I want to get some first hand experience of people running different cards like high performance cards for rendering with a much weaker card used as display adapter.
Unless it is a pure render box you will also want to use the higher tier card for viewport, substance painter, games or something else. So, do you actually have to connect both cards to the monitor and can switch in software which card to use or do you have to replug the weak card each time you start a render for example?
It’s going to depend on what your workflow looks like and how you like to render. I have a GTX 980 that I use to run my monitors and then an unconnected RTX 2070 to render with. The 980 is more than powerful enough for EEVEE and the monitors, so I never have to switch with the 2070. Also most software that uses CUDA will allow you to point them to the GPU to use. Cycles, Octane, Substance Painter, Davinci Resolve all allow this, so I just point them to the 2070.
But if you want to use the 2070 in something non Cuda you have to swirch cables? Thats the dealbreaker for me.
I can understand your point on that for sure. 
As I purchased my 2070 just for CUDA rendering, there is nothing non-CUDA that I would run on it. Anything non-CUDA I just run on my 980 and it does just fine for my needs. It’s far more important for me to have the full memory and speed of the 2070 then have it’s resources used up by the OS or other applications.
If you are referring to for example OpenGL for Eevee, you investigate about render offload.
On Linux, having intel iGPU as primary display and using PRIME, I can launch Blender so that OpenGL is handled by nvidia card using render offload. I suppose that having two nvidia cards it should also be possible, that is, choosing which card will handle OpenGL when launching an application.
But this perhaps would be better to ask on nvidia forums, for example here for Linux:
I’m not sure where for Windows, maybe in this OpenGL section:
Edit:
If I remember correctly, someone had said that when you have two nvidia cards you could choose with which card to launch an application, and these OpenGL settings were made from the nvidia settings utility. You look for discussions about how to render with Eevee in multiple instances with different cards.
Edit 2:
Here it is for Windows:
I dont use Evee but in general the dropdown selection “OpenGL render on “GPU”…” from rightclicking on Windows executables only seems to work if both cards are connected to the monitor. Selecting an unconnected card simply results in the one thats connected to the display being selected.
You research well about it and ask on nvidia forums. It seems silly to me that I can do it with the monitor connected only to my intel iGPU (motherboard) using nvidia PRIME render offload, while users with more than one nvidia card cannot.
Tbf I have not heard of nvidia PRIME until now. Also seems to be a Linux thing. And I only use Windows on my daily.
Prime technology is for Linux. Optimus technology is for Windows. I talk about it because it is designed by nvidia for computers with an intel iGPU and I can test it because I have an intel iGPU. I don’t have more than one nvidia card so I don’t know about it. I’m just saying that it seems silly to me that nvidia can develop a technology to be able to do it with an intel iGPU, but not between nvidia cards. This is why you should ask on nvidia forums.
Yeah I get your point. Will ask somewhere else and post my results here later.