Sorry for the thread resurrection.
Could you please shred some light on how this really works in today's GPUs?
I vaguely remember (from what I heard years ago, so don't mind me if I'm wrong) that in XNA the contents were always discarded because of Xbox 360 limitations. So, if you wanted to preserve contents, the system had first to make a copy of the RenderTarget, switch RenderTarget and then copy the content over the RT.
But I suppose that nowadays (at least on PS4,X1, Switch, PC, but maybe not mobile) this is no longer a problem in terms of speed penalty. Is this right?
I'm trying to improve the impostor rendering of some objects in the distance, and it'd be great if I could use a big rendertarget and assign a little space 128x128 for each impostor, and render all of them without having to switch textures every time I render one.