I'm really curious, why Sony just don't go with the current Cell with all the SPUs and faster clock paired with another NV GPU, like the upcoming GK110.
Cell is pretty awful to program. It has monster performance on software that benefits its architecture, thanks to the super low latency and high bandwidth local stores in the cell SPUs, but on software that isn't easily squished into that 256k of RAM it is not nearly as quick... There's a lot of main memory latency on cell as well thanks to the ring bus (1000+ cycles I've read on this board).
Also, you'd need to re-engineer cell to accept some other form of RAM than XDR, as that is kinda dead right now. There's no roadmap for faster XDR than what currently exist AFAIK, and having even more flops all sharing the same 22GB/s as in current PS3 wouldn't be much of an increase.
Better possibility of BC with PS3 and they can sort of continue with all their software investments in Cell and NV GPU.
Well, uh, yeah that is somewhat of a benefit in cell's plus column, but overall I think sony would prefer to just forget cell altogether. That one was Ken Kutaragi's baby, and he's long gone from sony by now.
Its not A10, its A10 based. Sony is putting custom GPU instead 7660 as I doubt they want to add discrete GPU to the APU which will only add up to motherboard complexity when they can just put better GPU in APU.
Not sure how much "complexity" really that would be, considering consoles have consisted of a separate CPU and GPU pretty much ever since the original playstation era. Heck, separate CPU and graphics chips have been the day since forever really.
A PCIe link between APU and GPU, and some power regulation for an external GPU isn't much in the way of complexity really. Any decent integrated GPU would lead to a very large APU chip, which wouldn't neccesarily be a cost benefit overall, considering manufacturing yields and such.