Originally Posted by Thunderbird
I guess when you mix vdpau with opengl using glx_ext_texture_from_pixmap a memory copy needs to happen. Is it possible to prevent this? A friend of mine is considering using nvidia hardware in a broadcast product in which he in the end wants to render 4 720p streams simultaneously. I know right now vdpau offers only one stream, but would the hardware be fast enough for this? (a pci-express 2.0 link would likely be required for the bandwidth)
Yes, at the moment a copy needs to be performed, from the VDPAU output surface to the X pixmap that TFP uses. We're aware this isn't as perfect as it could be. Also note that there's currently no synchronization between VDPAU performing the copy and TFP reading the X pixmap.
As an aside, it sounds like your friend is creating a professional product. I'd advise him/her to contact the NVIDIA professional services group to discuss his/her requirements. That group may be able to help prioritize any engineering effort for his/her project.