View Full Version : The troubles with nodes...

09-15-2008, 04:11 AM
I need a stained-glass window for a church I'm working on, and it seemed the perfect task for nodes (which I know very little about and I scarcely use at all, because I'm still having problems understanding how the blasted thing work) so I took a deep breath and went for it.

After a dozen or so fruitless attempts and after having tried every single mysteriously-named node, I ended up with a rather simple set-up consisting of two cells, one penrose rhomb and one planks2d. Test render - sweet! It does indeed look like a stained glass window... gee, rendering is a bit slow though, isn't it.

Turns out it's unbearably slow. 5 minutes at 640*480 with no tracing at all, 20 minutes (20!) with tracing on (depth at 3), AA level 4 and AS at 0.3; HOURS when I turn on radiosity. I thought it was the model, so I spent an afternoon otmizing it - no go. I thought it was my radiosity setup so I reduced it to the very minimum - no go.

I turned off the nodal surface on a whim: 20 seconds without radiosity, 3 minutes with. What?!

Now, being node-impaired, I'm willing to assume I'm doing something wrong - and I like to know what - because otherwise the entire nodal system would be unusable in production.


09-15-2008, 04:32 AM
Post a reasonble facsimile of your object/scene for the crowd here to test.
Lots of ideas on what might be the issue; and I don't believe it's nodes as much as it is what you are asking of the rendering engine.
IF you are reflecting and refracting through some thickly colored glass, then yes, it's going to be slow. Optimizing such a scene seems to be a hands-on-endeavor. At the least, one in which one needs to see all the parameters and how they are working together.

09-15-2008, 12:38 PM
penrose rhomb from dpont? It's indeed very slow procedural texture. But you can bake it to regular texture and use with normal mapping.

09-15-2008, 01:38 PM
Penrose rhomb ?
It is rather from Derek Ledbetter and original Roger Penrose's algorithm...

Don't know if you really need to use normal mapping,
but yes if your procedurals are evaluated a lot
through nodal multi-connection or transparency/refractions
and radiosity in the scene, the context could explain
the slowness.
UV map the glass and bake the colors, you could
also bake separatly the stained border/gap with
different settings to get a bump map.


09-16-2008, 02:02 AM
Yes, thank you - instead of baking the object I built a suitable imagemap using Genetica and it works like a charm - although the nodal surface was a little more...glasslike for some reason, but it must be seen from a distance anyway.

Unfortunately I can't post anything about the actual object because it's for an architectural design competition, and I really wouldn't want to be sued by the client - that could ruin our relationship somewhat :) - but I appreciate the fact that there seems to be a logical explanation. I know there's another thread for this, but I really would like a more detailed nodes manual now - something that explains this kind of behaviour to the Maths-impaired Club, which I should probably be President of.

Thanks again!