Yes, the two skus worked from the beginning, and then the conservative approach dumped in favor of the future proof PS5, was my point.
As for the all the Gonzalo and Github leaks, they could be very well controlled leaks _ remember that people found Github is not really datamine.
Why nothing leaked yet about the big ps5 chip, is just a guess but, wasn't a team of AMD engineers sent to work directly with Sony on the chip? Then people started to speculate it was on Navi, and Navi will be exclusive lol _ would it be possible that they could conduct all their testings themselves, not needing to be done in AMD? I don't know, lots of possibilities and only time can tell.
Only thing that makes me doubt the Github stuff being a controlled leak was, there was Xbox-related info in there, too. Wouldn't Sony be in legal trouble if they did something like that, exposing MS's data? And likewise MS if they did that exposing Sony's? Unless they both corroborated together to leak the benchmarks, but why do that when it put PS5 in unfavorable light?
If AMD did the leak intentionally, and they didn't get approval from both Sony and MS to do it, then they'd just mess up goodwill with at least one of them. If AMD got them both to agree to it, why would Sony agree to that?
...who knows, maybe the duopoly is real after all
I think the trick of Sony (Secret sauce) will be to free the CPU from load all the data from SSD, using the GPU too, mainly to load heavy textures. We need to remember that uncompress data at 2GBps or 4GBps rate will consume a lot of CPU resources, and what is an advantage can become a problem too. So divide this load between CPU and GPU it seems an obvious choice to do not create a new nextgen bottleneck. So the architecture will be more important than the isolated component specs.
For that to work the GPU would need a CPU or MPU built into it, or the system itself would need an MPU co-processor interfacing from the main CPU on a main port, to the GPU (possibly on a PCIe lane, or SPI; depends on how fast it needs to send commands and how much data had to be sent per clock).
Keep in mind the purpose of the GPU is mainly to produce the graphics and then output them to the screen. Think of it as the color palette, and the CPU as the paintbrush, while the television is the canvas. With HUMA and unified memory pools, these GPUs can be used for certain non-graphics tasks too, like logic/A.I and the such, but at the end of the day they still have to receive the commands from a CPU. GPUs have schedulers and units that issue commands to parts of the GPU to generate the polygons, texture them, anti-aliase etc. but those are more like DSPs in theory, specifically dedicated to only those graphics/logic-orientated commands for speed and efficiency purposes.
Basically, there's no way to program a GPU itself as if it were a CPU; all code still passes through the CPU even if only initially before the CPU commands sub-processors like the GPU to pick up the task from there. But, I guess it could be theoretically possible to customize the GPU with units dedicated to texture cache access and read/write to/from the customized SSDs.
Don't know how they would do that, or if it would be possible with Navi's architectural design, but any kind of effort would need to be collaborative from start to finish between Sony and AMD.