This is a post most half way experienced 3D modelling tool users will shake their head about.
It's targeted to people new to modelling who need a high resolution model to generate a smooth bumpmap from.
Let's start with a very simple panel I made in Lightwave:
It's a box 256m * 512m beveled out a bit and with smoothed edges.
Then I did cut it into segements to maintain 3 & 4 point polygons to be able to switch to subdivision modelling mode later (tab key toggles between poly editing mode and subdivision mode)
I then beveled out the newly created segments to make the surface a bit more interesting (would you like to read a small beginner tut to see me splitting up a flat plane into two triangles?...)
My subdivsion level (set in general preferences is very high for this project: 12)
The result of the amazing 264 poly panel looks something like this:
Now I hit the magic tab key in modeller an turn on subdivision mode:
Notice any difference? Sure you do
All and everything is nicely smoothed out, while the polycount still stays at 264...
Just in case you're wondering why I did cut the model up and smoothed the edges: Turning on subdivision mode with only a cube in vew will turn it almost into a sphere. So cutting the model up will reduce the smoothing effect of 90 degrees angled surfaces by creating smaller angles to interpolate.
Just maka a cube, hit tab and go to Subdivide: faceted - you'll see what I mean (subdivide metaform will increase the effect of smoothness when freezing the subdivision model later)
So while we're talking about subdivide metaform, let's see what it does to our panel:
See the difference? No? Well there is none really, besides the fact that the polycount has quadrupled.
Why am I doing this obviously stupid move?
I'm trying to create a mesh as unoptimized as possible as a base for a normal map. The less faceted the model is, the smoother my normal map will turn out.
If you want to hand the model over to Lightwave Layout for rendering purposes you could've stopped reading after "hit tab key".
Now we'll apply subdivide metaform one more time (again without any noteable visual impact) and then, after saving the model to hdd, we attempt to run a little stress test for our cute little pc: We will freeze the smooth subdivison model to a polygon model by hitting "ctrl+d".
Depending on your cpu speed and ram, you can now either make a cup of coffe, call microsoft tech center or wait a little till the smooth mesh got subdivied to a high resolution polycount model.
Assuming everything went well, you'll see something similar like this on your screen:
See the difference?
The blueish subdivision control cage is gone, the model is still smooth and the poly count has jumped to over 600k.
Now we hit the save button again and push the "triple" button.
This will convert any polygons with more than 3 vertex points into triangles, the kind of polygon shape every gaming engine loves!
This will also eliminate any nonplanar polygons (since a triangle can't be nonplanar:) )
This again will increase your polycount to another new record like displayed below:
1.2 million polies! 100k more and you have the same resolution Epic announced during a GeforceFX event to use for their new UnrealX engine when creating normal maps for their player models.
And we waste it for only a simple panel meant to get displayed on only two triangles ingame - now does that mean we're cooler than game developers?
In a common sense of biased uneducated webforum discussions: Yes.
In any other other sense of an educated mind: No.
Keep in mind that realtime rendering is all smoke and mirrors.
By exctracting a normal map from a 3D object we simply create an illusion, the illusion of geometry nicely lit in your game engine of choice.
The only greater illusion you can create with a 3D program and capeable gaming engines is by joining the PR team to grease the hype machine...
To show you that you actually did some ressource intensive operations:
So keeping two of these models in memory will eat up nice 670megs of RAM (multiply this with 15 and you will have the amount of memory needed to play with 30 high resolution player models on screen - remember this info and use it whenever a lamer keeps claiming that the new xyz engine will display such an amount of player models onscreen, hence he usually "0wnz j00", but most likely he doesn't even know that his 32bit pc can only address about 4gigs of RAM, which would enable him to load max. 12 models into memory...)
Excuse me the stupid notes above
Oh, before I forget it - this is our new lovely normal map: