the calculation in parallel. Modern AI language models and image generators are trained on clusters of hundreds of GPUs working together.
This “general-purpose GPU computing” (sometimes called GPGPU) means that when you hear about a company building a supercomputer for AI research, the majority of its processing power comes from GPUs rather than traditional CPUs. In your exam, this context may appear in ‘Discuss Impacts’ questions about AI or high-performance computing.