The OP is being a little disingenuous because I'm not sure Nvidia is supported anyway, but you lose some (not enough to go from 1080TI to 1060) performance. Something like 5-10%, depending on if you send the video signal to an external monitor or back to your laptop's screen.
For gaming. Also don't forget the annoying latency. But if you need to transfer from/to GPU memory all the time, like in large Deep Learning training datasets, you downgrade your GPU significantly; that's why I wrote 1080Ti -> 1060. You already have high end PCIe GPUs starved and waiting on memory transfers in Deep Learning all the time.
You could have observed something similar with e.g. SATA 1/2/3 SSDs and M.2 PCIe. For normal workloads, each new generation was slightly better performing, i.e. booting OS etc. But once you went into processing 4K RAW video, only M.2 PCIe was usable.
