OSPRay Multi Node Question

Hey all, looking at the updated 5.9 release notes I noted this paragraph under OSPRay updates

image

Does these mean we can now run ParaView with OSPRay sessions in parallel with multiple nodes? I tried loading these env variables and running that and got a bunch of errors, and wanted to see exactly what that meant before doing a bunch of debugging.

I was under the impression after talking to @Dave_DeMarle last year that multi-node raytracing was not possible due to problems sharing lighting information across nodes. At least when we talked, it was an industry wide problem that had not been solved. Has this changed?

When I have tried it in the past, it seems that each node would take a subset of the dataset, and apply the raytracing to each subset, but not be able to combine them back together.

If this has been solved, that would be HUGE and definitely speed up render times.

Thanks,

Unfortunately the offload mode mentioned in the release notes does not support data that does not fit on memory in a single node. With offload mode the visible data is broadcast to multiple nodes in a cluster so that all of them have a copy of the entire visible scene. Then each node ray traces only a portion of the pixels instead of the whole image. With higher quality graphics settings, for example when the number of samples per pixel or ambient occlusion queries are large, or the scene has things like area lights (ie soft shadows) and depth of field (camera focus) turned on, splitting up the work this way tends to speed up the rendering process more or less linearly with the number of nodes working on it.

Hey Dave, that makes more sense. So if my dataset can fit on a single node’s memory (it usually can), this should be able to speed up my headless renderings? I’m guessing if I run on 4 nodes, it will be slightly less than 1/4 the render time, but 4 times the total memory cost?

Sounds like a great feature, and I need to get this working on my system.

Appreciate you taking the time to answer.