According to a fresh report coming from 3DCenter.org, AMD has somewhat changed its roadmap, pulling HBM2-equipped Vega GPU to October 2016. While Polaris Ellesmere and Baffin GPUs sounds good on paper, these will target a completely different part of the market.
In case you missed it earlier, Polaris could offer performance similar to 390/390X, leaving Nvidia to reap the spoils of the high-end market with its GTX 1080 and GTX 1070 graphics cards. AMD has its horse for the race with the high-end Vega GPU but it was not scheduled for release before early next year.
AMD Vega GPU is a multi-chip module design based on a 14nm FinFET manufacturing process, just like Polaris GPU. It should be based on AMD's 5th generation Graphics Core Next (GCN) architecture and, more importantly, comes with 2nd generation High Bandwidth Memory (HBM2).
According to earlier leaks and rumors, AMD actually has two versions of the Vega GPU, Vega 10 and Vega 11, same as it did with the Polaris GPU.
The Vega 10 GPU is rumored to be a smaller chip with up to 4096 Stream Processors and this is the chip that AMD needs in order to compete with Nvidia's new GP104 GPU and Geforce GTX 1080/1070 graphics cards. The Vega 11, is a bigger chip, rumored to come with up to 6144 Stream Processors and compete with Nvidia's future GP100 flagship graphics card.
Both Vega 10 and Vega 11 should feature 4096-bit memory interface and HBM2 memory, if SK Hynix or Samsung manage to get those chip ready. Earlier rumors suggested that SK Hynix will begin mass production of its 4GB and 8GB HBM2 stacks in the third quarter while Samsung already started mass production of the 20nm 4GB HBM2 chips earlier in March and should get 8GB HBM2 chips ready later this year.
Hopefully, AMD will manage to speed up the Vega GPU schedule and get those chips out as soon as possible. October timeframe sounds like a good move as AMD could have those cards on retail/e-tail shelves in time for Christmas shopping season, which will certainly make its partners quite happy.