With first gen HBM, memory clock steps are at a certain amount: 500Mhz, 545Mhz, 600Mhz etc. like explained here http://www.overclock.net/t/1592384/fiji-bios-editing-fury-fury-x-nano-radeon-pro-duo/820#post_25241381 Maybe with HBM2 it is the same? And this might be the issue.

I had the same issue with RX 580 and Z97 Motherboard. I read about this problem here as well. The issue is that the algorithm is creating some kind of interference with the motherboard when under load which causes the the Wificard to lose connection. I tried to change the overclocking (load) of the card and it made it better. This somewhat solved the problem. Other and better solutions would be to have a different Wifi card (eg USB one) or use PCIe X1 to 16X risers and move the card from the motherboard away.

Good luck

Actually I have experienced this as well. With some (AMD) cards when they are running at 100% I can lower the voltage a bit more than when starting them from 0 to 100%. I think it is a normal behaviour and the problem is that the power delivery on the GPUs is not fast enough to keep up with the power demand. Maybe a future BIOS update for your cards will resolve this issue if the manufacturer sees it. I think there is nothing wrong with your cards.

You have a very decent PSU. Keep in mind, that if the PSU pulls 1000w from the wall this means your rig is pulling arround (1000 * 94% efficiency = ) 940w. Your PSU should be able to deliver 1200w output, so arround 1275w from the wall. I think you are good to go from that perspective. I guess your PSU is faulty or the connections to the GPUs are bad and the voltage drop there is to big. Do you have any PCIe 8 Pin splitters on it? Maybe their Ohm resistance is too big.

Use Wattman in the AMD Settings and you should be able to set it as low as possible. Or BIOS-mod the card.