How is the gpu power dispatched between preview & render ?

[Jukyan]

Hi,
I'm working on a very big 3d project (15k master for 5 full HD beamers), rendered with Octane. Project for which i'm about to buy a quad gpu new workstation.
I need to know how octane for C4d works to get decided if I buy a second rendering-dedicated computer, or if I can render & work with the same computer.

I'll constantly need to use the Live Viewer to preview my work & materials.
But how does it work when I render something at the same time ? Is it possible to dedicate one gpu to the live viewer and the others to the rendering ?
But then I loose its rendering capacity when I'm not previewing something : /
Or how does the live viewer rendering interfere with the queued rendering ? Is there some memory / perormance interference ?
I'll need all the possible power for the background render, the live viewer performance is not so important, but must be able to have the same memory limitation as the rendering. (I'll have a very high poly scene with many trees instances.)

Is it possible to dedicate only a gtx1070 to the live viewer, and use 3xgtx1080 for rendering ? Is there a better option ?
Or is it just better to have a second computer only doing the rendering with 4 gpus, and my workstation with just one gpu for live-viewer ?
The cost is not so important, only the efficiency is. In any case I'd prefer to have only one computer than having to manage a secondary one, if its possible to have an optimum performance with only one.

So, performance-wise, what is the best option ??

[aoktar]

For performance you can set priority for monitoring gpu. This is first options to try. Much better to disable one gpu and leave that for OS+C4D. It have to be monitor connected gpu. Also this gives full VRAM usage.

Also you can check LV's update options. And settings's performance options. You can disable geometry, etc.. update controls for performance or you can disable "all movable" option for manually assignments.
Another trick is to use render_instances. Less geometry compile times, vram usages. Plugin spends some short time on iterations of objects/materials. Rest is updating section. It can take much time for updated geometries. You can see this phases from info line onscreen.