Hynix released Q1 2017 databook for graphic memory.
https://www.skhynix.com/static/filedata/fileDownload.do?seq=366
Noticed how Hynix HBM2 memory parts changed since Q3 2016:
Q3 2016
4GB 256GB/s 2.0Gbps H5VR32ESM4H-20C 5mKGSD 4Hi stack, VDD/VDDQ=1.2V Q3'16
4GB 204GB/s 1.6Gbps H5VR32ESM4H-12C 5mKGSD 4Hi stack, VDD/VDDQ=1.2V Q3'16
Q4 2016
4GB 256GB/s 2.0Gbps H5VR32ESM4H-20C 5mKGSD 4Hi stack, VDD/VDDQ=1.2V TBD
Q1 2017
4GB 204.8GB/s 1.6Gbps H5VR32ESM4H-H1K 5mKGSD 4Hi stack, VDD/VDDQ=1.2V Q1'17
RTG will prob use the only final production part H5VR32ESM4H-H1K for 8GB Vega high end consumer graphic card but with 409.6GB bandwidth. That would be slower than last gen Fury X 4GB HBM1 with 512GB bandwidth and also slower than Titan X Pascal 12GB GDDR5X's 480GB bandwidth. I cant imagined RTG to have cut down Vega graphic card with 4GB HBM2 memory that would have 204.8GB bandwidth that will be very expensive to make than much cheaper GDDR5 on RX 480 with large 256GB bandwidth and RX 470 with 211.2GB bandwidth.
Made no sense for RTG to go with low speed HBM2 after many failures, delays and setbacks, they should had scrapped HBM2 years ago and gone to GDDR5X instead.