I really hope AMD would be able to implement HBM for Raven Ridge, but I fear the technology is still too high-end to be used in a APU (market segment / price point) at the time RR is out. HBM or a extremely fast on-die frame buffer cache is mandatory in order to kick-(re)start the APU concept.
If AMD fails to solve the bandwidth issue, they will loose their only remaining edge over Intel. AMD has run out of time and unless the solution is introduced in Raven Ridge APU / SoCs the whole concept of AMD APUs will die along with this failure. After that the target markets for their APUs will be solely in developing countries.
In the other hand, if AMD is infact able to introduce HBM or a on-die frame buffer in Raven Ridge, the performance improvements for APUs will be simply tremendous. Even with the bandwidth available from a single HBM stack (128GB/s @ 500MHz) AMD could increase the GPU resources up to 2.5x (over current generation APUs).
However based on the history I´m not holding my breath for this to happen, at least not in Raven Ridge :'( Wouldn´t be too surprised if Raven Ridge would be restricted to DDR-2400 (DDR4) too D: