Has anyone here shipped parts off an implicit/SDF pipeline end-to-end, not just for visualization? I’m prototyping a JAX-compiled SDF kernel that builds a about 420k-cell TPMS lattice and writes STEP AP242 in about 6 minutes on an RTX 4090, but Fusion 360’s CAM chokes on the resulting tessellation. Before I invest in a USD + GPU mesher or jump straight to 3MF/AMF for additive, I’d love to hear what’s worked for you in production.
I’d skip STEP for the TPMS and do an adaptive isosurface + error-bounded decimation, then write 3MF straight to the slicer; OpenVDB’s VolumeToMesh is a solid starting point: GitHub - AcademySoftwareFoundation/openvdb: OpenVDB - Sparse volume data structure and tools. If you must stay in Fusion, import as a mesh body and use Manufacturing Model > Reduce with a fixed mm tolerance so CAM stops “choking” on the triangle count. Is this purely additive, or do you also need post-print machining features?
For additive, bypass the heavy mesh/B-Rep and write 3MF “Slices” straight from the SDF: do GPU marching squares per Z and dump polylines into the slice stack (spec: https://github.com/3MFConsortium/spec_slices), or fall back to Materialise.cli if needed. It’s kept our TPMS jobs small and avoids Fusion’s mesh limits — think mailing pancakes instead of the whole griddle. What slicer/printer are you targeting?