I'm going to be honest with you. I believe that this was a test GPU for what AMD has coming next year.
All AMD and Nvidia GPU's will be going to 14/16nm technology probably starting in the second quarter of 2016. The Fiji GPU was testing HBM, and probably some other things that we do not even know about. AMD has not only learned a whole lot about what they were testing, but they expanded the supply chain for all of the new GPU technologies that they are using. For example, Until July, the TSV channels that carry the signals between the stacks of memory, were not even in full production. AMD had to pay to create the production lines that put all the various components on the interposer (huge piece of silicon) that is the base of the Fiji GPU. They have spent years and many millions of dollars getting everything into place.
But the real goal in this entire process was to get everything ready for 14/16nm production. That is when they can make the chips they have been drooling over for years. Initially, they expected to be able to do some of this on 20nm, but TMSC never created the fab technologies they needed to make high performance GPU's on the 20nm node. So they had to keep using 28nm. Until now. They have been busy getting 14/16nm silicon from the fab, and testing it for awhile now.
AMD has stated that their biggest GPU will have between 15 Billion and 18 Billion transistors. Compare that to 8 Billion in Fiji. They have also said that the new smaller process will be roughly 65% faster with the same number of transistors. So double the number of transistor, and you will likely be doubling the speed of the GPU.
By the time the new GPU's are ready to go into production, HBM2 will be ready. With HBM2, AMD will be able to put as much as 32GB of memory on a single video card. HBM2 will not only have larger capacities, but there will also be higher speeds, which will allow higher bandwidth. I have seen numbers between 1.2TB and 1.5TB per second tossed around. Both numbers are likely to be true, just with different speed HBM2.
And power consumption should drop quite a bit more for the same number of transistors.
So add up all of this, and we should have some incredible cards that blow away anything we have today. Obviously, we do not know the exact shipping dates yet. But AMD is focusing on the high end due to the profit margins. They may actually be working on multiple GPU's at once. But I do think we are within 6 to 8 months of seeing some of these becoming available.