Platforms | Wobbl

Platform	CalculiX	ngspice
macOS Apple Silicon	native	Intel build under Rosetta 2
macOS Intel	native	native
Linux x86_64	native	native
Linux arm64	native	not packaged, install it with your package manager and Wobbl finds it
Windows x64	native	native

The tests run on Ubuntu, macOS and Windows with Python 3.12 on every push, and on Ubuntu with Python 3.10 and 3.13 as well, including the end to end studies with the real solvers.

Apple Silicon

conda-forge builds ngspice for Intel Macs only, so wobbl setup installs that build into ~/.wobbl/env-x86 and macOS runs it through Rosetta 2. If Rosetta is not installed yet, setup’s own solver test fails and tells you to run this.

Terminal

$ softwareupdate --install-rosetta --agree-to-license

CalculiX is a native arm64 build, so strength and heat studies run at full speed.

Linux arm64

conda-forge has no ngspice for this platform. Strength and heat work as usual, and for circuits you install ngspice yourself, after which Wobbl finds it on your PATH.

Terminal

$ sudo apt install ngspice

If it lives somewhere unusual, set WOBBL_NGSPICE to the full path of the binary.

Headless Linux

Wobbl draws its heatmaps with VTK off screen. On a server with no display, VTK needs a few system libraries for OpenGL and fonts, and the Ubuntu test machines install these.

Terminal

$ sudo apt-get install -y libglu1-mesa libxrender1 libxcursor1 libxft2 libxinerama1 libgl1 xvfb

When VTK cannot render at all, Wobbl falls back to matplotlib for the images and keeps going, and wobbl doctor says which renderer is in use.

Speed

CalculiX runs on up to four threads. The time budget of a strength study is spent by a cost model calibrated on a 2018 six core laptop with an Intel i7-8750H, so a faster machine reaches the same mesh levels and simply finishes sooner. A strength study on a real part takes one to seven minutes on that laptop, and simple parts take well under a minute.