Quote:
Originally Posted by aufkrawall
I can hardly believe these figures. Almost twice as fast, just because of different API?
I'd really rather assume that is somehow a counting/calculation error.
You could check this with madVR settings which almost cost twice as much, e.g. more Needi3 neurons if interop performance is sufficient.
|
I'm just reporting my findings for my rig. Not trying to convince anybody or anything. Different hardware using different drivers using different settings will produce different results.
Quote:
Originally Posted by huhn
you should check the power states. dxva has rarely a clear effect on it. on the other side cuvid forces high power state.
just check it with GPU-z
|
Well what I've noticed is that when I use D3D9 old-path it maintains 450mhz/1250mhz throughout the video, where as D3D9 new-path or D3D11 with present a frame for every vsync, it fluctuates from 450mhz/1250mz to 1050mhz/1250mhz. Not sure why this causes the reported rendering times by MadVR to go down so drastically.
I'm going to try not using FSE while using D3D11 windowed mode to see if that bug that always forces me to go back to FSE is squashed. (the composition rate changes randomly and this mismatch causes the videos to be choppy. eg when the composition rate is 23.976 and I am watching a 60p video clip)
You were right that the dxva copy-back doesn't seem to affect the clock rates much. I think it's best for the card to stay at a single clock rate, say the 450mhz/1250mhz mid state instead of constantly fluctuating. low power state for my card is 300mhz/150mhz