A CPU handles almost any computing task with a handful of powerful cores, working through jobs in sequence. A GPU runs thousands of simpler cores simultaneously, each solving one piece of the same problem at the same moment. Knowing why computers need both is essential for GCSE computer science.
What is a CPU and what does it do?
The Central Processing Unit (CPU) is the main processor of a computer. It fetches instructions from memory, decodes them, executes them, and stores results — repeating this fetch-decode-execute cycle billions of times per second. CPUs are designed for sequential, varied workloads: running the operating system, responding to user input, managing files, and executing application logic.
A modern CPU typically has between 4 and 32 cores, each capable of complex decision-making, branch prediction, and large private caches. A single high-end CPU core is extraordinarily capable — but there are relatively few of them.
What is a GPU and what does it do?
The Graphics Processing Unit (GPU) was originally designed to render the millions of pixels on a screen — each pixel's colour, brightness, and shading needs to be computed independently. Because the calculation for pixel 1 does not depend on the calculation for pixel 2, all of them can be computed simultaneously.
This led to GPUs being built with thousands of small, simpler cores — each less capable than a CPU core, but enormously numerous. A consumer GPU may have 3,000–16,000 shader cores (depending on model and generation). Their strength is processing the same operation on thousands of data points at the same time.
How do CPU and GPU differ in core design?
| Property | CPU | GPU |
|---|---|---|
| Number of cores | 4–32 (consumer) | Thousands (3,000–16,000+) |
| Core complexity | High — handles diverse, branchy tasks | Lower — optimised for simple, repetitive maths |
| Clock speed | Typically 3–5 GHz | Typically 1–2 GHz |
| Cache per core | Large (several MB per core) | Small |
| Memory bandwidth | Moderate | Very high (needed for pixel/matrix data) |
| Best workload | Sequential, varied, logic-heavy | Parallel, repetitive, data-heavy |
A useful analogy: a CPU is like a team of ten expert surgeons — each one can handle complex, individual cases. A GPU is like a factory floor of ten thousand workers each doing one assembly step — extraordinary throughput on repetitive work, poor at improvisation.
Why do games need a GPU?
Rendering a 3D scene involves computing the position, lighting, texture, and shadow of every triangle in the scene for every frame — at 60 frames per second, that might mean transforming hundreds of millions of triangles per second. Each triangle's calculation is largely independent: the lighting on one triangle does not affect the lighting on another. The GPU's parallel cores handle these independent calculations simultaneously, producing each frame in milliseconds.
Without a GPU, the CPU would handle rendering, but its few cores would be overwhelmed — frame rates would drop to single digits, making games unplayable.
Why is the GPU used for AI and machine learning?
Training a machine learning model requires multiplying enormous matrices of numbers — millions of weights by millions of data values — and doing this repeatedly across millions of training examples. These matrix multiplications are exactly the kind of repetitive, data-parallel maths that GPUs excel at.
Training a large neural network on a CPU might take months; the same task on a modern GPU takes days or hours. This is why AI research accelerated dramatically once researchers discovered GPUs could be used for general-purpose computing (a technique called GPGPU — General-Purpose computing on GPU).
When is the CPU better than the GPU?
Despite GPUs' raw throughput, CPUs remain essential for:
- Running the operating system and managing resources
- Handling user input and application logic
- Tasks that depend heavily on the result of the previous step (sequential dependencies)
- Workloads requiring complex branch logic (e.g.
if-elsechains based on unpredictable data) - Small datasets where the overhead of sending data to the GPU outweighs the parallel speedup
In practice, high-performance systems use both — the CPU handles orchestration and serial logic whilst the GPU handles bulk parallel computation.
Frequently asked questions
Does every computer have a GPU?
Most computers have some form of GPU, but not all are dedicated cards. Many CPUs include integrated graphics — a smaller GPU built directly into the same chip as the CPU, sharing the system's RAM. These are adequate for everyday computing and video playback but not powerful enough for modern games or deep learning. A discrete GPU (a separate card with its own video RAM) is needed for those workloads.
Is a higher GPU core count always better?
Not necessarily. Core count is only one factor: clock speed, memory bandwidth, memory capacity, and architectural efficiency all matter. A GPU with 8,000 newer-generation cores may significantly outperform one with 12,000 older-generation cores. At GCSE, the key concept is that GPUs have many more cores than CPUs, and this mass parallelism is their defining characteristic.
What is the connection between GPU and parallel processing?
GPU architecture is a concrete implementation of data parallelism — the same operation applied simultaneously to many data elements. This is one type of parallel processing. At GCSE, understanding that the GPU's thousands of cores can all execute the same instruction on different data simultaneously is the core concept. This is sometimes called SIMD (Single Instruction, Multiple Data) architecture.
Why would someone put multiple GPUs in one computer?
For demanding workloads — such as training very large AI models, rendering feature films, or scientific simulations — a single GPU's VRAM (video memory) may be too small to hold all the data, or its compute throughput insufficient. Multiple GPUs can work in parallel on different portions of the data. Cloud computing providers offer machines with 8 or even 16 GPUs for exactly this purpose.
Understand the components of a modern computer system in depth — Professor Turing at aitutors.me uses analogies and worked comparisons to make every concept clear.