I'm still floating between GAF and Xbox heavy influenced forum and trying to wrap my head around the "tech" talk.
Basically, I think, what I'm getting from info over thier is that thr Xbox is a balance of components that will, if I'm correct, will make it stronger?
While the PS4 is unbalanced with power in some parts but weak in others, there for making it unbalanced and in reality weaker.
I heard them refer to it as using a powerful hammer to swat flys instead or a fly swatter, or in terms of horspower in a car as a metaphor. ..saying as they go faster the PS4 actually loses power due to bottlenecks..
Is this nonsense? What exactly is the bottom line between these two?
Disregard the term 'balance'. It is meaningless. It's used is just there to seed the doubt that the PS4 is somehow unbalanced. It's not a technical term, just a carefully selected marketing one. And the Xbone is the more complicated design, due to its RAM.
The PS4 trumps the Xbone is all areas. It has a single pool of fast unified RAM. The Xbone as its main slow RAM, and a small chunk of fast RAM. Devs have to juggle what goes where, and as per the OP, are finding this to be a pain a big bottleneck.
The PS4 has 50% more CUs (the 'cores' of the GPU). With the recent 6% clock increase of the Xbone, this means that the peak performance of the PS4 is 40-50% more than the Xbone.
The PS4 has custom made chnages to its GPU to give it far more GPGPU capabilities. The PS4 has 8 ACEs (the things that manage GPGPU requests) and 64 queues (not to mention 18 CUs capable of running GPGPU). The Xbone has the standard AMD number of 2 ACEs and 2 queues, with only 14 CUs to run GPGPU on.
Many folks, from Cerny to devs, have said that making use of GPGPU will be the main thing that drives improvements this gen. The PS4 has significantly more room to grow.
The PS4 also has modifications to make sure it is hUMA compliant. The Xbone does not (nor could it, with having two pools of RAM). hUMA is the big direction that AMD have been heading towards.
Both consoles have a bunch of other DSPs for things like audio, decompression and security.
Multiple sources have said the PS4 is easier to develop for. Not only because it has ~50% more horsepower. Having one unified pool of RAM cuts out the need for devs to micro-manage what goes where, removing a big bottleneck (the same bottleneck the PS3 suffered with its two pools of RAM this gen).
Once Xbone devs get better that managing eSRAM and main RAM, there will be improvements and the days of sub-1080p games should be over. But the Xbone has far less customisation than the PS4, with less focus on GPGPU for long term gains.
Both consoles will see improvements over the coming years, but there is more untapped performance in the PS4 than in the Xbone.
Microsoft's Techs found that more CUs doesn't necessarily improve performance. They realised that multicore processors have problems with getting full performance with more CUs. So, instead of adding more CUs, they actually locked 2 CUs, and overclocked the GPU by 6.6%. Sony having more CUs doesn't naturally equate to better performance. As stated, multicore processors have complications with more CUs, regardless of the ROPs. In other words, the more CUs you have, the more problems you have. Sony has more CUs, therefore, it goes to say, they have more problems utilising the direct benefits of those additional CUs. Another thing to consider is, the Xbox CPU is faster than the PS4's, and it also doesn't have to do any audio processing, which the PS4 does. Also, the Xbox One's multitasking does not affect allocated DDR3 RAM, there are processes and other RAM on the board for that purpose. In other words, reading the specs alone won't give you a real idea of efficiency. The PS4 may have higher specs, but the Xbox One is more efficient.
There is a misunderstanding around what MS were saying around CUs. MS were trying to play of the fact that the PS4 have 50% more CUs by saying that power doesn't scale linearly with CUs. So the PS4 with 50% more CUs might only get 25% more performance in real gaming applications. But that's still 25%! Pretty massive, and more likely to be higher.
You don't have 'more problems' with more CUs. Adding more CUs is how you get different spec AMD cards in PCs. At a basic level, the more CUs, the more power. It just doesn't scale 1:1, but more is always better as far as performance goes. MS claimed that a slight clock bump was more benefital than enabling the two disabled CUs, but they were probably never in the position to enable them anyway, unless every single Xbone off the production like was tested to have 14 working CUs. Unlikely, as they had two spare to improve yields (an APU can be made with two faulty CUs and the chip still passes - it only needs 12 working).