GPU or CPU Prismatic binaries are packaged on conda-forge for Windows, MacOS and Linux. They can be installed easily in an Anaconda/Miniconda distribution using the conda or mamba package manager.

The different variants of prismatic are available in 4 packages:

  • prismatic_cli to install the command line version
  • prismatic_gui to install the GUI version
  • pyprismatic to install the python binding
  • prismatic to install all 3 others packages

Each of these packages are available with CPU or GPU capabilities (prismatic doesn’t support GPU on MacOS). To install one of the packages, the Anaconda navigator can be used. Alternatively, one of the following commands can be ran from an Anaconda command prompt (on Windows) or a terminal (MacOS or Linux):

conda install prismatic -c conda-forge

If a CUDA compatible GPU is available, the GPU packages will be selected by default and a cudatoolkit compatible with your GPU (Kepler and above) will be installed automatically. Otherwise, the CPU will be selected. To switch or explicitely select CPU packages, see instructions below.

The prismatic package is a meta-package which install the packages of each variant. If you don’t want to install all variants, you can install them separately.

To install prismatic CLI (prismatic-cli and prismatic-double for double precision):

conda install prismatic_cli -c conda-forge

To install prismatic GUI (prismatic-gui) with start menu shortcut on windows:

conda install prismatic_gui -c conda-forge

To install pyprismatic:

conda install pyprismatic -c conda-forge

Select CPU or GPU build

To select CPU only packages, use the following syntax with any of the conda packages mentioned above:

conda install prismatic=*=cpu* -c conda-forge

GPU packages are prioritise by default and it can happen that this priority is overwritten by another priority when conda or mamba resolves the dependencies tree. To pin to a specific computing backend (CPU or GPU), use the pinning mechanism of conda, either through:

conda config --env --add pinned_packages prismatic=*=gpu*

or by adding the package name and its pinning to the following to the env_path/conda-meta/pinned file:

prismatic=*=gpu*