Network Rendering

[Spotter]

I'm interested in network rendering for Indigo.

How is the workload distributed? Does each node have a full copy of a scene in memory? Do the nodes communicate with each other to complete a task, or does the main client PC simply divide the task up and each node does it's own portion?

Say, for example I have a high res scene with lots of polys. Normally, I would need a PC with 8GB of memory to render this. Would it be possible to have 3 networked nodes with 4GB of memory each to complete the task?

[fused]

Hey Spotter,

each node has a full copy of the scene in its memory. The rays bounce around in the whole scene, so they have to. The nodes do not communicate with each other, they only communicate with the master.

Say, for example I have a high res scene with lots of polys. Normally, I would need a PC with 8GB of memory to render this. Would it be possible to have 3 networked nodes with 4GB of memory each to complete the task?

No, unfortunately not. If the scene needs 8 gigs of ram, each render node will need 8 gigs of ram.

[Zom-B]

regarding RAM usage there are some things that can be done to reduce it:

Scene Size
- Use instancing as much as possible!
- if Indigo Subdiv is used, try to set it to view dependend & Pixel treshold to 2
- Ultra High res EnvMaps eat up a lot of RAM, reduce size if not visible as BG
- Texture Resolution can be sometimes reduced. Image compression isn't interesting, Indigo always take the pixel data RAW and uses same amount of RAM for a TIFF or highly compressed JPG.
- try to use ISL shaders for noise, gradients etc. instead of HighRes textures

Render Buffer
- Super sampling raises the Buffer by render internally in higher resolution, try to reduce it!
- Light Layers: Each light Layer doubles the Render Buffer
- If rendering with Sun, take care to render Sun & Env Color in 1 Layer

Other
- use your renderslave with indigo_console to save a few MB


Basically the render Buffer is the biggest RAM killer!