View Full Version : Optimization for high poly scenes?

09-21-2006, 04:32 AM
Hi I always seem to go overboard with the poly count in my objects/scenes. I dont feel that its overboard but I am constantly running out of memory, when I reach about 1.6million polys in a scene with a few image maps (usually large ones) thats it im screwed!
Im just wondering if people could point out some hints and tips for maximizing memory usage without culling important details from scenes.
Is there something i can do in the LW settings to give me a bit more resources or maximise my resources?
I am not too concearend with render times as I am just working on stills atm.
The specs for my machine are as below.

09-21-2006, 05:00 AM
I hit this a while back. You can make HUGE savings in memory if you change your image maps to 256 colour PNG indexed. Of course, this is completely lossless with shadow bump and specular maps. With a good image program you are unlikely to norice the reduction in colour depth.


09-21-2006, 12:26 PM
The best method for optimising scenes for my work, is to plan out the animation or scene before hand, before commiting to any 3D work. You only need to model high detail for the stuff that ends up in front of the camera. Medium range shots you can drop the modelled details and use normal and bump maps. For the stuff thats in the backgound, often just a lores object/basic map will suffice.

Best thing to do, is decide what resolution your rendering at before hand, then run a few tests on different detail levels. Do I need 32 sides to define curves or 16 for this resolution and framing?

I have found too, that over modeling can cause problems with animation. If you have a high poly object thats way off in the distance and only being defined by a few pixels, things like microbevels can cause flicker if AA isn't cranked to the max and sometimes even at highest levels it can't be fixed.

So I always remember the 6 P's before I start a project - Perfect Planning Prevents P*ss Poor Performance :)

09-21-2006, 11:17 PM
starbase1 - thanks for the tip, ill give it a try.

stevecullum - cheers Im aware of what you say, I guess im more wondering how people can push lightwave to render 10mil polys plus and still use plenty of image maps etc. The max I can ever seem to squeze out of lightwave is about 2mil. I think my machine is pretty decent but I still seem to fall behind in poly counts etc. I like to make scenes with a lot of minute details, if those details are small ill use small image maps less polys etc, but even so if you want 500 or so of these small details scattered throughout your scene it starts chewing up resources its a real shame LW still has no hardware instancing.

09-22-2006, 03:52 AM
The poly limit in lightwave is alot lower than on other apps sadly, but I have rendered 7 million using procedurals - that was in LW 64 on 4GB Ram.

More ram and 64 bits is the only real answer afaik, but then I rarely need to render those kind levels. Most of my work is rendered in seperate passes and compostited later on. Once we have true instancing (and its coming soon according to LW website) hopefully the poly counts will be up there with the rest of the pack!

09-22-2006, 06:43 AM
SteveCullum -

Once we have true instancing (and its coming soon according to LW website)
Wow, where abouts?

09-22-2006, 07:25 AM

Will LightWave v9 have instancing?
Instancing will be added in a later update in the v9.x cycle. We'll be making the core changes that will allow us to provide instancing immediately as our first effort after the release of v9.0. These same changes to the core operation of LightWave will fix a number of long-standing issues in the code and provide the basis for a complete change in the workflow of the application, including a variety of other long-requested capabilities such as history, parametric evaluation, scene graph and more.


09-22-2006, 12:21 PM
check my optimization list here (http://www.cgunderground.com/?q=node/165), that help you to speed up your scene.

09-22-2006, 12:27 PM
Cheers Steve :)

09-22-2006, 03:13 PM
When i did the powerball project I had scenes in excess of 6million polygons with high res textures throughout.

My piece of advice is this, use subpatches on everything, but be VERY conservative with the base cages, basically i work in such a way that with subpatches at level 1 my models are only passably smooth, while on level 2 they are ok but not always perfect, they look perfect at level 3 (upclose).

this way, you can always crank the subdivisions waaaay down for items that are really far away, or crank them up if they get close on a shot by shot basis.

with the new adaptive subdivision with LW 9, this should work even better, but i dont really trust the current implementation, im moving on to xsi anyway

09-22-2006, 04:40 PM
I'm sure you're rendering out high-res images as well, that also eats memory, you should be rendering with F10 for that. If you can render in slices with limited region that would help.
You could use some more in your current system, LW can use 2gb, but your system takes a good 1/2 gb, right?

Another option which would save you lots of texture memory is InfiniMap Pro. I just did a test with the demo version, and rendered a polygon with a 21,000x21,000 24 bit image on it, and LW only used 270mb to render it. The plugin only loads the piece of the image that it needs per pixel.

09-23-2006, 12:19 PM
Thanks people this is quite helpful.

Stooch - Yeah I dont trust the APS much either, its great in theory but in practice I think it coiuld be implimented a lot better. I find the gradient control a bit scetchy & I think it could be explained a bit better in the documentation as well. Its near impossible to get dense detail just where you need it without getting excessive detail throughout the rest of your object. I have also noticed that when you animate an object with APS control on the camera and render it out, that the Detail stays in the same spot on the object that the camera was first focused on... so as your object rotates for example the detailed area rotates away revealing horrible low detail geomitry. Dodgy...

toby - Whats the diference with F10 render? InfiniMap Pro sounds good ill have a look.

09-23-2006, 12:48 PM
Whats the diference with F10 render?
F9 takes up memory to display the image, and for a big image it takes lots of it - remember that memory requirements with high-res images go up geometrically, double the size is 4 times the memory. F10 can make a big difference, and it's a standard technique for print res rendering.

09-23-2006, 12:51 PM
Thanks for that toby, all these years and I never realised... that should save me a few dramas in itself.