Due to work pressure I've been working all day and all night so all I've really had chance to do is a little research. I read a nice article on how valve have implemented multithreading in the steam engine via a hybrid approach of both coarse and fine multithreading, separating each major subsystem so it can run on a separate thread, then further threading each subsystem so each one can take advantage of more cores if available. I wonder whether openmp has had a role to play in any of it. Either way, multithreading certainly paves the way for some interesting tricks.

Speaking of which, that brings me neatly to my next topic of interest - id's tech5 engine. There has been an interesting discussion on the message boards here regarding virtualised textures.

I'm certain texture atlases come in to the equation, because there's no other way you can draw a whole scene in -3- draw calls. These cannot be conventional texture atlases though, I think they're generated on the fly. The problem with this is 2-fold, one is a practical consideration, the other is speed. Practically, a renderable object in a traditional renderer needs to know which texture it is using, and texuv's need to matched to the texture. If you're going to do this dynamically your process may swap your texture and uvs to another map. In terms of performance you may only construct your texture off the gpu and then do an upload and swap over just a few frames in a thread.

The 'mega textures' in tech5 I've heard people saying that it's just geometry clipmaps. Again I think it may be a variation on a standard technique where the base technique is a geometry clipmapping, but maybe with another threaded twist like dynamic streaming the data from the clipmap.

Whatever the result turns out to be I can imagine how much work it would take to put together an engine with the supporting toolchain of that magnitude, and the previews are nothing short of brilliant! Really inspiring work from the I'd guys, and I can't wait to try out the games!!