Originally Posted by ladcrooks View Post
shame new tech, HBM2, is going to be hindered by cost - just like SSD, was years ago, and now look!
Wouldn't it be great if the vga card could feed off your ram ? And then you decide how much memory you want - 64 GB wink.gif
Then you will get someone as they always do ' Will it play Crysis? '
Honestly I don't see GDDR5X as being usable on any chip designed to use HBM, there's just no point. A GDDR5X chip would require a third type of GPU headder and while it's CLAIMED that it can hit 12 to 16, I doubt it will make much difference compared to HBM2. 5X cards are gonna be more pricey than 5 cards and probably only result in a 30% increase in speed while taking market share away from HBM2 and GDDR5 cards - resulting in an over-complex market just like when we were transitioning from GDDR3/4 to 5. It may well be worthwhile to produce 5X cards if it gives more in the +70% range or it may be worthwhile to run higher speed HBM ram on an interposer.
The problem with HBM2 is that to effectively use it you need a whole sheeton of shaders, there's no point in having an 8GB depth 4096 bus if the read will simply outrun the cores.
The biggest improvement we can get right now in GPU is the addition of a SEPARATE frame-buffer on-card so the creation of frames isn't interrupted by the reading of them. Xfire/SLi could then direct-write the frame buffer from additional cards without interrupting the primary card.
Gsynch on monitors was just a gag by nV to try to make monitors proprietary to nV cards but if AMD took it to court the Fed would knock nV's proprietary block off.
Just another case of nV offloading what THEIR hardware is supposed to do on other hardware.
We need that separate frame buffer and we need it able to handle direct from bus writes.
We need that separate independent frame buffer more than GDDR5X, HBM2, or 14nm dies.