Start here · Lesson 1 of 18

Why use a GPU?

A CPU runs your app. A GPU helps with lots of similar calculations at once.

Open the animated version

How it works, step by step

  1. Start with four sums

    Match numbers in the same place. These four sums are independent: none needs another sum's answer.

    1 + 10, 2 + 20, 3 + 30 and 4 + 40.

  2. Share out the work

    The CPU gives the GPU the job. The GPU can work on separate sums at the same time.

    All four sums can make progress without waiting for one another.

  3. Put the answers together

    The app now has all four answers. AI models do far more number work, which is why a GPU can help.

    The completed list is [11, 22, 33, 44].

In plain language

A CPU runs your apps and helps organize their work. A GPU works with it to handle many similar calculations at once.

A way to picture it

One cook organizes a meal while a team chops lots of vegetables. The team helps most when there is plenty of work that can be done at the same time.

A worked example

Solve four separate sums: 1 + 10 = 11, 2 + 20 = 22, 3 + 30 = 33 and 4 + 40 = 44. The answers are [11, 22, 33, 44]. No sum needs another sum's answer first.

Keep in mind

CPUs can also do several jobs at once. A tiny job may finish sooner on a CPU because preparing GPU work and moving its data also take time. These four sums are a teaching example, not a speed test.

What these words mean

CPU
The general-purpose processor that runs apps and helps organize work. CPUs can also do several jobs at once.
GPU
A processor built to handle many similar calculations at once. Setting up a tiny GPU job can take more time than it saves.
Parallel
Different pieces of work making progress at the same time.
Independent
A piece of work can be done without waiting for another piece's answer.

Which work can be shared out most easily?

Four sums that do not need each other's answers. Each separate sum can be solved on its own. CPUs can do this too. Four sums teach the idea; they do not show that a GPU is faster.

Where this comes from

This explanation is checked against primary documentation. The small arithmetic examples are ours and are not hardware measurements or vendor benchmarks.

Read the primary source