"install 6 x VGA cards on the black slots first, and install others in the white slots."
So it seems the answer is in the fact. With this mb you have to populate the parent PCI sockets first ( the black ones) then go back and fill in the white ones as they shares PCI-e lanes but are ordered by the first back being 1.0 the first white being 1.1. Second back (16x 3.0) being 2.0 and the white one above that being 2.1..
You get the idea. since I had 11 cards I was using them in order leaving PCI 7.0 unused whilst using 7.1
This seems to be the issue. So I moved GPU 10 down to the last black PCI port and the rig has been up and running for 3 hours with a higher over clock than I've had in over a month with no issues.