View on GitHub

Graphitti

A project to facilitate construction of high-performance neural simulations

1.2 Installation

1.2.1 Necessary Hardware/Software

Graphitti is designed to be easy to use and fast to simulate with, but given its scope and flexibility, there are some tradeoffs.

First, and perhaps most importantly, for the speedups that we desire, we found that CUDA was the most reasonable way to go. Hence,  if you want to use Graphitti for migrating your model to GPUs, you will need the following:

To become a Graphitti user or collaborator, you might also need:

Of course, Graphitti is totally open source. If you wanted, you could modify Graphitti and make an OpenCL version.

1.2.2 Download Graphitti

In order to get started with Graphitti, you will need to build it from scratch, which means getting its source codes. You can either download Graphitti source codes as a zip file of a stable release (See the release page) or fork the development version from Graphitti GitHub repository (See Fork and clone Graphitti).

1.2.2.1 Fork and clone Graphitti

If you are a Github user, you can simply fork and clone Graphitti. If you are new to Github, follow our Wiki page on Contribute to Graphitti open source project. You can also go over our Git Crash Course for some useful tips.

1.2.3 Install Graphitti

In order to compile and run Graphitti, you will need to set up a couple things in the CMakeLists.txt first.

  1. Change to Graphitti directory in your terminal

    $ cd Graphitti/build
    
  2. Open CMakeLists.txt and change the following parameters:

    If you are using CUDA,

    • change the CUDA library directory: ENABLE_CUDA YES
    • you might also need to add your CUDA home directory into the PATH environment variable
  3. Graphitti is written in C++11 and CUDA C/C++. Make sure you have all these dependencies in order to compile Graphitti:

    • make
    • g++
    • h5c++: compile script for HDF5 C++ programs
    • nvcc: if you are using GPU for high performance, nvcc is the compiler by Nvidia for use with CUDA

» Next: 1.3 Quickstart


« Go back to User Documentation page


« Go back to Graphitti home page