Took me a couple of minutes to get my head around the water flow path caused by the end-of-card inlets.
It seems to cool half of the first VRM bank on the way in, then a quarter of the flow is diverted for all of the other VRM bank before rejoining at the GPU/VRAM coldplate, and then cools the other half of the first VRM bank on the way out?
Given how much machining and effort went into routing out two-layer, multi O-ring channels for the VRMs, I'm really quite surprised that the area of liquid contact of those VRM plates is both tiny and polished smooth. It seems like a waste of all that effort to get LOTS of water at a high-flow/low-impedence past those VRMs and then barely make the effort to do anything with it.
Perhaps this is just a side effect of very limited clearance around those VRMs indeed - I guess the clearance is so ridiculously tight that it mandated this little notch in one of the VRM plates:
View attachment 204059