Thanks for the overview.
To ask more questions then, you say there's no point in using FP16 ever, does that mean what was posted on B3D incorrect and it was using dual-rate FP16? Wouldn't the lower precision make the calculation quicker anyway?
It's not like I know much about it, just all this area interests me. I'm also presuming that these sort of tweaks in a game engine is probably the lowest you can go in-terms of the stack in an engine? Would this mean any performance gains you can find using dual rate FP16 would have to go through extensive testing due to how many areas it would actually touch?
Fair enough. When we deal with clients, it's always usually very showy and always tries to create a certain image so I respect your stance on it.
Let's hope we get some other media soon