Looking Glass is targeted at extremely low latency use requirements on the local computer, it is not designed to stream over a network or pipe but rather through a block of shared memory. In current testing at a refresh rate of 60Hz it is possible to obtain equal or better then 16 milliseconds of latency with the guest. If the user doesn't care for VSYNC this can be further reduced to under a few milliseconds on average.
Unlike network based streaming applications, Looking Glass does not use any form of compression or color space conversion, all frames are transferred to the viewer (client application) in 32-bit RGBA without any transformations or modifications. This is possible through the use of a shared memory segment which enables extremely high throughput low latency guest to host communication.
Anything recent graphics device that supports OpenGL should do fine, the client application does not perform any heavy graphical work.
At this time only Windows 10 is supported with either any video card supporting DXGI Desktop Duplication or the NVIDIA Capture API (Professional cards only such as Quadro).
Note: The NVIDIA Capture API is disabled for now, when Looking Glass nears a beta release it will be updated and made available again.