For an upcoming project will deploy a Threadripper workstation with 2x RTX 3080 Ti GPUs, one used for compute the other for graphics.
Will work on it a team of 3 specialists from different locations, so the possibility to get different virtual desktop is fine. Still due the kind of application they need to access the GPU hardware both for CUDA compute and also Vulkan hw accelerated graphics.
I’m a bit confused if this might be possible under NoMachine Workstation. Looking at this feature request in the comment is written dismissed out of scope, but in the status implemented, also I’m not sure is relevant for my case.
Just wanted to add the system will be dual boot Windows 10 and Kubuntu 20.04 LTS
My interest lie on the Linux side, due I know virtual desktop it’s not supported on Windows.
Also want to better specify I’m not especially interested into hardware based video encoding (even if will be useful of course) but the possibility to access trough the virtual desktop directly to hardware rendering features trough Vulkan and compute trough CUDA. This not necessarily having a vGPU partitioning (reserved to Nvidia workstation GPUs ad far I know) but just sharing the whole hardware resource among the different virtual desktops.
Sorry, I overlooked your exact requirements. At the moment it is possible to accelerate OpenGL apps by means of VirtualGL but not Vulkan, as a “VirtualVulkan” doesn’t exist. The reasons were discussed in this VirtualGL’s project page: https://github.com/VirtualGL/virtualgl/issues/37.
As far as I know, nothing should prevent applications running in a NoMachine virtual desktop from accessing the GPU for CUDA computing. We tried a simple CUDA program and it works, with the help of VirtualGL, since the CUDA program was using OpenGL for the final rendering.