simple TERRAIN?

Terrain CLOD: A properly implemented CLOD algorithm still beats the pants off a brute force system. This is due to the shear volume of geometry that a terrain contains. If you only render 10% of it, then your card is free to do other things. Not to mention the market penetration of T&L cards is still miserably low (give it another year…).

Texturing: It would be nigh-impossible to texture a terrain properly with auto-gen tex coords. Too many edge-blending issues. And a good texture LOD algorithm looks fabulous compared to a detail texture over a super-streched color map.

TIN: Too complex, too much memory, too many ‘sliver’ triangles…

VIPM: Why? Isn’t ‘View Dependent’ the whole point? shrug Depends on the project I guess.

VDPM: Looks promissing, I need to read more.

Geomip-mapping: Does not allow for strong terrain roughness metrics. Works fabulously for ‘smooth’ terrain though. Try it with ‘spikey’ terrain and you’ll understand.

Finally, robert_s: You do not need to draw the wireframe for any of the terrain. Instead, you want to draw only those parts of the terrain in the view frustum. Read my Gamasutra article to get an idea of the terminology used in Terrain Rendering. You’ll be performing two phases: Tessellation and Rendering. During Tessellation, you decide which areas of the terrain to render, and how many triangles to use for it, this decision is called the Split Metric - and is also where you perform frustum culling.

I keep a fairly up-to-date bibliography of Terrain Rendering on my website, I recommend purusing it before beginning your engine. ( Also check out the cannonical site (


FYI: Empire/Rowan Software is thinking about releasing their Battle of Britain flightsimulator as open source.

See this discussion:

See here for infos about BoB:

This could become a nice starting point, right now BoB is using DirectX as far as I know, but this could be changed. The flight model and engine are state of the art and can support several 100 planes at once in one game.

BryanNC said:
VIPM: Why? Isn’t ‘View Dependent’ the whole point? shrug Depends on the project I guess.

Well, the advantage is that you touch very little data each frame, and only indices, which means you don’t have to write to agp mem which is nice. And not being view dependent isn’t a very bad thing, especially if youre close to the ground most of the time. Then silhouette edges are fairly static (ridges and the like) so you can just add a cost to removing detail there, not at run time but off line. Of course there is some view dependence, you use lower LOD for tiles that are farther away, but you never alter silhouettes or something like that.

Thx all! you helped me a lot!