Reducing object load and voxelization times

[brue]

Hey Guys,
Currently I have a scene that takes in total 8 minutes to load into octane maya. 5 1/2 minutes for the object sort and another 2 1/2 minutes for the voxelization. Any tips on reducing this time ?

[brue]

One idea that came up would be the ability to cache the object load in order to not take the hit every time...

[glimpse]

=) hi there,

what CPU do You have? Some say, that disableing HT helps to reduce time to 50%, 'cos if HT is on You're running only 1thread what is effectivelly half core.
How much RAM do You have?..Maybe Your system starting to move info from RAM to SSD, even worse to HDD..

Anyway, these are just few thoughts to start with.

& Plz, put Your PC info to Your signature.

Cheers

[brue]

Thanks for the HT tip, I'll try that. The majority of the time appears to be in the object sort as it's loaded into memory. I haven't played with octane stand alone curious if this is the case of the maya parser, and what could be done to multithread this, or what the bottleneck is.

[brue]

The scene went from 8 minutes 17 to 7 minutes 3, without hyperthreading.

[glimpse]

not so much of improvement =( less than 10%..

regarding to Your system You clearly don't have any shortage of RAM.

how much the scene weights? what is the object? some sculpt?

[brue]

any ideas as to why the object sort and voxelization process is single threaded ? The scene is just large with about 5 million polygons but a lot of unique objects. It fits in gpu memory its just the build process to load into gpu memory is quite long and not threaded.

[face]

The transfer to the GPU is only single threaded, that´s the nature of the render engine.
But so much i know, Otoy has a plan to make it multithreaded...

face

[TBFX]

Hey Guys,
Currently I have a scene that takes in total 8 minutes to load into octane maya. 5 1/2 minutes for the object sort and another 2 1/2 minutes for the voxelization. Any tips on reducing this time ?

Hi brue,

Depending on the size and complexity (poly count, texture size/count) of the scene this sort of transfer/voxelization time is mot unheard of though with the integrated plugin it would take a very complex scene I think. The standalone is generally slower to load as it has to read and interperate the OBJ file where as the plugins already know where all the geo is in the host application.

In the Octane render globals there is an Animation Mode dropdown that lets you select Full, Camera only or Movable proxies. If nothing is animated in your scene other than the camera you can set this to Camera only and then it will only perform the transfer/voxelization once on the first frame rendered. If objects are animated then set the Octane: Geometry type attribute on their shape node to Movable proxy and use the Movable proxies animation mode, this will only reload the animated meshes. Be aware though that once you have movable proxies in the scene you will get a drop in render speed but this may still give you an overall faster combined frame render time.

T.

[brue]

Yep know about the animatable proxies for animation. Its primarily for the larger film scenes that I need to deal with. Trying to look into the reality of using octane as a production renderer for an onset workflow. An 8 minute load time for a standard scene is rather heavy, especially when the process could in theory be sped up by an order of magnitude by threading.