This feature is not used on desktop systems but on laptop PCs that have an iGPU from an Intel or AMD processor and a dedicated NVIDIA GeForce GPU.

Technically this can be used on desktop systems. When you have your monitor(s) plugged into the iGPU or another dedicated GPU, but you attempt to access another GPU (for graphics/display tasks), the OS will route that data from the monitor-less GPU through that iGPU/dGPU. This is my set up with a 9070 XT and RTX 5090, where the 9070 XT has all my monitors plugged into it. If I force a game to run on the RTX 5090, the performance loss seems to be from 5% to 15%, which is a nice trade off since the RTX 5090 can now idle at around 5W or such.

The MUX switch used in laptops from 2020 to now is a physical IC that switches the eDP/internal HDMI connection to the internal display on a laptop when the dGPU is used, so the dGPU directly sends data to the screen without having to go through the iGPU. The performance loss for the old school passthrough mode this used to be up to 15%, but with the not-so-new CASO changes in Windows 11 (and improved iGPU integration in newer APUs/CPUs), the loss is now around 3% to 5%, so using a MUX switch is not as needed as before but I believe its cheap for all OEMs to implement now so probably doesn't matter (having a MUX switch was considered a "high-end feature" in the performance laptops of 2018 to late 2020).