This has like zero usability for home users, right?
It's just sharing GPGPU power between tasks(from different users) ,right? A single user running multiple GPGPU tasks would see no difference,other then artificial performance walls.
As @at80eighty mentioned, AMD's solution is called MxGPU, and is leveraging SR-IOV to split a GPU into up to 16 virtual GPU devices per pGPU (Dual GPU cards can do 32 users). NVIDIA's approach is Software based, using a virtual driver on each client to interface with the graphics system controlled at the Hypervisor level. This Hypervisor driver communicates with the installed GPUs to complete GPU work on behalf of the client driver and return the results. AMD's solution is Hardware based as passes a virtualized version of the card directly to the VM. They both have their pros and cons. Intel doesn't really have anything for this market, but they don't really have a GPU with enough performance to consider splitting it up amongst multiple users..
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.