The reason behind the PS4 having 8GB of unified memory, is because it runs x86 cores. So software can't use beyond the 3.75GB limitation.
Jaguar cores are AMD64-based (i.e. x86-64). It's been a heck of a long time since AMD64 came out, but if I remember correctly, a CPU implementing it can reference something around 2^44 bytes of memory (not 2^64). I believe that's the right number as it results in 16TB of memory, and I recall reading about the number 16 in relation to the new memory capacity.
Also 4k playback requires at least 4GB of its own.
Uhh... why? If that's true, then 1080p playback would require 1GB of RAM since it has 1/4 the number of pixels. 3840 * 2160 * 24 = 199,065,600 ~= 190MB frame buffer.
Speculation suggests this will improve gameplay performance, I for one agree (not drastically, but at least a few frames).
It would be nice to see some statistics on how many sort of GPGPU events are going on during today's modern games. Is it simply limited to things such as PhysX, TresFX, etc?
How PS4 Could Overpower a More Powerful PC
Pretty much summarizes the dozens of pages here
http://e-mpire.com/content.php/2056
I didn't even have to read past the first paragraph to spot sensationalism.
In addition, the components in PS4 are manufactured with high bandwidth in mind. Much higher than PC.
Not true. The memory bus in many of the higher-end graphics cards (especially from AMD) are definitely faster than the PS4's
combined memory bus.
The PC GPUs have their own dedicated pool of GDDR5, while the CPU has it's pool of system RAM that is usually a good bit slower.
This is true; however, the problem is with the takeaway, which is referenced further down. They seem to be suggesting that there's a
problem with the CPU having less bandwidth. The fact of the matter is that there
isn't a problem unless you're encoding or using an APU. We're talking about gaming and discrete cards, so it doesn't matter.
I don't think anyone is arguing about the unified memory architecture combined with an APU being more efficient for sharing data, but I still don't see any proof that this will magically make it into a paragon of geometry processing excellence. When it comes down to it, simply because you can execute GPGPU commands faster may help you get to the point where you need to build the frame faster, but high-end GPUs still have more logical cores to actually build the geometry. No matter how many links you provide, you cannot necessarily overcome the hardware in these high-end, (relatively) expensive GPUs. The article does make a
very obvious point (to the point where I'd issue them a Captain Obvious award) that the PS4 will be faster than a PC of the same price.