I hope you can come up with a solution or even a workaround to help us out. The keeping resident video memory That are not helping with the display. And the other secondary (up to 12 or more ) gpu's are dedicated to rendering purposes. We generally leave the display card(s) just for displays. In our (me and people who are doing GPU rendering) workflows. I also found an article about Process residency budget regarding the changes in WDDM 2.0 GPU oriented computations is mainly done by using nvidia cards. I have no knowledge of other manifacturers like AMD having the issue. If you have a 24 GB Quaddro or Tesla GPU you end up having ~4.5 GB VRAM unavailable If you have an 8 GB (1070) card then you end up having ~1.3GB VRAM unavailable

If you have a gpu card with 4GB then you end up having 700MB VRAM unavailable We use gets all the GPU's work on the single rendered image. They are working as separate cards but rendering software The video cards are not connected as SLI configuration. But also all GPU's that are not connected to any display, just acting as dedicated CUDA computing (render) cards are suffering from this issue. This problem occurs using in every nvidia graphics card. This is a very well know problem people in my industry. The application (octane render) reports an amount of unavailable GPU memory that can not be accessed.

We did not have this issue with windows 7. Problem takes place after upgrading to Windows 10 (from win7). I'm using nvidia graphics cards for 3D rendering using CUDA computing.

Daz3d not using gpu