On Maxwell (example:M10) CUDA is available just in full-chip 8Q configurations. But as for I can understand, NVENC is not "hardwired" to CUDA units. So, if one could use NVENC directly, that could give much more flexibility since much more guests could be run simultaneously. Is that technically possible? Are there some samples or documentation?
Didn’t get your question. Nvidia Encoder is a specific hardware ASIC on the GPU for defined Codecs. It has nothing to do with CUDA. And for sure it can be used for several simultaneous sessions. This is exactly what is done in the VDI space for Remoting protocols like Citrix HDX3D Pro or VMWare Blast