What is a Voxel Renderer?

I often see both scholars and novices alike point out what a voxel engine is and what it is not. 'Minecraft does not use voxels', 'Resogun does not use voxels', and '3D Dot Heroes does not use voxels' is tossed around a lot. I agreed with them for a long while, basing my opinion on the fact that many, if not all of the examples thrown around are hardware accelerated and as such are rendered using triangles as rendering primitives. After actually writing my own voxel renderer I came to a different conclusion about what constitutes a voxel renderer by separating the term 'voxel', which is a spatial data structure, from 'renderer', which is an algorithm used for the visualization of spatial data. In this post I will attempt show the logic of that argument by examining the individual terms of 'voxel renderer' with the hopes of showing people that the concept is far less binary than originally thought.

The first component of the term 'voxel renderer' is of course 'voxel'. Thus, we should have a look at what the basic concept of voxels means. Simply put voxels are volumetric data stored in a three dimensional grid, meant to convey a three dimensional image. It can therefore be said that a voxel is to three dimensions what a pixel is to two dimensions. Therefore, a way to think of a single voxel is to imagine a singularly colored, axis aligned cube (although it is ultimately the renderer that decides how to visualize the input data). It should also be noted that a cube is really only a convenient way of imagining a voxel. A cube is a volume defined by six points in an unrestricted three dimensional space, while a voxel is a single grid cell in an axis aligned space. As such cubes are free to move around the space, while a voxel is forever confined to the location of the cell in which it resides. Because of this we can really only say that a voxel is represented by a cube rather than actually being a cube.

Voxels can often contain more data than a color. They can also contain many properties that describe the surface that the voxels are representing such as surface normals and roughness. However, voxels will not need to contain information about position or size since their position is defined by their relation to one another in the voxel grid, while their size is homogenous along the individual axis. However, voxel volumes can be made to rotate and move independently of other, separate voxel volumes by attaching individual transformation parameters to each volume for the engine to regard when handling the data.

Given the information provided above it should be obvious that any game storing spatial data in a three dimensional grid can be said to use voxels in one form or another. Yet, another interesting facet of voxels is the rendering of voxels. A game engine encompasses many components that work together in tandem. Undeniably, one of those components is the renderer, which is often the single most characteristic and generally identifiable component in an engine. 'Renderer' is an umbrella term for all kinds of visualization algorithms; rasterizers, ray casters, ray tracers, splatters and so on, but it also has a much broader significance by encompassing things like transformation, hidden surface removal, and lighting and shadowing amongst other things.

Taken at face value, 'voxel renderer' does not make any pretense to dictate the actual algorithm used in the rendering process. In fact, it does not dictate anything except the predominant use of voxels in the process. A predominant use of voxels should be an obvious prerequisite when determining if a renderer is a voxel renderer since the term voxel is used as a classification of the renderer. Classifying by something other than a main characteristic feature would be misleading. However, 'renderer' is a much too open ended term to actually mean much (try substituting the word 'pixel' for 'voxel' in 'voxel renderer', and it should become abundantly clear just how without any meaning term is). Additionally, 'voxel renderer', as it stands, has not been used in such a was as to load it with a specific meaning regarding what method of delivery is employed in the rendering process. Therefore, the only prerequisite for a correct use of the term 'voxel renderer' is then the use of voxels as the main source of data used for rendering.

This leads to the question 'when are voxels considered the main source of rendering information'? It is common knowledge that modern day graphics cards provide an ad hoc solution for rendering triangles. Barring the use of OpenCL or the like, any rendering algorithm that seeks to accelerate the process via hardware has to ultimately render primitives as triangles. However, a voxel is converted into triangles merely as a final step to in the rendering process, which arguably is the smallest segment of the rendering pipeline. Aside from that, the mere fact that a voxel is rendered as a set of triangles does not mean that the end result is not a voxel. Rasterization is merely a delivery method for rendering voxels and should not, in and of itself, exclude a renderer that otherwise predominantly uses voxels from being considered a voxel renderer.

In conclusion, voxels are a form of spatial data, while rendering is a form of visualization algorithm. Voxel rendering is a form of algorithm (unspecified) used for the visualization of voxels. Thus, many different things can be considered voxel renderers, as long as there is a predominant, albeit not necessarily persistent, use of voxels throughout the renderer so as to not undermine the classifying term. Other than that, the definition is very open ended and may be used liberally until practice establishes a narrower use for the term.