The system is a dual socket Xenon E5 2670 with 32GB of ram, with two R9 390. The two graphics cards are not in a cross fire configuration. They are meant to server two separate seats when running linux. When booting windows 10 with the crimson 16.4.1 drivers installed, the system either freezes before showing the login screen or some time after logging in, within the first few seconds. The second R9 390 is in a PCIE slot that is connected to the second CPU. There is not enough room left in the board to get them both attached to the first CPU. Some times Win10 manages to notice the boot problem and it switches back to the generic software driver, then I can see that the amdkmdap processes caused a lot of event logs - reading stopped responding and has successfully recovered from the last boot.
It seems that only the displays froze but not the entire system. I still hear win10 wrangle that poor harddisk while the screens show no change and no reaction to mouse movement.
Both graphics cards work fine with linux. I can also get win10 to boot properly when the crimson driver is not active, or when I only have one graphics card installed in the system.
Researching these forums I found notes on changing the registry to disable ultra low power states ULPS. I did disable all enabling settings, but left the timeouts and the NA settings as the were. I did that with only one card installed. Now two reboot later I inserted the second card again. System freezes again.
Is there something I can do to get better error logs from the amd drivers? Something that would help resolve this problem...
Should I try Catalyst?
My plan B is trying to run win10 within a VM and using VT-d to attach just one graphics card to the VM...