Please note that this package is still under developement and has not been released yet.
Brian2CUDA is an extention of the spiking neural network simulator Brian2, implementing a Brian2 standalone device to generate C++/CUDA code to run simluations on NVIDIA general purpose graphics processing units (GPGPUs).
Use your Brian2 code (see Brian2 documentation) and modify the imports to:
from brian2 import * import brian2cuda set_device("cuda_standalone")
- Python2: The current version only supports Python2. For the first release, we will support (only) Python3 (see issue #161)
- Brian2: The correct version with which this implementation is working is stored in a submodule in
- Matplotlib, Seaborn (these dependencies will be removed for the release).
After cloning brian2cuda, you can initialise the submodule from inside this repository:
cd brian2cuda git submodule update --init frozen_repos/brian2
Now you can install the correct Brian2 and brian2cuda versions using pip (Be careful if you already have a Brian2 version installed in your current Python environment!):
pip install . pip install ./frozen_repos/brian2
Or just add them to your
PYTHONPATH (in which case you need to install their dependencies manually).
Comparison with brian2genn
git submodule update --init frozen_repos/brian2genn git submodule update --init frozen_repos/genn
If you are using LINUX and your CUDA is installed in
/usr/local/cuda/, you can now just source
init_genn to set the enironmental variables needed for GeNN to work:
Otherwise modify the
init_genn accordingly or follow the instructions from the GeNN repository
Now you can install
brian2genn either with pip:
pip install ./frozen_repos/brian2genn
or just add it to your