I will try to ask few simple question.
It is possible to get information from the GPU about what has beendrawed on screen
It is possible to read some part of memory from the GPU
It is possible for shader to be readable.
PS : i am new in deep learning of 3D, i am used with java3D. So i am try to anderstand what can be done by GPU or not. My goal is to implement picking. So i need to know if i have to organize my object information hierarchy in CPU processing or if i can expect some work from GPU, like ray tracing information test from a shader for example.