Skip to content
AnimationStudiosBLUX Physics SuiteGet BLUX
Discord

Solver environment

Open Edit > Preferences > Add-ons > Blux Physics Suite. The Solver box controls how the GPU worker is started.

At the top of the page, Refresh Status re-reads everything shown on the page: the kernel cache size, SciPy status, cached frames and folder checks. The page does not poll the disk on its own.

The Python interpreter that runs the solver worker. It must have the GPU runtime (Taichi) installed.

  • Leave it empty to let BLUX search: the BLUX_PYTHON_PATH environment variable, then PATH, then common install locations.
  • Blender’s own interpreter is never used, because the GPU runtime is not bundled with it.
  • The file must be named like a Python executable (python, pythonw, py, pypy, with an optional version and .exe). Anything else is rejected.

If the path you enter is not valid, a red message explains why and shows which interpreter BLUX keeps using instead. The invalid value is not applied.

Where the CUDA, Vulkan, CPU fallback chain starts.

Option Behaviour
Automatic (CUDA, then Vulkan, then CPU) Try CUDA first and fall back through Vulkan to CPU. Recommended. This is the default.
CUDA (NVIDIA) Start at CUDA. Falls back if no CUDA device or driver is available.
Vulkan (AMD / Intel / NVIDIA) Start at Vulkan. Use this on a non-NVIDIA GPU.
CPU only Run on the CPU. Much slower, but works everywhere. Useful for diagnosing a driver problem.

An explicit choice is honoured exactly: choosing CPU never silently upgrades to a GPU backend.

How long BLUX waits for the first simulation step, which compiles the GPU kernels. Default 3600 seconds, minimum 120.

A cold start of a large Blaze setup with the multigrid pressure solver measured 513 s to its first frame on a GTX 1060, which is why the default is generous. Later starts use the kernel cache and take far less. Raise it only if a very slow machine times out.

On by default. Compiles the scene’s GPU kernels in the background when a file is opened or the solver changes, so Run starts without waiting. Blender stays responsive meanwhile, and compiled kernels are kept on disk. See Kernel cache and warm-up.

Runs a check in the background that the chosen interpreter can run the BLUX solver. The result shows under the button:

  • OK with the interpreter path: Taichi imports and the solver can start.
  • An error message with the hint Install it with: pip install taichi: the interpreter was found but Taichi is missing or broken.

If a different working interpreter is found on your system, a Found: row appears with a Use It button that saves it as the Interpreter.

See also Troubleshooting: cannot connect to the worker.