Concepts explained
Why the Drawing Chip Became the Brain
A part built to draw game screens now runs AI, folds proteins and forecasts weather. Not by accident: drawing and those calculations turned out to share one shape — a simple job, millions of times over, all at once.
One chef, thousands of helpers
The processor of piece five — the CPU — is one highly capable chef. It can cook anything and shines at work with a complicated order: simmer the sauce, taste, change the next step as things develop.
Drawing a screen is a different sort of job. Millions of dots need colouring, each dot's sum is simple, and — decisively — the dots have nothing to do with each other. Painting the top-left needs no result from the bottom-right.
That is how a game screen can be redrawn whole, dozens of times a second: one helper per dot, near enough.
Scientists appear at the game shop
In the early 2000s, odd customers began buying gaming parts: physicists, astronomers, people doing financial sums.
What they had noticed is the kernel of this piece: a great share of the world's big calculations have the same shape as drawing. A galaxy simulation repeats one small sum over millions of stars; a weather forecast repeats one over millions of cells of sky. Simple job × millions, mutually independent.
The trouble: the gaming chip understood only one request — 'draw'. So the early scientists disguised their sums as pictures: numbers dressed up as colour data going in, the rendered 'image' read back out as numbers.
The chip company watched this happen. In 2006 it opened a door for giving the chip calculations straight, no disguise. The moment the drawing chip officially became a calculating machine.
A little further in
Why not just add more chefs?
A question remains: why not simply pack in thousands of capable chefs (CPUs) instead?
Because some work refuses. A thousand chefs cannot boil one pot of soup any faster; boiling time passes in sequence only. The world's tasks split into work that must queue and work that can be shared out at once — the former wants one fast chef, the latter wants many hands.
And one more customer arrived
In the early 2010s a final customer found the helper brigade: people working on artificial intelligence. The artificial neural networks they were building turned out, on inspection, to be multiply-and-add × billions, mutually independent — precisely the shape a GPU loves.
In 2012 a program trained on two gaming chips upended an image-recognition contest. That story — how a machine learns — is the next piece.
The question that remainsDrawing explosions, folding proteins and learning language all shared one computational shape. Is the world simply built that way — or have we merely become able to solve only the problems of that shape?