Since a couple of version I noticed a pretty poor performance using network rendering. Our current setup:
Workstation with:
1080ti
2080ti
Rendernode 1:
2070 Super
1080
Rendernode 2:
1070
I'm currently rendering several shots with various VRAM consumption and I'm noticing a very poor scaling in terms of renderspeed. For example:
Workstation + Rendernode 1 --> 1:32
Workstaion + Rendernode 1+2 --> 1:47
The reason seems to be that the weakest rendernode has to finish their part, even if all other Nodes are already done and idling.
As far as I know this has to do with the parallel samples/tile samples, but I just can't to get a "perfect" settings.
So my question is, are there any magic numbers to punch in?
Network shouldn't be a problem, we got 100MB constantly.
Another observation: turning on/off rendernodes equals in different rendertimes. The given example above went up to 1:59 instead of 1:32 after testing the Rendernode 2.
Overall I hope there are some performance and usability boost in the making. 2 major improvemtns would be:
- headless rendering with C4D
- turning on/off GPUs at nodes just like in the host App.
- Ability to render frames per node (I know this works with managers like deadline)
Thanks in advance.
