The big idea: A CPU has a few powerful cores built to run one instruction stream as fast as possible.
A GPU has thousands of simple cores built to run the same operation on thousands of values at once.
Neither is better. They are built for opposite shapes of work.
| CPU | GPU | |
|---|---|---|
| Cores | A few — typically 4 to 16 | Thousands, each far simpler |
| Optimised for | Latency — finish one task fast | Throughput — finish many at once |
| Handles branching | Very well — branch prediction, deep caches | Poorly — cores run in lockstep |
| Suits | Operating system, logic, decisions | Graphics, matrices, machine learning |
| Control logic | Large share of the chip | Small — more room for arithmetic units |
Free preview
This is the free notes preview
You're reading the free notes. Aimnova Pro unlocks the full study experience — and you can try it with your first topic free to keep:
- FlashcardsLock in vocabulary and key terms with spaced repetition.
- Practice questionsAnswer exam-style questions and get instant AI marking.
- Mock exams & past-paper vaultSit full mocks and see exactly how examiners award marks.
- Personalised study planA daily plan built around your exam date and weak areas.
The CPU's problem
- Code is full of decisions — if, while, function calls
- The next instruction often depends on the last result
- So it spends transistors on prediction and caching
- Making one stream fast is the whole design
The GPU's problem
- Shade a million pixels · multiply a large matrix
- Every element is independent of every other
- So thousands of simple cores can each take one
- No prediction needed — there are no branches to predict
The consequence
- A GPU is useless for sequential, branch-heavy code
- A CPU is slow at a million identical operations
- Real systems use both, for the work each suits
Independence is the test: Ask: can these operations be done in any order, at the same time, without needing each other's results?
Yes → a GPU can help. No → more cores change nothing, and the CPU's single-thread speed is what matters.
Feeling unprepared for exams?
Get a clear study plan, practice with real questions, and know exactly where you stand before exam day. No more guessing.
Machine learning is the current example
- Training is dominated by matrix arithmetic, and every element is independent
- So a GPU can be orders of magnitude faster than a CPU at it
- Inference — one input, one pass — is far smaller and often runs fine on a CPU or a phone
- Which is why training needs a data centre and using the model does not
Do not say "a GPU is faster": It is faster at parallel work. On a single sequential task a CPU core beats a GPU core comfortably, because the GPU core is deliberately simpler.
The qualification is usually the mark.
How this is tested — you must explain the difference from the SHAPE of the work, not from raw speed. It comes up two ways:
Paper 1 Section A
- Explain the differences between the CPU and the GPU, 3-4 marks
- State one task suited to each
- Give one architectural difference
Paper 1 Section B — case study
- Justify hardware for a described workload
- Explain why more cores do not always help
The classic trap: Answering "a GPU is faster". It is faster only where the work is parallel and independent. On sequential, branch-heavy code the CPU wins — and saying so is what earns the marks.
A company trains a machine learning model and also runs a database server. Explain which processor suits each, referring to how the work is structured.
Model answer plan
See the mark-by-mark plan — for / against / judgement, with marking guidance — in study mode.