Embarrassingly Parallel - Intro to Realtime 3D Computer Graphics

Many video games and engines can support amazing three-dimensional enviroments and models consisting of millions of triangles shown on screen at the same time, while at the same time managing to process input, handle physics and logic, and render a new frame in (ideally) less than 20 milliseconds. Naturally, this involves a ton of optimizations and little tricks, ranging from more simple things, like only rendering parts of objects the virtual camera can see, up to more complicated things, like reducing the complexity of (more distant) models without affecting their visuals as much. By far the biggest idea that enables rendering these massives scenes at decent framerates is simplifying the problem as much as possible, such that it becomes so easy to distribute over many parallel processing units, or simply "embarrassingly parallel". Combined with modern GPUs, which usually feature thousand of relatively simple parallel processing units, this is essentially what realtime graphics are: break down everything into simple geometric primitives (triangles) and fake the rest as best you can. We will cover the basics of a 3D rendering pipeline, from projecting triangles to the screen, mapping them to pixels (rasterization), and applying additional effects (shading). This will be somewhat technical, but hopefully still digestible, and as a treat, there will be lots of eye candy. Additionally, we will also take a look at how more complicated effects, like shadows, are realized in these pipelines. Finally, we will peak under the hood of GPUs: how they work and what they can handle best.