Many calculations at once

Some work has a different shape.

To draw a picture, a computer performs similar calculations for many pixels. Most of those calculations do not need to wait for the one beside them.

The graphics processing unit, shortened to GPU, is built to handle many pieces of similar work at the same time. This is called parallel work.

Imagine one painter colouring every tile in a mosaic, then imagine hundreds of painters each taking a tile. The second group can finish much faster when the tiles can be handled independently.

This will matter later. Training AI systems involves enormous amounts of similar mathematics that can be done in parallel.

Many calculations at once

Some work has a different shape.

To draw a picture, a computer performs similar calculations for many pixels. Most of those calculations do not need to wait for the one beside them.

The graphics processing unit, shortened to GPU, is built to handle many pieces of similar work at the same time. This is called parallel work.

Imagine one painter colouring every tile in a mosaic, then imagine hundreds of painters each taking a tile. The second group can finish much faster when the tiles can be handled independently.

This will matter later. Training AI systems involves enormous amounts of similar mathematics that can be done in parallel.