Super cool.
I for one can't wait to use C++ like CUDA can. Type traits and static / compile time polymorphism - some fools think C is the way to go but it becomes hard to make generic code and you tend to get higher performance and more flexible code if you (ab)use c++ templates right. Less macro tom-foolery too and all those defines which need to get set on the host side (for every language that's insantiating the opencl) w/ every given type.
My advice is put 2.1 out pronto and start sucking more researchers, automotive and military customers back from NVidia - potentially before you've got every bit of Vulkan perfected.
Looking forward to OpenGL's successor not being so gosh darn complicated too. The math is fine but soooooo many gotchas, it's tremendously off-putting/unproductive to use directly for me.