|
Pemalite said:
|
Actually, Zen has very small cores (4.9mm^2 reportedly). That would make a fictional 8core Neo (or even a 6core unit would suffice) feasible inside a Neo SoC. However, all that cache ultimately makes the whole package too big/expensive for a console. Also AMD in its roadmaps for the Zen architecture does not show third party access to the Zen architecture until 2017. The design of the Neo SoC was likely finished in (late?) 2015 so that simply doesn't compute for Zen.
The idea that Sony would have any viable way to change the SoC design "after the Scorpio announcement" is laughable at best. Yes there are tons of rumours (more precisely: fanboy wet dreams) floating around that "Sony has two cpu designs to choose - and after the Scorio announcement blablabla..". Of course Sony had several designs for the cpu. When they started to play with ideas about a Neo. That probably was in 2013/14 when they started seriously thinking about a beefed up PS4. If you are lucky, you can change a key component inside a SoC and you are back where you were after 6-8 months at best.
The next idiotic rumours came up when "4k gaming" was the new buzzword. The Neo is not equipped for 4k gaming. For 4k gaming you do not only need TFlops (Compute Units), you need enough TMUs and ROPs and memory bandwidth. NIVidia's new 10x0 gpu line has 48-64 ROPs (with double raster power), high clock rates and high memory bandwitdh to get into the 4k terrain. The Neo SoC most likely has none of all that. Spicing up the PS4 memory bandwidth to a possible 220GB/s gives you more room for compute usage (good for PSVR use and cpu/gpu sharing of the bus), but by far not enough for "true" 4k gaming.
Why don't people just wait a week or so to see what Sony has to say about the Neo (provided that announcement really opens up the hardware)? Personally I can see a Neo SoC with upclocked Jaguar cores (we know the 14/16nm design was done because of XBox Slim) and a gpu with double the CUs (excellent for gpgpu stuff) and unknown TMU count and same 32 ROPs due to 256 bit memory bus. That would be completely in line with the original 4.2TF rumour.
