HBM memory integrates tightly with the GPU die, by sitting right next to it on the silicon package. This approach promises smaller PCB real estate usage, but AMD has opted to keep their card full size. Arranged around the GPU is almost nothing but voltage regulation circuitry, probably to keep sufficient space for the heatsink and fan.
The IR35217 voltage controller seems to be a new design for RX Vega. It feeds its six PWM phase signals into six IR 3598 phase doublers on the back side of the PCB. These six chips now drive twelve power phases, each made up of an IR6894 and IR6211.
The HBM2 memory chips are made by Samsung. You see two stacks, which add up to a total memory capacity of 8 GB. If you take a closer look at the picture above, you will notice that the space between the GPU and memory chips has been filled up with a clear/gray substance. This is in contrast to the Radeon RX 56 GPU from our other review, which has an empty space here, just like we've seen on the R9 Fury X. A hint to these differences could be the "made in Taiwan" print on the GPU of Vega 64, whereas Vega 56 uses a GPU "made in Korea".
AMD's Vega 10 graphics processor is produced on a 14 nanometer process using 12 billion transistors occupying a die size of 486 mm².