consider zero-copy rendering

[bmegli]

In hardware accelerated video processing (depending on hardware) we usually end up with decoded video data already on GPU side.

It is possible to use this data directly for rendering, without transferring video data from GPU to CPU and from CPU to GPU again (e.g. as texture for rendering).

This technique, called zero-copy, is generally most efficient and fastest way to decode & render video data. Unfortunately it also needs tight integration with low level display abstractions used by the OS (e.g. video data from VAAPI surface directly used from EGL with DRM or X11 display),

I don't see possibility to do this with Unity, not without access to its closed source part.

If you think otherwise, please kindly let me know.