OMG, just watched the video. The technology is very very interesting indeed... dat example of running geometry on one card and doing global illumination / shadArZ / AI on another card or between a few other cards is really quite an interesting departure from the present status quo.
Questions like "what FPS do you get" will become pretty much irrelevant in that situation, since you'll probably be getting like over 9000 literally on a low-end card, easily on pure wireframe geometry and textures... but maybe the lighting around the game changes at 10 fps?
It'll also maybe enable mixing different GPUs in some situations, if you're brave enough...
Then again, this is all ONLY possible if you have a LOT of development time to dedicate to this kind of stuff IMO. Basically, you're developing your own GPU driver, in some respects
. Forget about indie studios using some complicated technique, unless they have some super special talent in their team, which is very unlikely. Then again, maybe AMD will have magic tools that make development easy? Not likely.
Speaking of super special talent, when that driver guy started talking about queues and the rendering pipeline, my brain started slowly twisting around itself. I hope that man gets over 100K for what he does - he certainly deserves it.
... and then the biggest question of all, will this really take off? Will people spend extra time developing for mantle when nvidia doesn't support it, and the other big question... what will nvidia do next year with Maxwell and its "unified memory."
