BlendLuxCore Development

[juangea]

What versions do support that and what version don’t?

Also could be great to be able to enable/disable that export if it slows down a lot the export process

[B.Y.O.B.]

What versions do support that and what version don’t?

Currently, fast export is supported for 2.82.7 and 2.83.*
This can be checked here: https://github.com/LuxCoreRender/BlendL ... rter.py#L9

Also could be great to be able to enable/disable that export if it slows down a lot the export process

There is a warning if custom normals increase the export time of a mesh by more than 0.3 seconds. In such a case, you can disable usage of custom normals from the mesh in Blender, e.g. by deleting them or by disabling the weighted normal modifier or whatever else is creating them.

[juangea]

Ah, ok, si I assume that refers to the normals modifiers and the baked normals into the mesh, but the normal smoothing with angle works ok and does not affect, right?

[B.Y.O.B.]

Ah, ok, si I assume that refers to the normals modifiers and the baked normals into the mesh, but the normal smoothing with angle works ok and does not affect, right?

Yes, I mean the feature that is labeled "custom split normals" here: https://docs.blender.org/manual/en/late ... it-normals

[juangea]

Yep, that was what I was thinking about, perfect, then it's not a big deal, at least for us, since usually that kind of smoothing detail is done in some very specific models

[B.Y.O.B.]

I have extended the FAQ with a few new topics: https://wiki.luxcorerender.org/BlendLuxCore_FAQ

1.2 Is there something like Cycles' Layer Weight/Fresnel/Facing node in LuxCore?
1.3 Can Cycles materials be used in BlendLuxCore?
1.4 Can 2D/3D mapping be driven by textures in LuxCore?
1.5 What can be done against fireflies/hotpixels?

If you can think of more topics that come up often and should be included there, let me know.

[Dade]

I was recently trying to found a work around to CUDA/hardware texture map limit of supporting only 1, 2 or 4 channels texture maps and I looked at Cycles sources to check how they were solving this problem: apparently they don't.

I'm guessing Cycles stores RGB (i.e. 3 channels images) as 4 channels, wasting a 33% of GPU memory. Can someone confirm this ? It shouldn't be hard to test by using a very large texture map with 3 or 4 channels and checking the amount of GPU memory used: if the amount is the same, my theory is confirmed.

[TAO]

Not sure about that but OpenColorIO color space can be a good solution, and as I can see many renderers and software support it natively.

[B.Y.O.B.]

I'm guessing Cycles stores RGB (i.e. 3 channels images) as 4 channels, wasting a 33% of GPU memory. Can someone confirm this ? It shouldn't be hard to test by using a very large texture map with 3 or 4 channels and checking the amount of GPU memory used: if the amount is the same, my theory is confirmed.

I did a test with the following scene:
Render on GPU only (CUDA) with Cycles.
Three versions of a 16000x16000 pixel image: 1 channel (black/white), 3 channels (RGB), 4 channels (RGBA).
Output resolution 100x100.
I have no precise measurements of the used GPU RAM, but the task manager graph shows that your suspicion is probably right, as there's no difference in memory usage between 3 and 4 channel images.

[TAO]

Maybe if you try the same test in a bigger image size the difference can be more obvious.
The memory usage for the compare part is so small that hardly can be noticed.