I recently read this paper with results that suggest that a quadtree based terrain algorithm is better in general than a block based algorithm.

The quadtree algorithm uses a reusable vertex and index buffer for each block. (That is, the number of triangles in each terrain block is the same.)

Further, the tests in the paper show that using vertex textures with the reusable vertex buffer is slower than using two vertex streams -- a reusable vertex buffer in one stream, and height data in the second.

I am wondering if anyone has experience with the difference between using two vertex streams for terrain rendering versus using one vertex stream with vertex texture lookups for the height data (on the xbox360 in particular).

I have implemented a GPU terrain rendering algorithm using a quadtree that uses vertex texture lookups much like the one in Game Programming Gems 6 section 5.5. After reading that article, I tried implementing two vertex streams, but my implementation with two streams is slower so far (just wrote that over the past hour, so there is much optimization left to be done I'm sure). However, this is on my laptop only, as will not have access to my xbox for a few days.

So does anyone know which way is likely to be faster on the xbox360? Maybe vertex texture lookups are particularly fast on the xbox, I don't know...