User Tools

Site Tools


build:minreq

This is an old revision of the document!


External dependencies

Minimum requirements

Hardware

Abinit requires at least 1 CPU, 4Gb of RAM and 4Gb of disk space to perform the calculations of the tutorials and the test suite. The most complex calculations may require more than 10,000 CPUs, more than 1Tb of RAM, and a few Tb of disk space.

Any recent desktop or laptop computer is in principle able to perform calculations with Abinit.

Compilers

To be correctly built, Abinit requires a working development environment providing both a C and a Fortran compiler.

On desktop and laptop computers, we recommend:

Libraries

Abinit will work only if linear algebra libraries are present on the computer. Only 2 linear algebra components are mandatory:

Even if you are free to install the reference implementation from Netlib, we highly recommend you to use an optimized implementation instead, in particular if you intend to perform complex calculations. A full list of supported libraries is available on the Configuring linear algebra page.

If you are using Linux, the best way to install linear algebra libraries is to use the package manager of your distribution to get some of them and/or install the Intel Math Kernel Library.

For MacOS X, many libraries can be installed using Macports or Homebrew.

The University of Tennessee has also some useful information on how to install linear algebra libraries on Windows.

Parallelism

OpenMP and multithreading

If your computer has multiple cores, you might want to take benefit from all of them instead of running Abinit on only one of them. There are 2 ways of achieving it: OpenMP and Multithreading.

Various parts of Abinit have been optimized to work with OpenMP. In addition, the latter is usually directly provided by compilers and does not require the installation of additional libraries. The build system of Abinit is able to set the corresponding compile flags automatically for a dozen popular compilers if you ask for OpenMP at configure time. Should this not be the case for you, please consult the documentation of your compilers and contact us once you have found and tried the flags to provide.

Abinit does not provide support for multithreading but can benefit from multithreaded versions of its external dependencies, as long as the multithreaded operatons are entirely performed externally. Please note that Abinit is not developed with thread-safety in mind and use multithreading with a lot of care.

MPI

GPU

Enhanced components

Linear algebra

FFT

build/minreq.1461129817.txt.gz · Last modified: 2016/04/20 07:23 by Yann Pouillon