I just tested DXMD while in DX12 mode. Instead of doing my normal run around I just stared at a wall (so no NPC's would walk in front of me and so FPS was steady).
GPU usage from MSI Afterburner was @ ~96% the whole time.
Gotta say I just found out about the MSI Afterburner remote server & phone app and it made testing this very nice. Since I was just looking at a wall my fps was completely stable but there was still high CPU usage and other game logic / NPCs going on behind me. I was able to modify the clocks / voltages from my phone and then record the changes without having to touch my computer at all
Anyway, going from the stock of 1000 (and my own undervolt) to 1150 (15% OC) had 11% gain in this game, still with ~96% GPU utilization. The power usage I took from my UPS so I'm not sure how accurate it is, plus it includes other things like my monitors. It does seem to have jumps in ranges for what it reports off, but I included it for anyone curious. "Idle" usage is around 170w (browsing web etc). The 15% OC used ~41 more watts going from -72 to a +72 voltage modifier.
I tried doing HBM OCing (500, 525, 550), but with the Relive drivers AMD seems to have locked them down again, and even enabling both settings did allow me to OC it finally, but it made absolutely no difference, so I'm not sure if it was just a display change or if the memory isn't bottlenecking at all so made no difference. Since it wasn't making any difference I didn't test any further.
Well, ROTTR was actually the only game I have installed with a benchmark, doh. 1440p, mostly max settings. Average of 6 runs. Fury air.
1100MHz Core / 545MHz Memory / 60 CUs = 53.85 fps avg
1100MHz Core / 545MHz Memory / 56 CUs = 53.06 fps avg
1000MHz Core / 500MHz Memory / 56 CUs = 49.73 fps avg
So 6.7% gain for 10% Core / 9% Mem OC. Partial unlock doesn't do much, just another 1.5% over default CUs. But that's an 8.3% total gain over a stock reference Fury in this (only one sadly) game.
Can you try w/o doing memory OC? I found that ocing the memory had 0 impact when doing my testing.