Conclusion
As we've pointed out, a lot of challenges still need to be met before ray tracing can become a credible alternative to rasterization for real-time rendering. And when you think about it, is that really desirable? The advantages of ray tracing aren't really revolutionary enough to justify the impact of its cost on performance. The algorithm's strong points essentially concern reflections and transparence, since those are the two effects that are most difficult to render with current rasterization algorithms. But that's not really as big a disadvantage as you might think. The world around us is not really made up of very shiny or very transparent objects, and because of that, our eyes can easily be satisfied with rough approximations.
A look at the simulations used in recent auto-racing games like Gran Turismo and Forza is enough to realize that, in spite of the totally fake reflections on the bodywork, the overall rendering is very satisfactory. The exact reflection of a side-view mirror on the paintwork wouldn't be enough to create the impression that a major new step has been made towards photorealism.
Most people think of ray tracing as being intrinsically better than rasterization based on images generated by offline rendering engines, the results of which are greatly superior to what any new game can even dream of doing. But that impression has more to do with the confusion surrounding the ray tracing algorithm. The images people compare to those with rasterization effects are actually a combination of several techniques, such as ray tracing for direct reflections, radiosity for diffuse reflections, photon mapping for caustics, etc. All these techniques are combined to come as close as possible to the rendering equation written by Kajiya.
In its basic version, ray tracing, as far as the attempts currently being made to implement it in real time go, is suitable only for perfect reflections and hard shadows. Doom 3 proved a few years ago that it was possible to create a robust 3D engine that handles dynamic shadows perfectly with rasterization, but in retrospect, it also showed that hard shadows aren't really realistic.
To create soft shadows or diffuse reflections (like those you see in brushed metal, for example), more advanced ray tracing techniques like path tracing or distributed ray tracing are needed. But such techniques require a much greater number of rays and are still far from being feasible in real time.
Some people feel that eventually so much processing power will be available that the performance advantage of rasterization will no longer be a determining factor. Applying the law of diminishing returns, they say, the performance gain with rasterization will quickly be forgotten when compared to the elegance of ray tracing, just as the point was reached where the performance gain from programming in assembly language wasn't enough to compensate for the advantages of programming with high-level languages.
However, we're not yet convinced. In any case, we're still far from the time when we'll be able to sacrifice performance for elegance and simplicity. Just look at what's happened in the last 10 years in the world of offline rendering. While one frame from the movie Toy Story took an average of two hours to be created, a frame from Ratatouille took six and a half hours, despite processing power that was multiplied by a factor of more than 400 in between the two movies. In other words, the more processing power and resources you give artists, the quicker they'll absorb it.
If a company like Pixar, which can afford to devote several hours of processing to produce one frame, chooses to use ray tracing sparingly because of its impact on performance, it follows that the time when we'll have enough processing power in the world of real-time 3D to be able to afford to do all the rendering using ray tracing is far off. And people will surely have better things to do with that processing power.
