It depends on what you do. For gaming for example, the difference will be absolutely staggering. For machine learning, even more so (since Vega has more than twice the amount of FP16 ALUs). If you are using your dGPU for photo and video processing, not so much, since these tasks don't scale too well to begin with. But GPU-accelerated rendering? 50% increase is not unlikely.
[doublepost=1541094533][/doublepost]
I really think you will be in for a surprise. Just the move from GDDR5 to HBM2 will be huge. The current Polaris 555/560 are bandwidth-limited in most tasks and show big boosts in performance when the memory is overlocked. HMB2 will completely remove that limitation. Vega is already much more efficient than Polaris (just check the performance of the Ryzen APU parts), and — since HBM2 barely uses any energy, more of the thermal headroom can be allocated to the GPU. Combine all that and improvements in the ballpark of 50% compared to Polaris, at the same 35W TDP, are fairly realistic.
But of course, this is all just a conjecture. We have to wait and see, and then we will know.