Heavy scene with Instances and MDD optimisation - How ? Tweet

[JPhi]

Hi all,

I have questions about how optimize heavy lightwave scene with several (lot of ?) different instances drived with different MDD.

Is it better to have lot of small objects (an small MDD), or is it better to have less objects (and less MDD) but bigger objects (and bigger MDD) to instance ?

This to solve the 2 following problems :

- First : how to use as low as possible amount of cpu ram when converting lw scene to octane scene (the fisrt step that take place when doing F10). Ram usage can raise so far than it become impossible to complete. (A 16, 20 Go ram scene (windows and lightwave included) can be impossible to convert with 96 Go of ram.

- Second : how to optimize loading time in GPU VRam (the second step after scene convertion and before rendering begins, and the same between frames to render). This loading time can become enormous and not usable in production.


This is for a production very near from its dead line.


Thanx.

[juanjgon]

Hi,

Do you have a lot of small objects, instances or fibers in near the same place in the scene? This problem sounds to me like a scene pre-processing issue related to having a lot of geometry in near the same place, in a small zone in the scene. Build the scene acceleration structures can take a lot of time and memory in this case.

If you enable the log file, you could see more information about the scene extraction, preprocessing and render times. If you attach it here, I could take a look at it.

Thanks,
-Juanjo

[JPhi]

Hi juanjgon,

Thanx for your fast support, as usual.

This scene is a big crowd in town.
MDD have to be used in local space. So every object (character) which have to be instanced is built at the origin in modeler.
That means that about all those objects are in the same place in layout before instancing (lot of overlapping here). Instances does not overlap but are most of the time very near to each other.
More than that, because of lack of instances of instance in lw, characters are cut in several pieces (each one with its own MDD) for creating more variety when instanced. And yes, those pieces nearly overlap too.
Not mentioning the need of offseting in time several time every animation, so this becomes very very heavy. (I know, lightwave isn't great tool for this job, but I had to use it. No choice in this production).

Could it help if I move null parents of those objetcs so they are far away from each other in world space and correct this for instances whith instances offsets ? (I can't change position in local space because I need MDD to work in local space).

I should be able to give you a log file tomorrow.

Thanx again.

[juanjgon]

Move the null parents of those objects is not going to help, I think. The problem is all the massive instanced geometry near the same place. The pre-processing time (and the memory needed) for this kind of scenes can be really high.

Thanks,
-Juanjo

[promity]

You can load your big objects separately and convert each one in to Octane materials. Than collect it in final scene.
I had a fairly large project with MDD - 10 models of a sports device from the Solidworker and 10 people who work on this device - each doing the same set of movements, but shifted in time through the MDD. To the movements of people the movements of the device are synchronized - in the same way they are shifted along the time scale. If I did not use instances, then the scene would be about 20 million polygons, which is quite a lot for an animation project.

[JPhi]

Hi Promity,

What do you mean by convert each object in to octane material, then collect it in final scene ?
Creating an orbx file for each object and then import them in octane stand alone and create final project in octane stand alone ?
Or is this a fonctionnality into lightwave I don't know about ?


Hi Juangjon,

I couldn't give you a log file yet, because we are solving now some crash that happens on our bigger octane render station.
I will give it to you as soon as possible.

Thanks for everyone's help. Very appreciated.
I like this forum.