Support Multi-GPU cluster environment

I am aware of vrcluster and am looking at it’s source, but my environment is unique and I’m not sure vrcluster is a good implementation for my application. I am hoping someone from Epic will be able to tell me if what I have in mind is even possible with Unreal.

I have visualization projects for a couple different clusters. Both clusters run Linux and have multi cpu, multi gpu nodes. Both are projector/display arrays with GPUs dedicated to individual displays/projectors. So as an example, if I have a 4x4 array of displays, I have 5 nodes, a server, and then four client nodes with four gpus per node. Each gpu is dedicated to one specific display in the array.

My initial thought, not having experience with this and wanting this system to be as painless as possible for people who work with it, Is to make a server using Unreal, that spins up 4 clients on each node, each of those clients using one of the four gpus in the node. Is this possible? Is it possible to have four instances of a “game” running on one machine, each using a specific GPU device? Or a single client game that is rendering to four gpus (not sli)? The hardware is more than sufficient to handle four instances of these visualization projects, I’d just like to know if this is even feasible or if someone can point me in the right direction if my idea is off base.

I’m looking at doing something similar, did you end up having any luck? I can imagine you can use the consolevariables.ini to select the correct adapter or use the nvidia control panel to force it for each ‘game’ as long as they are named independently. But if there is a better way to do it, that would be great to know!