This is a really impressive demo. Most virtual globes (e.g. Google Earth) separate the terrain, surface image and building data. Normally, these are sent to the client separately and merged in the graphics card: the surface image is texture mapped onto the terrain, and then the building data is drawn separately on top. Special routines are used to draw trees (e.g. billboards).<p>What Nokia have done here is to merge everything - terrain, surface image, buildings and trees - into the same model. They're still using the classic chunked level of detail approach, just with more complex models, which the graphics card handles with ease.<p>This requires more work on the server side to prepare the data, but once it is done it is <i>really</i> fast for the client. The main disadvantage is that the data ends up being very static - you can't move objects around, for example.<p>P.S. I'm currently working on open source WebGL globes like OpenWebGlobe (www.openwebglobe.org) and WebGLEarth (www.webglearth.org). If you're interested in this sort of thing, I recommend reading www.virtualglobebook.com .