GPU Paint Race One fast painter or 64 slow ones? Race them and see why AI runs on GPUs.

1. How to play

   • Read the job, then pick a crew: CPU or GPU. On job 4 you can also split the job between them.
   • Press Race. Both walls fill at the same time. Skip to end jumps to the finish.
   • The lights come on with the tick counts. Pick the winner to clear the job; Race again to try another pick.
   • Toy numbers, so you can count them: the CPU paints 1 pixel a tick. The GPU crew has 64 painters; each step they all paint one pixel together and it takes 4 ticks, plus 16 ticks to send the job over first. Real GPUs have thousands of painters.

2. What the race teaches

   • A CPU has a few fast, clever cores. A GPU has thousands of simpler ones that all do the same step at once.
   • Jobs where every piece is the same small sum, like pixels or AI math, split across the GPU's painters.
   • Jobs where each step waits for the one before don't split. Extra painters just wait.
   • AI models are millions of multiply-and-add sums, which is why they train and run on GPUs.

3. No screens? Play it unplugged

   • No screens? Draw a 10×10 grid. One player colors every square alone while the class times it.
   • Cut a second grid into strips, one per player. A caller says one color at a time, everyone colors only that color, and nobody moves on until all hands are up. Time it.
   • Now a trail: each square gets the number of the square before it plus one. Split it into strips and see why the team can't go faster.

Downloads

• Grown-up guide

  Plain paper · 1 page · 8.5 × 11

  How to play, what it teaches, the tick count for every job, and a no-screens version with a paper grid and a caller.

  Open Download