NVIDIA - Ray Tracing and Rasterization

Cool:

http://www.pcper.com/article.php?aid=530

is raytracing the be-all and end-all of graphics then. IS that the only thing we're missing for full photorealism in games?

I think real-time raytracing in games will happen at some point in the near future.

If it doesn't, then I never said that.

yeah, I think it will.

The videos show it being done using qukae3 in real time. It's only a matter of time before GPU manufacturers upgrade the gfx cards to handle this. The reason CPU's haven't been able to do it is due to them being multipurpose chips, having the GPU do this along with the other graphical effects will be much more effifcient.

As long as we can have raytracing along with all the effects we're getting now in games like crysis and cod4 then we'll be laughing.

Q3 in realtime was 50ghz of processing power so no you won't see it that soon.(36 months at best if the modified moore's law holds)

CUDA is definitely a great possibility to offload the stuff onto a very math oriented platform which is high performant. Problem is that there are still ATI users and those would remain on regular shaders.

So I think its more likely that a hybrid approach will become standard which uses addition CPU cores + pixel shaders for raytrace and rasterization alltogether. At least for the gaming market.

3D rendering suites most likely will integrate it for the NVIDIA users which will give ATI a fair breakin unless they adapt CUDA or have a similar framework

Games running in real time raytracing, that must be for year 2020...

Not really ...
An 8800GTX has 16 parallel high performance floating point cpus which can handle 4 threads parallel (real not virtual) ...
Problem is just that CUDA is quite new and it is totally different programming wise than a regular CPU. But once its concepts are understood well enough, I would guess that even a 9600GT could more or less "easily" handle regular raytracing (without nifty features) in realtime.

Hmm, don't see the point in it really. There's probably always going to be faster ways to fake most effects in games without full blown ray tracing. Seems like a waste or resourced at least for games. But great if people use the GPU to speed up regular pre rendered stuff.

There's already a renderer that uses the GPU to great effect, a year or so ago a 7xxx nvidia card was about 8 - 11 times faster than the high end dual core CPU's of the time.

I expect people to fake effects and make better use of the GPU for for physics and materials in games for quite some time.

I hope they start to use it for usefull stuff.

More realistic games = yawns-ville. :) Having said that, if it means that you don't have to jump through the GPU hoops to get desired effects (water,shadows etc) then that would be cool. Code like this would be ace:

scene = LoadMesh("scene.x")
LightsOn(scene)

Beaker, that is pretty much how Leadwerks Engine 2.0 works. The lighting is as simple as Blitz3D lights, but with shadows.

I think this is the only advantage of raytracing - you can very easily get realistic lighting working, just as long as material settings, etc. are set up. The problem is, with this simplicity comes a massive reduction in performance, when compared to an almost identically appearing rasterized scene (according to John Carmack, at least).

I think non-hybrid raytracing will never become popular until the computing power of CPU/GPUs far exceed the requirements of anything you can possibly imagine wanting to render. Otherwise there will always be visual quality benefits to rasterization since you can get a lot more detail drawn with the same number of computations.

I don't think I'd like all of my games to be photo realistic. I would prefer to gave better game play rather than photographic look.

Barney

Just imagine GTA V with photo-realistic graphics.

Oooh, slapping baby oil on at the thought of that.

Making games photorealistic is merely an exercise in artistic skill. Raytracing any given existing game would only show up flaws in design and give an overall worse visual experience. Raytracing Quake, for instance, would leave a lot of areas poorly lit, or overlit. You need to really think about lighting, materials and colour perception to make it work without it looking bad.

You'd need top-notch artists being able to play with at least 1GB of working area for textures and vertex data to pull off a really good looking raytraced game, if photorealism was the aim.

Raytracing simpler, more stylised graphics could lead to some interesting visuals, but the market for such things is so tiny that it's not going to drive GPU hardware in that direction for some time.

If raytracing goes realtime, it will likely be as a side-effect of more efficient GPU design rather than an end result of that design process.

If it's stylized graphics you want, I think rasterized rendering (pixel shaders, etc.) is far superior to raytracing, since you can program it to look any way you want. Raytracing, on the other hand, attempts to simulate light realistically, which would not fit well to cartoon-style games, for example.

I think just because of the constant relentless march of `whats the next big new thing` we will see raytracing more and more despite some people's aversion to it and eventually I'm sure will be as commonplace as the whole polygon thing is now. I'll give it a few more years though.

John: then you never had Gile[s] run on Global Illumnation with Cartoon Lights :)
It gives a really nice style and you can set the number of light levels :)

I do not think Ray-tracing will be used in real-time for another 20 years, if ever. It does not scale well with geommetric complexity. A rasterizer only has to deal with a set number of fragments, but as geometric complexity increases, raytracing processing demands will increase linearly.

In other words, if you have twice as many polys in the scene, part of the raytracing routine will be twice as slow.

Don't be so sure Josh :D, think of how far computers have come in the last 20 years.