CP2K biomolecular QM/MM benchmarking data

Dataset containing raw CP2K output logs produced by running benchmarks from the BioExcel QM/MM benchmark suite (https://doi.org/10.5281/zenodo.6591692), provided as part of BioExcel-2 project deliverable D1.6. 

An analysis of (a subset of) these benchmark results can be found in https://doi.org/10.5281/zenodo.6591574 

Machines used for benchmarking:

Cirrus@EPCC (HPE SGI ICE XA):

https://www.cirrus.ac.uk

• Infiniband interconnect
• CPU compute nodes:
  • 2 x 18-core Intel Xeon (Broadwell) E5-2695, 2.1 GHz
  • 256GB RAM
• GPU compute nodes:
  • 2 x NVIDIA Tesla V100 (Volta) SXM2-16GB
  • 2 x 20-core Intel Xeon (Cascade Lake) Gold 6248, 2.4 GHz
  • 384GB RAM

ARCHER2@EPCC (HPE CRAY EX):

https://www.archer2.ac.uk/

• Interconnect: HPE Cray Slingshot
• Compute nodes:
  • 2 x 64-core AMD EPYC (Zen2 Rome) 7742, 2.25GHz
  • 256GB RAM

Benchmarking protocol

Repeated runs to rule out machine noise variability were performed for each benchmark on each machine for both 1 MD step and 6 MD steps. Subsequent analysis and visualisation of parallel scaling of average runtime per MD step and subroutine-level profiling was performed using our analysis script available from:

https://doi.org/10.5281/zenodo.6591681

Results on Cirrus CPU nodes are for CP2K release version 8.1, whilst results on Cirrus GPU nodes and ARCHER2 are with CP2K version 8.2.