questions about Cuda utilization of dual (NON-SLI) GPU resources in "newer" cards

Hopefully this thread can serve as a general informational resource for others who wish to combine multiple NVIDIA brand GPU’s (I’m a fanboy).

I have a 1660 Super and I’m having a 1080 arriving soon and I want to know if they:

* The VRAM is additive or parallel for both gaming AND development
* The 1660 can be used as a passthrough for video WHILE providing computational resources
    * play a FPS game (to stream to the AI) on the 1080
    * use the 1660 as PhysX AND AI operations
* anything else you can tell me that I obviously don't know because I'm just starting out with GPU programming...

If I tell tensorflow through CUDA to completley lock out, say, the 1660 for “stuff” and have the monitor hooked to THAT, will it do any harm?

Seriously, this is my FIRST ACTUAL POWERFUL COMPUTER! :)

Ryzen 7 2700x, 16GB-3200 DDR5, GT1660 Super, 20TB of storage across some 50 random hot swapping SATA drives both SSD and spinny-disc (it has flashy lights :) )