CASTEP

Benchmarks

Introduction

As well as ensuring your CASTEP build is correct, it can be useful to see how CASTEP performs on your computer. To this end, there are a number of benchmarks you can use to compare performance between different machines.

We have categorised the benchmarks as "small" (workstation-class), "medium" (cluster class) or "large" (large HPC-class) depending on the computational requirements. As faster computers become available, these classifications may change!

Small

Small benchmarks may be run on most machines, from desktop workstations (or even laptops) up to large clusters. Their memory requirements are quite low, and they will usually run quickly, but may not scale very well to large process counts simply because the amount of computational work required is not very large, so any time spent in communication becomes a problem.

Small benchmarks typically have only a few tens of atoms in a small simulation volume, and often have many k-points.

Medium

Medium benchmarks are designed to run quite quickly on a modest parallel computer, such as a small cluster, though they may run on workstations if they have plenty of RAM.

Medium benchmarks typically have a few hundred atoms and a small number of k-points. They should scale fairly well to a hundred processes or so, depending on the interconnect between the compute nodes.

  • al3x3 is a 270-atom sapphire surface, with a vacuum gap. There are only 2 k-points, so it is a good test of the performance of CASTEP's other parallelisation strategies.

Large

Large benchmarks are designed for use on large high-performance computers. They will typically have many atoms and a large simulation volume, and consequently very few k-points.

  • crambin is a 1280-atom simulation of a protein residue. There is only 1 k-point (the gamma-point), so its parallel performance is reliant on CASTEP's other parallelisation strategies.
  • DNA is a 1356-atom simulation of a DNA strand (poly-A) with counter-ions, in a large simulation box. There is only 1 k-point (the gamma-point), so, like the Crambin test, its parallel performance is reliant on CASTEP's other parallelisation strategies.