Is it possible to over provision the frame buffer memory allocated to a pool of vm's? For example, if I have two K2 cards in my server for a total of 16384MB frame buffer.
The NVIDIA GRID quick start guide shows the following choices to assign a desktop in vSphere 6. I am using Horizon View 6.1.
So, if I stay within the physical limits of frame buffer and I want to assign 1024MB to each desktop, I can then only provision 16 desktops.
But, can I over provision the frame buffer assignments, by creating a pool of say 24 desktops, each with 1024MB assigned?
I think I was able to answer my own question...which apparently is Yes, as long as you don't need to power on more VM's than the physical resources can support.
I tested this by making only one GPU available to assign as a "Shared PCI Device". I edited two desktop vm to use the grid_k280q profile (4096MB frame buffer). I was then only able to power on one VM. Attempt to power on the second VM resulted int this error:
I think I was able to answer my own question...which apparently is Yes, as long as you don't need to power on more VM's than the physical resources can support.
I tested this by making only one GPU available to assign as a "Shared PCI Device". I edited two desktop vm to use the grid_k280q profile (4096MB frame buffer). I was then only able to power on one VM. Attempt to power on the second VM resulted int this error: