A lot of these CSS posts reference the performance gains realized from using CSS transforms as opposed to canvas because the transforms are hardware-accelerated.
How long will it be before canvas gets hardware-accelerated? Is there a technical hurdle that needs to be overcome in getting (for example) V8 to talk directly to GPUs?
Canvas is still 2D though, so unless they develop a 3D drawingContext for canvas, you won't be able to do what's in this article with the same kind of hardware acceleration. With canvas you'd still need to run all your own poly drawing routines, while I imagine Webkit is able to take 3D CSS properties and map <div>'s straight to polys that render with hardware T&L pipelines. (Which makes me wonder, how the heck are they dealing with Z-order?)
It is similar in concept to Structure Synth ( http://structuresynth.sourceforge.net/ ) but in 2D and using CSS. I was thinking of making a browser-based 3D Structure Synth but not many people seem to like making art with code.
You could skip all the CSS transformation stuff and just serve the browser the images in the orientation you want them. In this projection and buildings-of-cubes, there's only 2 perspectives for a wall and 1 for ground/roofs.
Panning would work fine since the objects in this projection don't change anything but position with panning. Spinning would only work in 90 degree increments, but that's fine for many, maybe most cases you'd be doing this. Most isometric-ish projection systems I recall seeing don't spin / change perspective. I'd imagine that's because it's disturbing to look at since things don't get smaller as they get farther away. For whatever reason the static version doesn't induce the same reaction.
How long will it be before canvas gets hardware-accelerated? Is there a technical hurdle that needs to be overcome in getting (for example) V8 to talk directly to GPUs?