Rendering

A computer does not store a three-dimensional scene as an image, but as numbers. It records, for example, where the corners of an object lie, what color its surface has, and from where a lamp is shining. A screen cannot do anything with that. It only displays a grid of millions of tiny colored dots, the pixels. Rendering is the computational step that turns the numbers into exactly this grid of dots. You can imagine it like a photographer taking a picture of a set that has already been built: the set already exists, the image is only created through the shot.

Why every game and every film depends on it

Everything that moves on a screen and looks three-dimensional must be rendered. In a video game, this happens while you are playing. For the motion to appear smooth, at least 30 finished images per second are needed, or 60 or more for fast-paced games. The computer therefore has less than a fiftieth of a second per image. This is the reason why graphics cards are so expensive and so powerful.

For animated films, the opposite holds true. There, a single frame is allowed to take even hours, because no one is waiting for it. Large studios therefore run entire data centers for weeks on end, so-called render farms. The difference is called real-time rendering versus offline rendering. Both do the same thing, but with completely different time budgets and therefore very different image quality.

Economically, rendering is a significant cost factor. Computing time costs electricity and hardware. If a method achieves the same image quality with half the effort, a studio saves money directly. That’s why rendering techniques regularly show up in the quarterly reports of chip manufacturers.

From the triangle list to the finished pixel

Almost every 3D object consists of many small triangles. A car tire can be made up of several thousand of them. In the classic method, rasterization, the computer calculates for each triangle which pixels it covers on the screen. It then determines the color of those pixels. This is very fast, because many triangles can be processed at the same time.

The second major method is called ray tracing. In this process, the computer sends an imaginary ray of light from the virtual camera into the scene for each pixel and tracks what it hits and how it bounces onward. This produces reflections and shadows almost automatically, since the method imitates the physics of light. The price is enormous computational effort, because a single image requires millions of such rays.

This is where artificial intelligence comes into play. Modern graphics cards deliberately render an image at a lower resolution and let a trained model fill in the missing details. This upscaling is called, well, upscaling. Some systems even invent entire intermediate frames that were never classically computed. A common misconception is that this filling-in is itself already rendering. It is a downstream step that builds on an already rendered image.

From mobile games to furniture apps

In everyday life you encounter rendering constantly, usually without it being named as such. Every mobile game renders, every map view with 3D buildings renders, and so does trying on glasses in an online shop. Furniture retailers have long since stopped depicting their catalogs photographically. The sofas in the images are rendered models, because that’s cheaper than a photo studio.

In the news, the term mainly comes up around graphics cards and gaming consoles. When a manufacturer claims its new card delivers twice as many frames per second, that’s about rendering performance. Pay attention to whether the figure was measured with or without AI upscaling. That makes a big difference and is often hidden in the fine print.

Rendering must be distinguished from pure image generation by AI. An image generator creates an image directly from a text description, without a 3D scene ever having existed. Rendering, on the other hand, always presupposes an existing model that someone has built. The advantage of this is control: whoever owns the scene can later rotate the camera by ten degrees and show exactly the same sofa from a new angle.