This tool allows the user to execute a NURBS-based form-finding of shell structures. In addition to continuum shell form-finding, it provides an option to align stress and curvature directions (either conjugate or principal). This option strongly supports the design of grid shells.
The package also includes a user manual (pdf).
The example definitions are tested with Rhino 7, Grasshopper, Windows 11 Pro. The following two items are required to run the code.
Cuda-enabled Nvidia GPU
Cuda toolkit Library 12.x. (The user must install this library on their own)
Does the GPU version work in R8?
I unlocked the DLL and added the assembly to the C# component but still got many error messages.
They are in the same folder.
I think it is very likely you have downloaded an older version. Or you have downloaded an older version before and the C# nodes are referencing older KingOfMonster.dll.
ic, but you might want to remove the mac tag then.
i recently installed win 11 on my older intel macbook pro it has a NVIDIA GeForce GT 750M with 2GB of GDDR5 you think that could work? might not be the fastest but i believe it was CUDA capable, if i get to installing rhino there i might try.
If your GPU is Cuda-capable, the code should run theoretically.
In the script, you can find ‘RefineSrf’. Using this component, you can decrease the number of control points. I recommend at least 4GB VRAM, though.
Hi mikity! The results you show are super cool! Could you please share how this is different from just form finding with quad mesh, then converting to NURBS vis SUBD?
I’m also impressed that CUDA is used–would that mean some this form-finding process is parallelized?
Hi, if you use a quad mesh, the topology cannot change, or otherwise, a re-meshing algorithm is needed to change the topology. If you use NURBS, you don’t need to assume the topology of the mesh beforehand. After obtaining a NURSB surface as a solution, you can convert it to a mesh.
And yes, it’s using parallel computing a lot. But the Cuda part is actually not parallel. Because the Cuda toolkit library provides Lapack-ish APIs to solve linear equations, I’m simply using it to accelerate the numerical solution part.
