Install NEST GPU

Requirements

To build NEST GPU you need CMake (version 3.17 or higher). You also need the NVIDIA drivers for the GPU card installed in your machine and the NVIDIA CUDA development toolkit. To use the NEST GPU Python interface you need Python 3, Numpy, Scipy, Matplotlib and mpi4py for using MPI from Python.

We also recommend to install the OpenMP binary and development packages, which are needed for using MPI.

Note

Currently NEST GPU supports only NVIDIA GPUs and can only be installed on linux systems.

Installation instructions

The following instructions describe the steps to compile and install NEST GPU from source code. To adjust settings please see CMake Options instructions.

You can install the mandatory and recommended packages (see Requirements section) as follows:

sudo apt update

sudo apt install -y \
nvidia-cuda-toolkit \
openmpi-bin \
libopenmpi-dev \
libomp-dev \
python3-all-dev \
python3-numpy \
python3-scipy \
python3-matplotlib \
python3-mpi4py

If you are using Ubuntu, here you can find a guide to install the NVIDIA drivers for the GPU card on your machine.

Download the library (see Download section) and unpack the tarball (if needed):

tar -xzvf nest-gpu-x.tar.gz

Create a build directory:

mkdir nest-gpu-x-build

Change to the build directory:

cd nest-gpu-x-build

Configure NEST GPU. For additional cmake options see the CMake Options of this docuentation.

Without the additional options you can type:

cmake -DCMAKE_INSTALL_PREFIX:PATH=<nestgpu_install_dir> <nestgpu_source_dir>

Note

nestgpu_install_dir should be an absolute path.

Compile and install NEST GPU:

make
make install

NEST GPU should now be successfully installed on your system.

Environment variables

To specify where NEST GPU installation is located you have to use some environment variables. For your convenience, a shell script setting all required environment variables is provided in <nestgpu_install_dir>/bin/nestgpu_vars.sh. Setting the environment variables in your active shell session requires sourcing the script:

source <nestgpu_install_dir>/bin/nestgpu_vars.sh

You may want to include this line in your .bashrc file, so that the environment variables are set automatically whenever you open a new terminal.

The following variables are set in nestgpu_vars.sh:

Variable	Description
`PYTHONPATH`	Search path for non-standard Python module locations. Will be newly set or prepended to the already existing variable if it is already set.
`PATH`	Search path for binaries. Will be newly set or prepended to the already existing variable if it is already set.

If Python does not find the nestgpu module, your path variables may not be set correctly. This may also be the case if Python cannot load the nestgpu module due to missing or incompatible libraries.

Installation tests

To check the correctness of NEST GPU installation you can find some tests in the directory <nestgpu_source_dir>/python/test, where the <nestgpu_source_dir> is the install path given to cmake. Each Python script tests a specific feature of the library, and to perform all the tests you can run the bash scripts test_all.sh (which runs all the MPI tests that do not employ MPI) and test_mpi.sh. If everything worked well, for every test you should see a line indicating TEST PASSED or MPI TEST PASSED.

If some test did not pass, you can have a look at the log.txt file given in output by the bash scripts to see the output of the Python tests.