PDA

View Full Version : Radiosity optimization?



cj8n
05-13-2005, 09:55 AM
Hi,

I'm looking for suggestions on how to optimize radiostity rendering on my system. I have a dual 2.5ghz G5 with 4 GB of ram. I also have several slower PCs availalbe to render on. Should I set more threads or increase my memory segment limit? What ray settings should I use and do I need more than 1 indirect bounce? What anti-aliasing settings work well to reduce radiosity noise?

Thanks in advance,

Chris

cgolchert
05-13-2005, 10:09 AM
You didn't mention any of the really important stuff. :)


What is in your scene?
Is the scene going to be animated or just a still?
What major plugins are you using?
Are you ray tracing or just using radiosity?

cj8n
05-13-2005, 11:51 AM
the scene http://hdri.cgtechniques.com/~sibenik2/download/
still image to start, perhaps camera anim if I can get render times down
no major plugins
radiosity and ray tracing



You didn't mention any of the really important stuff. :)


What is in your scene?
Is the scene going to be animated or just a still?
What major plugins are you using?
Are you ray tracing or just using radiosity?

cgolchert
05-13-2005, 12:02 PM
If it is just camera movement you could bake the radiosity. You will need to create uvs for that or bake to vertex maps. Vertex maps will give you lower quality results but you don't have to create uvs for that.

Going through the scene turn off trace reflection... I couldn't find anything in the scene with a reflective setting.

Are you just playing with this scene or were you trying for something with it?

ps 2 threads should be good. I'd ahve to take another look at this scene at home to really tweak it. :)

cj8n
05-13-2005, 12:18 PM
just using the scene as a test to get a sense of radiosity and how it works

cgolchert
05-13-2005, 12:26 PM
Ok, that scene is really overkill for that. It was meant to almost overload the renderers to see what they kicked out. :)

To get a better understanding of it, you may want a simpler scene that renders faster so you can see the results faster.

Captain Obvious
05-13-2005, 12:52 PM
The radiosity modes in Lightwave:

Backdrop: as the name implies, this only calculates lighting from the backdrop, be it a pure white background or a complex HDRI. Faster than Monte Carlo, obviously, but also very limited.



Monte Carlo: produces great results in a very long time. Unlike the Interpolated mode, Monte Carlo will never introduce artifacts, only noise. It is great for details, too. With one bounce, it can be tolerable, but as you increase the number of bounces, the time to render inceases by an horrible amount. It also seems that it's very sensitive to the number of raytracing recursions in the scene (I turned down the number of recursions in one scene from the default 16 to 3, and the time to render was cut in half, all with two bounce Monte Carlo), but I haven't been able to confirm this. Rendering with more than three bounces with Monte Carlo is in my opinion not a viable option, ever.



Interpolated: Really fast, but takes a lot of work to get right. The key to the interpolated mode lies in the Tolerance setting, and this is the main thing you should tweak. When set to zero, it looks more or less like Monte Carlo radiosity, and takes a long time to render. With a tolerance level of one, it's can be almost as fast as rendering without radiosity. Kick ***. The problem then, however, is that all the detail become washed out. Shadows, corners, etc, won't have the same great natural look as they do with Monte Carlo. However, if you have a fairly complex scene with quite a few lights as it is, you can use Interpolated to increase how much the light "fills" the room, and make it look a lot better. A tolerance of about 0.1 or 0.2 can make some amount of detail appear, while still having acceptable render times.

The number of rays per evaluation controls how many rays are calculated, essentially. With a very low value, like 1x3 or 2x6, you will get blotches of light here and there. With a higher value, you will get the same blotches, but many more of them, thereby making the whole thing smoother. Turning up the number of rays will not make Interpolated better at details, though. You need to turn down the tolerance for that.

It is absolutely vital that you use several anti-aliasing passes with motion blur when using interpolated radiosity. The more passes, the better the radiosity will be.

Hmm. I'm going to try to write a guide for using Interpolated radiosity. I'll post a link later with some examples.

cj8n
05-13-2005, 12:57 PM
thank you, this is really helpful

JackDeL
05-15-2005, 09:45 AM
Very informative post Cap! I would be interested in that guide as well, Thanks! :)

Captain Obvious
05-15-2005, 02:18 PM
Very informative post Cap! I would be interested in that guide as well, Thanks! :)
It might take a while, though, since I have three exams and two assignments due next week...

JackDeL
05-15-2005, 06:53 PM
It might take a while, though, since I have three exams and two assignments due next week...

:eek: Yikes! Take your time Captain! Good luck on your exams! :)