I guess it's a possibility. We'll have to wait and see if their possible gains are greater than the costs of finding a way to theoretically double their FP performance without compromising other requirements such as die size or power consumption. 

I don't think we will see stacked ram in a future PS4. Not worth the effort. What you win in surface you loose in height, essentially.  By the time this becomes an option, we are looking at a potential PS5. Potential savings are in the SoC (which might be a multi-die assembly for the first year(s). This 28nm stuff is brand new (at least for GF, a potential manufacturer), untested technology so the yields are probably very low on the cpu and gpu dies. If you have a single large die and the cpu part is bad, you have to throw away the gpu part as well and vice-versa. If you have two separate dies, you only have to throw away the defective die, not both dies).

If you look at any PC dimm that has a heat spreader, you'll notice that the heat spreader is not in contact with the ram chips, but with the small pcb board the ram chips are soldered onto. Ram chips are basically cooled through their pins, the heat flows through the pins onto the pcb and barely through convection. So if you have stacked ram, you just insure the pins are "large enough" (whatever that means in the end) soldered onto a "large" heat spreader. Better airflow around the heat spreader will do the job (if I look at my board with four ddr3-1600 dimms, the air flow is pretty shitty but the heat spreaders apparently still do the job).

If you look at the schematics, you clearly see the stacked ram is NOT on the volta gpu. I think the first company to actually "stack*  is a future (lower power) Intel cpu. Technologically, they are maybe 16-18 months ahead of TSMC and probably 24-26 months ahead of GF.

When you say not worth the effort you mean basically just the space the ram takes away ?

But GDDR5 produces more heat and eats up alot more power than DDR3 or DDR4. If this could be reduced by lets say 20 Watts through stacked ram the powersupply could be made smaller and cheaper aswell as the cooling and the board. the number of chips might be less too DDR3/4 densities are easier to increase than GDDR5.  

Ofcourse if its not worth the effort Sony won't do it. But we don't know if it might be worth the effort someday in the future.

Sure guaranteed savings are in the SoC, but stacking could come out cheaper than using GDDR5 one day. If the process is proven and widespread. PS2 got redesigned after PS3 was on the market so who knows. 

MY biggest problem is that there is absolutley nothing that needs this sort of power, and to create a game that uses it would be cost prohibitive. This is why we are still stuck on crysis level graphics despite having the power to make crysis cry.

It will be pretty simple to use up all the bandwith if game developers can target Volta specs as baseline, through highres textures and high quality assets. The assets are already high quality but downscaled for use in games. But even without changing anything all that power will be used up easily.

4k gaming at 60fps should become reality with Volta. I am pretty sure 3D 4k next gen console titles with texture mods will make a Volta SLI system necessary. Game budgets wont even be influenced by that. 

It's totally exciting - we are in a new era of tech advancing super fast again.  IMO, things have been rather stalled the last 6-7 years, as tech was moving to more efficient and less raw power and speed.  It's kinda sad in a way new consoles are coming out right at the beginning of this, as in just a few years, things will be 10-100 times more powerful. 

Still, I will do my best to enjoy it now, and also when the new stuff comes along.

So 8-9 times by going from 28nm to 10nm, but what about from 28nm to 20nm which is around the corner? What sort of increase does that give in raw performance (ignoring any architectural advances)?

The first card to come out utilizing this will be launched in 2016 at the earliest, and it will probably be reserved for their $500 and above cards only at first. By the time this becomes more mainstream we will be talking about PS5 anyway.

Also just to compare:

X360 ''Xenos'' gpu has a bandwith of 22.8 GB/s. PS4 will, if it's indeed based on HD78xx technology about 153GB/s of bandwith. So that's about 8 times as much. Going from 153GB/s to 1000GB/s would be an increase off about 6.5 times. So no technology advancement is not speeding up.

28nm to 20 is double. It depends how good the process turns out, Its really hard to tell. 

Lets compare 40nm vs 28nm die size wattage included.

Top End GTX 580 for 40nm 244 Watts Die size 520mm2 Tflop 1.6

Top End GTX 680 for 28nm 195 Watts Die size 294mm2 Tflop 3

So they doubled Tflop within one shrink. But if you look at Wattage and die size then GTX 680 is not the equivalent of a GTX 580 actually that would be this:

Real Top End 28nm Geforce Titan 250 Watt Die size 560mm2 Tflop 4.5

They almost tripled the Flops within one die shrink. 28nm to 20nm should see about the same increase. 

Guess GTX 780 will be 6 Tflops and Super Titan 9-10 Tflops. Titan would have more flops if the whole chip would be working.