11 replies to this topic

[dj]

I have just noticed these pages:
http://www.heise.de/...r/meldung/35456
http://www.openrt.de/
http://graphics.cs.uni-sb.de/RTRT/

which talk about a raytracing algorithm that is physically correct and is around 30x faster than normal ray tracers. It was shown in the C-bit expo and was running on a cluster of 11 athlons at a framerate of 10fps. Does this mean that we will eventually be able to have games based on ray tracers if optimized even further?

[anubis]

Check out http://www.saarcor.de. Since some time the university of Saabrücken is working on a prototype for a ray tracing GPU. The papers for OpenRT are available (check out the publications page on openrt.de) and some people have already implemented real time ray tracing on common cpus (there was an IOTD on flipcode some time ago, which unfortunatly is inacessible right now).

To conclude : It's quite possible and has been done. It is not even that hard and the papers are pretty straightforward, if you which to try it yourself.

[Reedbeta]

In fact, raytracing has even been done on current GPUs, although it is currently not significantly faster than CPU raytracing, has low image quality, and is limited to tracing voxel grids. See this page for more info.

[bramz]

bah, RTRT is nothing new! :) ever seen the heaven 7 demo? or the nature sucks demo's? _that_ was cool! And even check out my own little RTRT demo: http://users.skynet.be/bdegreve/projects/p...acer/index.html nothing fancy I know, but hey, I'm still proud of it ;) But yeah, those links you showed are really cool. It is fascinating to see how hardware is catching up.

Bramz

[anubis]

The guys that did nature sucks claim that they have real time GI in the works (I think they said something about a release this summer). I'm looking forward to seeing this. The Realstorm benchmark was cool, too.

[Crash]

I'm a big fan of real time ray-tracing, but I don't really see it as being a viable solution in anything less than 8 years from now.

-M

[Crash]

Crash said:

I'm a big fan of real time ray-tracing, but I don't really see it as being a viable solution in anything less than 8 years from now.

-M

Now that I think about it, even that is probably overly optomistic :sad:

[davepermen]

this was rather pessimistic. www.intrace.com IS selling. they DO make money.. it is the paying back-support for the openrt.de and saarcor etc developments.
yes, it's clusters, but it's payable, at least for companies that want it. and with the hw in development, it'll be quite soon that there won't be clusters anymore, but SLI-style systems with 1, 2, ...... PCIe cards in a pc/server that trace at the same speed..

it just depends on whats your use of it. for games, on your home pc.. not that soon _except_ there will be a REvolution. as evolution won't get to raytracing anyways. and REvolutions are something that normally don't happen in capitalistic systems. the gpu vendors will happily continue where they are, and won't bother trying something really new.

thats the only reason it's not yet there.. as long as rastericers sell, they have no need to rewamp.

but it's right now available, companies use it for visualising, testing, etc.. possibly your new vw car got designed with this very raytracer, the app written in openrt.

:D

[cgk]

davepermen said:

the gpu vendors will happily continue where they are, and won't bother trying something really new.

thats the only reason it's not yet there.. as long as rastericers sell, they have no need to rewamp.

Actually, the main reason is probably that raster graphics are still _far_ superior to raytracing. And really, I don't see this changing anywhere in the foreseeable future. On the contrary, the new programmable rasterizers make many things possible, easy and cheap that would give raytracer programmers a hard time if they are looking for any remotely acceptable framerate (OpenRT requires massive precomputed datastructures for example, it's world is therefore mainly static).

And even then: Just look at the OpenRT screenshots and compare them to any ATI or NVidia demo. Raytracers always had some kind of fascination because of their "physical correctness", but after all, a good fake solution is much better than a bad "correct" one.

[skynet]

The _only_ tempting thing about Raytracing is the possibility to access the scene in your fragment shader. This allows to have the combined effects of different shaders on a surface (think about a bumpmapped surface refracted in another). This makes it very easy to plaster your world with arbitrary shaders without having to worry about how you create an effect that shows Shader A reflected on a Surface covered with Shader B.

Also, there are some misconceptions around.
1. "Raytracing has logarithmic complexity". Well, raytracing is _only_ logarithmic if you use hierarchical datastructures that accelerate the ray casting. The argument "Rasterizing has linear complexity" is wrong as well. Use clever algorithms and hierarchical datastructures and you can render scenes of arbitrary size with rasterization hardware as well. (Though, the mathematical prove that these algorithms are logarithmic is unlike harder). Also consider, "O(log n)" is only valid for a point-search in a largly balanced trees. Tracing a ray through a tree takes much longer.

2. "Raytracing is physically correct" Raytracing is not physically correct unless you use a physical correct Model of the light transport. The only thing where Raytracing beats Rasterization (in terms of "correctness") are _geometrically_ correct Reflections, because you have access to the scene in your shaders.

[davepermen]

heh, skynet: no

you lack a lot of knowledge and ideas on what a raytracer is capable, and what physically correct means..

and cgk:

hardware is only in one thing superior: performance. and thats because it's ultraoptimised hardware.

and rastericers in softwareterms have only one superiority to raytracers: they can start with less requirements, thus less steep requirements as minimum.

but.. well.. easy.. i have a 5hour set this evening, and i'm yet quite tired.. so i have other stuff in my head currently :D

[skynet]

Well, then please tell me what I´m missing.
I guess, "physical correct" has a stretchable meaning. Phong plus hardedged shadows certainly is not ;-)