HBM2 Memory is Stacked Directly Onto the GPU's Substrate Package.

HBM2
+Less Power,
+Easier to Cool,
+Smaller Memory Chip Footprint,
+1024-Bit Interface
+8GB Per Stack (GPU's Usually use 4 Stacks, so up to 32GB for HBM2)
+Faster Memory Lane's to GPU
-Higher Cost
-Lower Production Yield

GDDR5X
+Cheaper than HBM2
+Higher Production Yields
-More Voltage
-More Heat
-More PCB Space Used
-Slower and Longer Memory Lanes to GPU
-512-Bit Interface
-1 GB Per Chip Max


GDDR5X is Simply an Extension of GDDR5 to enable Extreme Clocks (10-12Gbps)(2500MHz - 3000MHz)

GDDR5X Total Bandwidth still doesnt even match the HBM1 Memory used on the AMD Fury GPU's
However HBM1 was limited to 4GB, Thus the Reason why AMD Chose to go w/ GDDR5X short term, Because 4GB isn't enough for some extreme Resolutions and VR Setups anymore. Next Gen Flagships will be 8, 12 or 16 GB.