I'm expecting a first Pascal to hit the market in 2Q16. But it's not a given that it will use HBM2 at all. GM200 vs Fiji situation is clearly showing that there's still some juice left in GDDR5 and considering that HBM2 is a new and expensive tech the first Pascal which will end up in GeForce may be using GDDR5 again.
We really don't have anything to even speculate about yet.
NV's roadmap highlight new tech, it doesn't state anything about the end products. 3D memory is a technology that may or may not be used in a product based on Pascal architecture. Pascal generation will have lots of chips in it and I'm pretty sure that most of them will end up with LPDDR3/GDDR5 and not HBM2. There is a reason why Fiji is the only GPU with HBM in AMD's 300 series. Same reason will apply to Pascal and HBM2 as well I think.
When you're thinking about bandwidth usage per frame you have to remember that for the next 3-5 years we'll be limited by that DDR3 pool of Xbox One here.
I think 2Q16 is very optimistic for Pascal release considering how much there has been issues with getting 16/14nm pipeline up, getting delayed regularly after 20nm was dud. GDDR5 isn't bandwidth limited, but takes a lot PCB space while being inefficient for cooling and requires more juice than HBM. HBM2, even 1 already, tossed all worries about bandwidth becoming issue out of the window while making cooling it more efficient and takes less juice to run. How expensive HBM is over GDDR5 I don't know for sure, but afaik it's supposedly cheaper solution than GDDR5 in long run.
Pascal is already taped, according somewhat reliable leaks, and it uses HBM2 in its design. HBM brings to table better power usage, hugely increased capacity over GDDR5 while using a lot space on PCB as it gets integrated into die. I will be very disappointed in Nvidia's Pascal line up if after all HBM2 hype they have been generating for it they use GDDR5 for all or mid to low tier cards. HBM will be future mem tech for GPU industry and introduction of 16/14nm is best time to embrace future. We get more powerful and efficient dies with superior mem tech.
Fiji architecture is only AMD arch to use HBM is because it's only AMD arch designed to support such memory. 300 -series is refresh of 200 -series and bases on Hawaii arch that is designed for GDDR5. You could argue that there isn't enough HBM to go around for larger scale release, but we really don't know any factual numbers on it and like I pointed out it's about arch supporting it. You just don't slap next-gen tech to last-gen arch and expect it to work.
VR development will get huge boost from possible performance given by Pascal, and AMD's rivaling next-gen arch, allowing higher resolution solutions etc. Not even mentioning it very likely will make single GPU 4k gaming possible for PC space, as I don't consider 30FPS with lowered settings to be
"4k PC gaming".