While the terrain chunking method works well in many cases, it dosent currently allow for LOD of any kind. Of course for smaller terrains this not an issue, or where the view distance is not very far. However, for large and/or highly detailed terrains, some sort of LOD is needed, particularly where a far view distance (several hundred thousand units for instance)is used.

May I suggest a future implementation of something like Geomipmapping , which is a simple and effective algorithm for patch-based LOD that is specifically designed for modern hardware?

Alternatively, there is the ClipMap method as well, though I personally find the geomipmapping method somewhat simpler and more intuitive.

As well as allowing for arbitrarily large and detailed heightmap-based terrain, this method can also be adapted for data streaming in the future.

I dont often make big suggestions for the renderer, but I think this is a good and well-considered request.

I couldnt find it but anyone knows what LOD technique does Oblivion use ?..I can see quite some terrain popping up(sometimes even altered geometry altogether) in it,but then I dont have a beast of a machine either.

Well it would almost always run better than the current chunked terrain method on any system. Plus LOD values can be changed depending on desired framerate.

For instance, if you render a 2 million polygon terrain with normal chunking at full view distance this will be very slow on the best hardware. With geomipmapping, it can be made to run fast on most any decent card with minimal loss of detail.

its a winning situation, no question. Terrain LOD is needed in a modern engine i believe. brute force only works with small levels with low farclip, even with chunking. Chunking as it stands is only good for increasing the terrain complexity, not improving framerate much beyond the gain by frustum culling the unseen chunks.

That's exactly the answer I was hoping for. Such a feature would indeed be nice then,

Cheers

Terrain chunking works very well with LOD. We just haven't implemented it because our tests have shown no significant frame rate improvement. However this might change with larger terrains.

Patching or "geomipmapping" was one of the first algorithms for terrain optimization, and is for various reasons not suited for modern 3D hardware. We tested it once with an early A5 version. It was slower than no optimization at all. I don't think that anyone still uses it today.

Chunking is by far the best terrain algorithm for T&L 3D cards. The only thing that we could improve is implementing caching, and possibly LOD, for extremely huge terrains.

JCL,

I have done extensive testing with chunked terrains of 1,000,000 vertices, and have found that performance is simply prohibitively slow (even on GeForce 8800) unless a very conservative farclip plane is used, such as 40,000 units given a scale of about a 150 unit-tall player model and a terrain size of about 1.1 million units wide.

i dont consider this vertex count for a terrain of this size to be particulary high. In fact I think its somewhat low.

I find it hard to accept that some form of LOD is not only beneficial but needed for a similar situation to run well on modern hardware.

My point is not a particular system, but that LOD is not only a good idea but neccessary. I cant believe that peformance with LOD is not going to be substantially bettter at scales and vertex counts like I have mentioned, and beyond--you are going to want very high vertex density near to camera.

We will implement chunked terrain LOD in a future version.

Great! thats all i wanted to hear

This is great to hear.
What type of LOD will it be?