I won't quote everything to keep this post concise.
Remember DDR != GDDR, DDR4 follows DDR3, GDDR5 has some similarities to DDR3 but is a separate technology.
An awful lot of software is actually making use of more than CPU cores now, aside from tons of well-known software using GPU acceleration, there's also dedicated hardware on newer APUs for various tasks like media encoding (things like screen mirroring/streaming for example). Some repetitive, heavy tasks are far more efficient when performed on GPUs, sometimes even more so on custom logic; that's why pretty much anything you'd expect to watch a HD video on will have hardware decoders for popular codecs like H.264. That's especially important for mobile use/battery life.
I doubt Intel would be doing something similar if it was a 'solution looking for a problem'. And look at the next consoles; the Xbox processor has been detailed - if the APU architecture wasn't needed, they could've gone for two smaller, separate processors (like so many previous consoles - even the Wii U with its tiny CPU uses separate dies on an MCM). The XB One die is bigger than a 7970! And don't forget it's a very high volume part too - they wouldn't use such a large die for no reason.
Also, I'm not aware of anywhere, apart from a few parts of synthetic benchmark suites, where the 5200 is significantly ahead of Richland. And as CAT said, the 5200 is far more expensive than any Richland chip.