JEMC said:
First of all, after writing that I was told that this wasn't true and that performance improved when moving from DX11 to DX12. That is true in some cases, but not in others: http://www.guru3d.com/articles-pages/hitman-2016-pc-graphics-performance-benchmark-review.html 1080p DX11 1080p DX12
Now, regarding your question. The problem we have is that there aren't many DX12 games available, and even less DX12 only games. But I would say that it's more a case of devs/publishers willing to invest in making proper DX12 modes versus others that only the bare minimum to be able to use the DX12 tag as a selling point. Part of the benefits of DX12 is to avoid CPU bottlenecks (that Mantle's primary task, and DX12 & Vulkan come in one way or the other from it), but also brings new techs and tricks, one of them being Asynchronous Computing or A-Sync Compute where AMD's GCN hardware excels. But that is only part of DX12. Honestly, I don't know how long it will take until DX12 is the base API for the upcoming games. If it was in MSoft's hands, it would have happened one year ago, but that's not how it works. In my opinion, I'd say that DX11 will still be the main platform the next three or four years, because the market is still dominated by DX11 capable cards. |
Thanks a lot.
Oh yeah, asynchronous compute was the buzzword thrown around so much. Problem is I don't know what it is. 
The thingy avoiding CPU bottlenecks in DX12, when that was announced I remember I thought it was totally useless and even counterproductive a couple years ago, because in the modern era nearly all games have been so clearly bottlenecked by the GPU, except for strategy games, simulation and sim-hybrids like Skyrim. But since then I've realized that CPU technology has lagged behind so much in recent years that today some even pure action games have become largely CPU dependent. Games still don't require much CPU tasks, but the graphics cards have improved so fast and the graphics engines with them that even the newest CPUs have a tough time to keep up. We just have to transfer some tasks over to the GPU.
Another thing I would like to understand better, is why CPU performance only increases by, what is it, like 25% every two years when GPU improvement is still fairly close to Moore's Law (double transistors every two years, which typically translates to perhaps 1.5-1.7x in raw performance).
I assume it has something to do with specialization versus parallellism for sure, but I wonder if there's other factors too.
Thoughts anybody?
