Page 1 of 1
Voxelisation time
Posted: Thu Jun 13, 2013 3:09 pm
by Otuama
I'm thinking about thinking of writing a quick tool that gives users an approx time a maxwell preview will take based on the max frame time setting & timeline.
I've been told that the voxelisation time is 5 secs but does that depend on the size etc of the scene. For example, do heavy scenes with many objects, high res meshes & varying materials have a longer voxelisation time?
Also, does the voxelisation happen each frame?
Re: Voxelisation time
Posted: Thu Jun 13, 2013 3:37 pm
by enrique
Voxelization time depends on the geometry of the object (the number of faces, and the way the faces are in the space affect that), the number of objects, the number of particles. Voxelization is just one of the steps. Rendering time depends as well on the number of files to be loaded, the size of the files, the number of channels to be loaded, the possible compression of those files, the hard disk (is there any other application trying to write or read stuff), the status of the CPU (any other application thirsty for some cycles?), the materials of your scene (take into account the extra geometry generated in rendering time with displacement, SSS material...), motion blur affects as well and changes with time. Consider as well that sometimes the render ends because you have reached the maximum number of seconds of the render and sometimes it happens because the maximum sampling level has been reached. Many more things have to be calculated in each render.
So forget about those 5 secs
Voxelization happens everytime something relevant to the render changes, so it will happen in every frame while doing a preview.
Maybe the naïve approach would be the best in this case: let's say we have a 100-frame animation to render and the first 10 frames took us 15 mins, we would assume that we'll still have to wait another 135 minutes (15/10*90). Do this in each frame.
Re: Voxelisation time
Posted: Thu Jun 13, 2013 4:06 pm
by Otuama
Thanks for the in-depth reply!
It'd be great if something like this could be brought into Rf somehow but there isn't much point it being done unless accurate-ish results were given.
Your method of rendering 10 frames first to gain a more accurate estimation is good. However, as python scripts don't run during the preview process, it wouldn't be possible.
That is unless the script itself runs the preview.