Dual x8 power connectors, is this a dual-GPU card?
Single GPU, just one Fury-sized (or slightly bigger) die. Same Stream Processor count, and the raw TFLOP rate increase is almost identical to the clock speed increase. Outside of the Single Precision performacne doubling to account for packed operations, 'big Vega' is looking to be a Fury clock-bump when it comes to consumer performance. The big question is memory: it's stated to have 16GB of HBM2, but the bus width compared to Fury is half (2048bit vs. 4086bit) which would indicate only two HBM2 dies vs 4 HBM dies. But neither SK-Hynix nor Samsung have even advertised 8-stack (8GB) HBM2 yet. Memory bandwidth takes a slight hit compared to Fury, but for consumer workloads memory bandwdith has never really been a limiting factor.
I can't think of any reason this card would have a TDP below the PCIe max of 300W, being aimed at the performance-above-all 'deep learning' market.
With RTG's new enterprise first, client later approach, I'm not holding my breathe for a consumer Vega release in Q2.
With this card just scraping in at the end of the '1H2017' shipping windows for Vega, I wouldn't bet on consumer cards coming until Q4. Possible a Titan-esque high-margin consumer 'halo' card using binned dies in Q3.