I ran across a few articles describing how Intel is trying to improve their integrated graphics, one of which is a method that people were quick to point out AMD has a similar method (the zero copy method). Something I find amusing about this is that an embedded systems application framework, implemented back at least in 2003, uses a zero copy method on events to schedule tasks. Namely, when a task has to do something from an event, the scheduler just passes a pointer to that task. This has the obvious value of saving on memory.
I figure, you know, desktops don't have to worry about such things. But apparently it's a big enough of a performance improvement that they're doing it. Then again, this sort of thing was also implemented back in 2006 or so in the form of geometry instancing.
I wonder what else we could pull from the embedded world...