I don't know if you guys see a big downside in this new API...
To be close to the hardware you need to eliminate all the abstraction level used to deal with different types of hardware... so this API will be limited to one unique kind of hardware (GCN).
So forget nVidia or any other GPU hardware.... low-level API can't deal with different hardware... that's kill the purpose of a low-level API. If you start to add abstraction level to support more hardware you will go back to DirectX/OpenGL.
The API is open but it can deal with only one hardware to maintain this low-level access.
For each similar hardware a new API will need to be created i you want this advantage... so for each Family in nVidia GPU will need to have it own API... same for AMD... each family a new API.
While that DirectX and OpenGL support every family of GPU from any company since... well... forever.
PS. This API will be dead when AMD moved to a new architecture in the future.
