CPU vs. GPU: what’s the difference?

What a CPU does

The CPU runs a loop forever: fetch the next instruction, decode it, execute it. It’s built to be good at anything, including long chains of decisions where each step depends on the last. Running the operating system, opening a document, and working out the rules of a game are CPU work.

Most CPUs today have somewhere between 4 and 16 cores, each a complete processor.

What a GPU does

Drawing one frame of a game means working out the color of about two million pixels on a 1080p screen, sixty or more times a second. Each pixel can be worked out on its own, so a GPU hands one to each of its thousands of cores and does them all together. It’s worse than a CPU at complicated one-at-a-time work and far better at the same simple job done millions of times.

Why AI runs on GPUs

Training an AI model is mostly multiplying huge grids of numbers, over and over. That has the same shape as drawing pixels: the same math, on a huge pile of data, all at once. In 2012, a neural network trained on two ordinary gaming graphics cards won a famous image contest by a wide margin, and AI researchers have bought GPUs ever since.

Integrated or dedicated?

• Integrated graphics

  A GPU built into the main processor chip. Fine for email, homework, video and light games.

• Dedicated graphics

  A separate, much more powerful GPU with its own memory. For modern 3D games, video editing and running AI tools yourself.