Performances

In this page we collect hints on how to use the features available in PLUMED to speed up your calculations. Please note that PLUMED performs many different tasks, it can calculate a number of different collective variables, functions of collective variables, bias, on-the-fly analysis, etc in a way that is compatible with a number of different molecular dynamics codes. This means that there cannot be a single strategy to speed up all the possible calculations.

Here you can find a step-by-step tutorial on optimizing PLUMED performances, discussing some of the topics below in more detail and using practical examples.

PLUMED makes use of MPI and OpenMP to parallelize some of its functions, try to always compile it with these features enabled. Furthermore, newer compilers with proper optimization flags can provide a dramatic boost to performances.

PLUMED collects atoms from an external code and sends back forces, so it is key to minimize the effect of PLUMED on highly parallel calculations to keep to the minimum the number of atoms used by PLUMED at every calculation step. The less is the number of atoms you need to send to PLUMED the less will be the overhead in the communication between PLUMED and the code.

In the following you can find specific strategies for specific calculations, these could help in taking the most by using PLUMED for your simulations.

• GROMACS and PLUMED with GPU
• Metadynamics
• Multiple time stepping
• Multicolvar
• Neighbor Lists
• OpenMP
• Secondary Structure
• Time your Input
• Making lepton library faster

Check also this tutorial: Optimizing PLUMED performance