"Ray tracing" that we hear a lot these days What the hell is this? (ITmedia PC USER)

The world's first RT-compatible GPU "GeForce RTX 20 series" (photo is GeForce RTX 2080 Founders Edition)

In 2018, NVIDIA announced the GPU "GeForce RTX 20 series" that supports real-time ray tracing (RT). This has brought about a major “change” in the graphics pipeline since the “programmable shader” that was born in 2000. [Image] The principle of ray tracing And in 2020, the RT function was finally installed on home game consoles. Yes, "PlayStation 5 (PS5)", "Xbox Series X" and "Xbox Series S". In the same year, AMD, which provides APUs (GPU integrated CPUs) for these game consoles, also introduced the RT-compatible GPU “Radeon RX 6000 series”. Although it is about two years behind NVIDIA, it is significant that the two major PC GPUs support RT. And this year in 2022, NVIDIA released the "GeForce RTX 3050" and AMD released the "Radeon RX 6500 XT", an entry-class RT-compatible GPU. The impression is that the “popularization of ray training” is finally in full swing. Well, what kind of technology is "ray tracing" in the first place? And what kind of change and evolution will it bring to game graphics? I would like to explain this in several parts.

The conventional 3D rendering method is the "rasterization method"

First, I will briefly explain the rendering (generation) method of 3D graphics. All GPUs installed in current PCs, mobile phones, game consoles, etc. basically render 3D graphics using a method called "rasterize". This name comes from the process of decomposing (rasterizing) polygons into pixels on the screen when drawing a 3D scene constructed using polygons (triangles) in units of polygons. Pixel groups decomposed from polygons are subjected to shading processing in a "pixel shader" in conjunction with lighting (light source) calculations and textures. This fixes the color and outputs it to the screen. In this processing system, 3D objects outside the screen are treated as "nonexistent" because they are not subject to processing. This is the greatest advantage of the rasterization method, and at the same time it is also its weakness. Let me explain with an example. Suppose now that there is a warrior facing the front on the screen. In the rasterization method, the front of the warrior is within the field of view from the viewpoint (camera), so it is drawn properly. On the other hand, since its back is out of sight, it is excluded from the drawing target as a "non-existent thing". Furthermore, the part of the mage's body that overlaps behind this warrior is also treated as "non-existent" (this is called "Early Culling"). Also, since the rasterization method does not have the concept of "indirect light", it can only perform lighting from direct light. "Shadows" as the amount of direct light hit can be created automatically, but "shadows" created by a third party's shielding cannot be created. Basically, the rasterization method is a drawing method that completely eliminates the mechanism for processing "involvement from third parties". Therefore, even if you can express the highlights of the reflected light as direct light on the warrior's armor, you can't express the "mirror image" of trees reflected in the armor. At this point, some people may be thinking, "Huh?" Recent game graphics have shadows, indirect lighting, and even mirror images. In fact, these elements seen in current game graphics are drawn "separately" using the rasterization method using the GPU each time.