Hi, I’ve tested nDisplay for creating setup with 2 PCs and 4 screens. Screens are placed in one row, displaying wide panorama. As I tested and compare it with setup based on one PC and 4 screens connected to one GPU it has the same performance as 2 PC with 4 screeens connected via nDisplay setup.
My questions are:
Does nDisplay boost performance by joining compute power of all nodes? So every instance is handled by each node?
If not, what can I do to use available hardware to boost performance in Unreal? I have no limits in hardware spec (2-3 PCs, every with 1080 SLI) but there are software limits. Also if nDisplay is using only master PC’s compute power so what is the point of joing like 3 PCs to 6 screens if I can use all available ports in 1-2 GPUs on single computer, or use splitter?
For question 1, Yes, it should. Just started investigating nDisplay myself, but from the unreal docs “Each of those Unreal Engine instances is responsible for rendering a single segment of the same 3D scene to a single screen or display”. The master is in charge of synching, and in theory could do both synch and render one of the instances, I think.
For question 2, what are the software limits? What FPS do you get? Do you have source, or is this 3rd party? If the latter, it could be the software design is the bottleneck. For example, if the software is not yet coded to use SLI, you won’t see much if any improvement by adding a second GPU. The newest version of NVAPI sounds like it would help with this.
For question 3, Multi View Rendering should help and should be doable on an RTX/Turing machine with Unreal if you build from source. Unfortunately, Nvidia has not yet updated their branch of the UE4 engine. It’s still on 4.18 and is using the older Pascal based VRWorks, not VRWorks 3.0. NvLink may also be helpful, since it would allow for 4 GPUs https://github.com/NvPhysX/UnrealEngine/tree/VRWorks-Graphics-4.18