Hi Michael,

Was there a speed difference that you noticed between the cgal and
manifold rendering? I'm not sure what happens inside openscad, but cgal
results are typically cleaner than manifold outside of openscad, but
there could also be additional cleanup in openscad code just for cgal.
Manifold calculations are 64 bit, but unless meshgl64, is used, vertices
are stored as 32bit. For complex geometry (many booleans, for example),
slivers and other degenerate cases can be generated.

Best wishes,

Ray

On 28/01/2026 04:37, Michael Marx (spintel) via Discuss wrote:

From: Raymond West via Discuss [mailto:discuss@lists.openscad.org]

Was there a speed difference that you noticed between the cgal and manifold rendering? I'm not sure what happens inside openscad, but cgal results are typically cleaner than manifold outside of openscad, but there could also be additional cleanup in openscad code just for cgal.
Manifold calculations are 64 bit, but unless meshgl64, is used, vertices are stored as 32bit. For complex geometry (many booleans, for example), slivers and other degenerate cases can be generated.

Best wishes,

Ray

On 28/01/2026 04:37, Michael Marx (spintel) via Discuss wrote:

Can Manifold rendering be tweaked?

It seems to like very long thin triangles, which historically have been problematic with floating point inaccuracies. IIRC

Via Manifold:

Can you try the following:

Export as 3mf
Import into a program using Manifold... somehow? Unfortunately
https://www.manifoldcad.org does not easily let you add files.
https://manifoldcad.org/docs/jsuser/functions/importManifold.html says
there is an import, maybe use some JS fetch to download it from a URL?
Alternatively, https://github.com/elalish/manifold says something about
a Python notebook?
See if it shows the same.

If it's the same or similar, can you call Simplify() on it and see if it
helps?

On 1/27/26 8:37 PM, Michael Marx (spintel) via Discuss wrote:

Oops...

Yes, I was surprised as there was a minkowski() over the whole thing, it finished much quicker than the ~40 minutes for CGAL.

I recall reading minkowski() was still CGAL. (?)

Both renders included degenerate faces, CGAL had a couple of self-intersections.
But Manifold had holes (ie 'non-manifold') and lots of self intersections:

From: Raymond West via Discuss [mailto:discuss@lists.openscad.org]

Was there a speed difference that you noticed between the cgal and manifold rendering? I'm not sure what happens inside openscad, but cgal results are typically cleaner than manifold outside of openscad, but there could also be additional cleanup in openscad code just for cgal.
Manifold calculations are 64 bit, but unless meshgl64, is used, vertices are stored as 32bit. For complex geometry (many booleans, for example), slivers and other degenerate cases can be generated.

Best wishes,

Ray

On 28/01/2026 04:37, Michael Marx (spintel) via Discuss wrote:

Can Manifold rendering be tweaked?

It seems to like very long thin triangles, which historically have been problematic with floating point inaccuracies. IIRC

Via Manifold:

Export as 3mf, loaded into MeshLab.

Long thin triangles, self-intersections (in orange), no holes (as was in the STL).

I don't think so?

From: Cory Cross via Discuss [mailto:discuss@lists.openscad.org]
Sent: Thursday, January 29, 2026 2:11 PM
To: Michael Marx (spintel) via Discuss
Cc: Cory Cross
Subject: [OpenSCAD] Re: Manifold - long thin triangles

Can you try the following:

Export as 3mf
Import into a program using Manifold... somehow? Unfortunately https://www.manifoldcad.org does not easily let you add files. https://manifoldcad.org/docs/jsuser/functions/importManifold.html says there is an import, maybe use some JS fetch to download it from a URL? Alternatively, https://github.com/elalish/manifold says something about a Python notebook?
See if it shows the same.

If it's the same or similar, can you call Simplify() on it and see if it helps?

On 1/27/26 8:37 PM, Michael Marx (spintel) via Discuss wrote:

Can Manifold rendering be tweaked?

It seems to like very long thin triangles, which historically have been problematic with floating point inaccuracies. IIRC

Via Manifold:

Via CGAL:

OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org

Would you be willing to send me the 3mf? I can try to load it in
Manifold https://github.com/elalish/manifold (not MeshLab) because
I'm wondering if Manifold's Simplify transformation will clean it up. If
so that's something easy for OpenSCAD to do.

• Cory

On 1/28/26 7:59 PM, Michael Marx (spintel) via Discuss wrote:

I've got a relatively complex (in terms of parts, at least for me) model
of a PNP assembly.  Using CGAL, rotating takes about 1/10 to 1/5 second
between positions.  Using Manifold, it rotates smoothly.

Harvey

On 1/28/2026 10:03 PM, Michael Marx (spintel) via Discuss wrote:

PrusaSlicer also has a Simplify Model feature.

On Wed, Jan 28, 2026 at 9:59 PM Michael Marx (spintel) via Discuss <
discuss@lists.openscad.org> wrote: