Hello there,
I’m having issues setting up properly nDisplay for a cluster using mGPU - in short : I can’t find a way to dispatch a viewport rendering to a secondary gpu even after following steps described on https://docs.unrealengine.com/5.1/en-US/ndisplay-overview-for-unreal-engine/#multi-gpusupport
Here is the setup :
- 2x Quadro P6000 (R513.91) + Quadro Sync II (v2.02), but they are not in SLI/NVLink.
- Each GPU outputs a 4K screen
- Unreal 5.1
- nDisplay config is one cluster node with two viewports. One viewport’s GPU index is set to 0 and the other one to 1. The config setting regarding Multi GPU Mode is set to enabled.
With this setup, it seems that only one GPU is working for both viewports (on task manager GPU 0 is at 100% and GPU 1 at 0%), resulting in bad framerate.
I currently have a workaround to create two separate cluster nodes on same IP with only one viewport for each, and add the custom command line dc_gpu=0 or dc_gpu=1 for either instance. This is not ideal as I would like to avoid synchronizing some logic in my project.
I noticed that the nDisplay configuration export json file (.nDisplay) has no entry related to Multi-GPU mode, how is that supposed to be taken into account ?
Feel free to ask for more information, and thank you in advance for your help!