I was brain storming the other day with a cup of Cafe Bustelo and this is what I arrived at:
With GDDR5X they have the potential to go up to 16 Gbps. We have yet to see 12 Gbps, 13 Gbps, 14 Gbps, 15 Gbps or 16 Gbps used or revealed. This GDDR6 operates immediately @ 16 Gbps.
NVIDIA is planning to release Volta, but we haven't heard anything regarding what succeeds Volta. Considering this, I immediately thought back to Kepler when they refreshed the 670 and 680 (760 and 770), and they dropped the 780 Ti and Titan Black with faster memory than the 780 and Titan.
My thoughts here is that, they retain the ability to use faster memory for a possible Volta refresh. This will allow them to have another product stack to release instead of rushing out Volta immediately with the fastest memory they can use.
2080: GDDR5X 12 Gbps
2070: GDDR5X 11 Gbps
2060: GDDR5X 10 Gbps
Then they release the Volta Titan (Cut Down): GDDR5X @ 15 Gbps
Time passes until 2019, and refresh;
Volta Titan Black (full GV102 chip): GDDR6 @ 16 Gbps (if it has GDDR6 on the box, it would look great, especially if all the other cards carried G5X; would be a huge differentiator)
3080 Ti (slightly cut GV102, same or slightly more than original cut down Volta Titan): GDDR5X @ 15 Gbps (to have the Volta Titan Black distinguish itself from the 3080 Ti)
3080 (more cut down GV102): GDDR5X 14 Gbps
3070 (full GV104 chip same as 2080): GDDR5X 13 Gbps
3060 (cut GV104 chip same as 2070): GDDR5X 12 Gbps
Volta is also rumored to be on 12 nm, meaning that refresh means the manufacturing process has matured, resulting in better binning, higher clock speeds and higher memory speed (770, 780 Ti, Titan Black - because remember Kepler was a die shrink). Another great example of process maturity is Maxwell. Nevertheless, NVIDIA hasn't done the whole refresh thing in ages and there is nothing that follows Volta as far as we know:
Which is the only reason why I am considering the refresh option, because we haven't heard so much as a peep about what comes after Volta, not even a mention of a code name.
Again, I do not think that the consumer side will see HBM used. GV100 will most likely use it, but for enterprise applications only. Since we have all of this new GDDR memory technology, I doubt that NVIDIA would opt for HBM instead, since this GDDR5X and GDDR6 has less cost involved.Edited by BiG StroOnZ - 4/24/17 at 1:44pm