Six Myths About Ray Tracing
Created on May 6, 2012, 6:21 p.m.
For a few years now people have been holding up Ray Tracing as the next big thing for computer games - saying that once we reach the processing power needed all games will be Ray Traced and we'll finally get cool effects like reflections and global illumination.
Many people have misunderstood notions of Ray Tracing and I don't blame them. It is a word which can mean a lot of things. In an attempt to give some clarity - here are six myths about Ray Tracing and the reasons why they are not true.
1. Ray Tracing is what they use in animated movies
Most CGI in movies from Lord of the Rings to Finding Nemo is rendered using RenderMan, a program developed and maintained by Pixar.
RenderMan has a huge array of tools but by far the bulk of the work is done via a normal Scanline Renderer which supports dynamic subdivision and programmable shaders - much like is seen in modern video games. Whole scenes are rendered purely using this system and only for particular effects are the alternatives rolled out.
For global illumination, reflections, and other complicated effects RenderMan provides a type of Photon Mapping. The scene is subdivided recursively into blocks and photons emitted. Then there is a Ray Collection phase that happens independently for each block.
This system allows them to work on scenes with large polygon and texture data which normally would not fit in main memory (the main reason standard ray tracing was not an option).
Pixar show us that great looking results don't automatically mean Ray Tracing, which has major (though not insurmountable) issues with memory and cache.
2. Ray Tracing scales better than Scanline Rendering
In general Ray Tracing is well known for giving diminishing returns. For an incremental improvement in accuracy, Ray Tracing requires you shoot out twice the number of rays. Even with extensions like Photon Mapping this still holds to some degree, and holds completely when we talk about memory and cache costs.
It is also common to hear that Ray Tracing is better suited to parallelism because each pixel is computed individually. Scanline rendering can only compute in parallel the pixels for each triangle in turn. This is true, but each ray trace is still propotional to the number of triangles in the scene and in reality computing each pixel absolutely in parallel only seeks to destroy the cache. [More]
As well as being kind to the cache, Scanline Rendering has a large number of parallel stages which Ray Tracing lacks. This means Scanline Rendering can be pipelined and throughput is maximised. Scanline Rendering may not be able to capture such a broad range of effects so easily, but it really does have a remarkable computational complexity and in practice Ray Tracing isn't any better.
3. Ray Tracing is more physically accurate
This is a difficult one. Light, materials, and the physics involved are all so complicated that the physical accuracy of any of our models used in current rendering is generally pretty bad.
The way we imagine Ray Tracing to be more physically accurate, is that we like to imagine a ray of light being traced through the air and intersecting with a surface. This is already the simplest part of the model, and a Scanline Renderer does it too, just in a different way.
The complex part of rendering is what happens to the ray after it hits the surface. For this none of our models are really that accurate and we use the same models for Scanline Rendering as in Ray Tracing. The truth here is that Ray Tracing is much more algorthmically simple than scanline rendering, which although obviously a huge advantage to implementation, doesn't mean it is more physically accuracy.
4. Ray Tracing allows for more dynamic elements in a scene
Unfortunately the opposite of this is true. Effective Ray Tracing requires you hold large structures subdividing the scene into sections. This has a terrible price to pay on memory but even worse is that if any movement occurs in the scene these structures have to be updated; a non-trivial task.
Similarly, smoke, fire, water and other weird effects lend themselves no more naturally to Ray Tracing than to Scanline Renderer. They have an odd set of material properties which cannot be easily simulated even via physics and often an artist approximation is best.
5. Ray Tracing can provide Unlimited Detail
Unlike the image which is painted by Euclidean Games, the Graphics Programming community is open and sharing. Just look toward SIGGRAPH - one of the biggest conferences in the world. This idea of a secret industry investment in polygons and rejecting Voxels is just crazy.
It seems Euclidean Games, unlike their contemporaries, had no intention of sharing their creation or publishing a paper (something very common for Graphics Programmers in Game Development) - and instead are happy to make patronising comments about developers use polygons on youtube.
Without going into the technical details (which I'm sure you can look up) Unlimited Detail has recieved crazy hype and should've been ringing your "too good too be true bells" before the video even started. Unlimited detail has all of the issues I've outlined above and more. I think it will be a long time before we see a full product using the technology.
6. Ray Tracing is The Future
It really is hard to tell what the future will be - and Ray Tracing is certainly a key thing to have in any Graphical Programming toolkit. But polygons and Scanline Rendering have served us pretty well so far and I'm sure they have a large number of years left in them.
As with most things, what will probably happen is there will emerge systems using a healthy mix and mash up of techniques with an added blend of nothing we've seen before. The future resides in algorithms, not buzzwords, and I'm sure it will be exciting whatever it is.