I'm measuring the CPU and NIC on the slave Windows machine.
I understand (from a high level) the differences in the protocols VNC/RDP. Upon further thought, I think that the difference in the CPU load has to do with the fact that VNC offloads the render to the GPU on the slave machine, whereas with RDP it is rendered on the local machine.
In terms of the network traffic, I think VNC is sending pictures/video so it is a "sustained bandwith" stream whereas RDP is sending bursts of traffic (the drawing instructions). So, although RDP is measuring instantaneously higher bitrates, it is likely sending much less data than VNC. It just depends on how WIN10 is measuring it.
The difference in CPU usage is worth noting, however, especially given the fact that alot of sample libary slaves are really working the CPU.
Neither one of those solutions is ideal, but given the differences in CPU load and the OpenGL errors/crashes I'm probably going to stick with VNC until I find something better. I left a VNC session open to monitor a VEPro slave for 12 hours, and it didn't crash, whereas it will crash about 90% of the time on RDP if left open/idle for more than about 2 hours. Anyway, I hope this thread is helpful to folks out there trying to get to the bottom of all of this.
Regards,
Jason Todd Shannon