see there is a common sense. Dedicated GPU is always better then integrated.
This simply does not have to be the case.
To talk between the CPU and dGPU and vRAM, traffic has to go over an external bus.
If all of that can be integrated on a single die,
so long as the resources are spent, the communications can be WAY faster.
The only issue so far is that intel and the rest haven't been willing to spend the resources on GPU on the processor. But the demand for actual CPU resources is just not there these days, its all about GPU.
If you think I'm talking crap, just look what happened to external math co proessors, external MMUs, external L2 and L3 CPU cache and external memory controllers. They're all on the CPU die now because it is faster.
Until recently, intel hasn't bothered to really try to build a proper GPU and haven't spent the die area and transistors on it (which is why their GPUs have been, to be blunt, crap).
Now they've decided to focus on it, AMD and Nvidia should be scared.