Say if you bought a dual core machine, and an SLI based graphics setup, so you had two 7800 GTXs for example. Wouldn't this technology allow one processor and one graphics card to be used by one OS, and the other two by another?
Not yet is the simple answer. It is one of those things for down the line for now when the PCI-Express 2.0 spec shows up, or whichever ends up supporting virtualisation.
SLI or Scan Line Integration is not 2 logically separate GPUs that can be individually assigned tasks. They are one logical unit at the driver and frame buffer level, supported by 2 sets of dedicated GPU hardware. If you separate them at the driver/frame buffer level you don't have SLI, then you just have 2 GPUs. So talking about SLI separate GPU use in a virtualized set of OSes just doesn't make sense.
You either go with SLI and have each OS use the GPU hardware when it's their turn, or you go with individual non-SLI (not individual parts of SLI) GPUs and dedicate them to an OS just like the other resources.
It requires hardware and driver support, both on the card and on the bus. PCI-E can provide the proper bus support, if you have the proper PCI-E configuration. There are additional hardware connections on/between the cards themselves.
Comments
Originally posted by mattyj
Say if you bought a dual core machine, and an SLI based graphics setup, so you had two 7800 GTXs for example. Wouldn't this technology allow one processor and one graphics card to be used by one OS, and the other two by another?
Not yet is the simple answer. It is one of those things for down the line for now when the PCI-Express 2.0 spec shows up, or whichever ends up supporting virtualisation.
You either go with SLI and have each OS use the GPU hardware when it's their turn, or you go with individual non-SLI (not individual parts of SLI) GPUs and dedicate them to an OS just like the other resources.