I thought I'd come back to add that there are AMD GPUs older than the Rx 200 series that will support DX12. I was going by the feature listing on Wikipedia, but today found out that any of their cards using the GCN architecture will be compatible. This includes some from the HD 7000 series and HD 8000 series, as well.
Also, the major new features of DX12 are aimed at increasing performance, for sure. But you have to think about what that means for game development... Take consoles for example, where they have a set configuration of hardware with a finite number of resources that they have to make the best use of. You have a "performance budget," which say you'll calibrate so that you end up running the game at 1080p with 60fps. You design the game and everything in it to look good, play well, and still be able to be run at that target resolution and speed. You might have to cut down on things like draw distance or number of NPCs visible at a time in order to meet that target, to come in "under budget." Or if you decide you can't cut back anything, you adjust your budget at the last minute, lol, which is why you see so many games running at sub-HD resolutions on consoles... They went over-budget, and had to adjust that to compensate and keep performance up.
Anyway, imagine if you could do everything you were doing, but suddenly you save a huge chunk of performance just by being able to do that stuff more efficiently. That is what they've been focusing on with the last few iterations of DirectX. The power of modern GPUs is insane, for what they're made to do they are tens or even hundreds of times more powerful than your CPU, a general purpose processor that has to handle so many other broad things. And they're only getting more powerful every year, exponentially so in some cases. The problem is that there are bottlenecks, and your GPU is not running at full capacity, not doing everything it could be doing all the time. Look at these charts (they're animated), borrowed from an article on AMD's website where they talk about such features in DX12:
Asynchronous shader operations -
http://i.imgur.com/XgiOwNI.gif Multi-threaded command buffer -
http://i.imgur.com/jYWEEtY.gif By letting the GPU do multiple things at once, as well as FINALLY taking advantage of multi-core CPUs to feed data and commands to the GPU, DX12 is supposed to drastically increase performance. That means much larger "performance budgets" for game developers to work with. More and better graphics effects, more stuff on screen, more power leftover for behind the scenes stuff like AI and physics, higher detail everything... All while maintaining good performance. This is how games get better, we need advances like this.