When Will Ray Tracing Replace Rasterization?

Good article. Rab1d-BDGR: the picture is most definitely CGI - too clear and clean to be a picture - this was a big thing back in nineties when I was still doing 3D using Imagine 2.0, Lightwave 3.5 on an Amiga 4000/040/25MHz (for a simple scene consisting of roughly 1000-2000 polygons - one frame 640x512 took about 30-60 minutes)... Anyway, the biggest problem with 3D rendering is that nature is not perfect - that's why new games still look "unrealistic" - they are too perfect, and the only thing that can help simulate the mother nature in current 3D engines are the dirty textures - which might be better or worse - hence why they will never be perfect by definition. Another thing is that it does not matter how perfect the engine will become - without viewing the effect in "3D" our brain will know it's "different" and therefore fail to produce "reality".

I dissagree with your last statement, as watching "Real" TV would seem unrealisitic, irrelevant to whether it had effects in it or not. I totaly agree with the dirtying of textures suggestion, and i think a few games companies are now starting to take this aspect seriously, As far as realism in Games, They should aim to simulate Movies rather than reality, as the majority of big feature films aren't accurate in anyway, the physics are always overblown and the effects normally much to big for the actual action. One of the most Cinematic games of late is Call Of Duty - Modern Warfare, which although technically no where near the likes of Crysis, actually "Feels" more real when its moving.
Raytracing won't become the normality in games for many many years, we are still rendering sequences which take over 3 hours per frame for a HD1080 movie. matching that fidelity on a single graphics card in realtime isn't going to happen soon.
The biggest steps forward we will be seeing will be in texture optimisations and geometry deformation, these will make the games more immersive and therefore suspend disbelief longer.
The other main factor, is the majority of Games Render pipelines DO use ray tracing, just not in realtime. Its all prebaked into either textures or pointclouds which can be accessed in realtime.
For anyone interested in highend game rendering, looking at spherical harmonics shows just how close you can make games look with baked light clouds - the Turtle renderer (used for mirrors edge and a few other titles) has a great realtime demo app showing off its ability.

The glasses/dice picture is definitely CGI - as someone who has been making stuff like this for years you get a 'feel' for what is real and what is fake, that picture just feels cgi - the surfaces are all pristine and the ambient light doesnt seem real, the dead give away is the ice cube though - in real life different parts of the cube would have different indexes of refraction based on how the cube froze, that cube is a uniform optical density.

Well, this article is interesting but it's very Renderman-centric. What about Mental Ray? It is built from the ground up as a Raytracer, unlike Renderman where it was tacked on afterwards. They each have their advantages, REYES having a distinct advantage at high detail geometry and 'post-effects' like depth of field, but Mental Ray seems a nearer fit than Renderman since it ray-tracing tends to have a greater return on improved threads etc.

There is another way to determine if the picture is GCI or real; what is m missing. And 9 out of 10, it's 'dirt'. In the glasses picture, for example, the glasses are sparkling clean. Yet they have been used. There is no moisture on the inside of the glass, there are no palm prints on the outside. And there are no lip marks, let alone lipstick marks.

Ashtray is unused, the table top is gleaming. But everything is consistenly clean. My kitchen surface is clean. But at eye level, there are variations. My glasses are clean, but the glass is uniformly sparkling.

I haven't read the article absorbing every word, and know very little about the subject. However, from "back in day" when r-t was the current buzz word (and we could do it on our Amigas), I do recall that different objects/surfaces had different values of reflectivity, which would be consistent across an object surface. Seems to me perhaps a small tweak would be to assign a surface map (matrix) to an object, with the refraction and reflection values the cell values. That would allow variation to be introduced across a surface. And if such maps were applied as layers, or masks, then one could model lipstick smudges, palmprints, etc.

Anyway, I'm not even an informed beginner so I'll go back to marvelling at the patience and skill of the designers(?) artists(?)

As someone who's currently working as a rendering coder in a game development company, I can only say: spot on, great article!

light levels in rasterised games are to flat. Ray tracing allows for a more realistic/dynamic treatment of light and dark.

How about reflected light? I.e. light reflecting from a surface into a shadow. If I read this correctly the secondary rays only trace back to light sources themselves to determine if the pixel receives a light from that source, but each pixel could actually have multiple paths of refelected ligth coming from a single source (e.g. via concave mirror). How it has been presented in this artice sound very simple, since the paths are here limited to finite number (primary and secondary rays) but when some of the secondaries can actually number infinite it would become somewhat more difficult... If you look at photo 19 (the blue car) you can see that the front bumper doesn't reflect any of the light back upwards in to the gap on top of it as it should. This is usually very visible in many of the ray traced examples I've seen so far.

Good article, truth is that ray tracing has some bad-to-solve issues and the fact of the matter is that rasterization is well understood and has clear pipelines in modern hardware, ray tracing will be increasingly used in the future, but it is not even starting in graphics, I've used a simple version of it to test surroundings -throw twenty rays in all directions, how many touch within five meters- for ambient occlusion and so on.
About storage structures, you guys should keep in mind that most physic systems require structures that are _very_ ray-tracing friendly, that particular aspect is not going to be a big problem.

The picture of the glasses & dice is from POVRay - http://hof.povray.org/

cube though - in real life different parts of the cube would have different indexes of refraction based on how the cube froze, that cube is a uniform optical density.