Passthru GPU, no iGPU, and hardware transcoding while gaming?

[BurntOC]

Hey all, so I would like to get this wise group's recommendation on what to do regarding my setup.  Here's what I have at the moment:

 

CPU:  3700X, until 5900X arrives

GPU1:  3060 Ti (in PCIEx16_1 slot)

GPU 2: GTX 1660 (in PCIEx16_2 slot), or I also have a GTX 1050 Ti spare

Intel i350-T4 Quad NIC (in PCIEx16_3)

32GB RAM, several drives, including 1 NVME mode

Running the latest 6.9 beta 35 version of Unraid.

1-2 VMs, about 15-20 containers including Plex, Jellyfin, and ChannelsDVR

 

My goal was to passthrough the 3060 Ti to a Windows VM so I could game on my Unraid server without running another box.  I figure worst case situation is I'm running something like Call of Duty Black Ops Cold War at 1440p epic or ultra settings , while also transcoding a couple of streams (media center streams from Plex, Channels, etc. to be clear)  if there's any post-processing/upscaling going on.  That's why I was passing the 3060 Ti through and still running the 1660, so it could handle the encoding (which I guess it does better than the 1050 due to Turing, anyway).  In my previous build, I had an iGPU and that worked just fine when I was passing the discrete GPU through.

 

Now I don't have  an iGPU any more.  I also can't seem to get the 2nd GPU to provide display out when booting with my STRIX X570E, but it does show its presence in the PCIEx16_2 slot.   I've been fighting it a bit, but I began wondering if, even though my CPU isn't an APU and doesn't have an iGPU if the hardware encoding and decoding capabiities are there such that I don't really need that 2nd GPU for handling that use case I laid out.  If the CPU handles that, thenI imagine I don't really need anything else other than to pass the 3060 Ti through to my VM.

 

I figure if I try it then it may work, but I wouldn't know if it was doing software encoding/decoding vs. hardware.  I don't really want the overhead to impact gaming performance, and I would like to know if the CPU can handle this as I imagine the Ryzen 5000 ought to be in the same ballpart as the Turing, at least.  That would also allow me to ditch the 2nd GPU altogether, saving energy, heat, and allowing me to run in x16 mode vs. my current x8/x8.

 

I would appreciate any insights!

 

 

Edited January 5, 2021 by BurntOC

[BurntOC]

So I got an initial response on Reddit that seems to make a lot of sense.  In a nutshell, the response was that:

 

Yes, with a 5900x I could probably power through, but there may be a more elegant way.  This would involve:

1. swap the 1660 to slot 1 and the 3060 Ti to slot 2 to address the POST issue and as the 1660 will be "always available".

2. install the nvidia driver plugin (which I'd not done as I thought it would conflict with the VM GPU passthrough for some reason).

3. reconfigure the VM as needed to address any bus/slot changes

4. configure containers to leverage the nvidia driver builds where available