This directory contains performance benchmarks for the DFT-FE code on various medium and large-scale benchmark systems. The performance of the benchmark systems will be tracked as the architectures evolve (upcoming exascale machines) and methodology and implementation enhancements in future DFT-FE versions while maintaining the same accuracy. Additional performance benchmarks will also be added in the future as new capabilites are added into DFT-FE for example hybrid exchange-correlation functionals, time-dependent DFT, enriched finite-elements for large-scale all-electron calculations etc. The input and output files for all performance benchmarks are provided in the corresponding folders. The file structure from the top-down follows the following order DFT-FE version-> Architecture -> Type of electronic-stucture calculation -> Nature of performance benchmarks -> different material systems.

All benchmark calculations are run using ONCV pseudopotentials and finite-element discretization parameters are commensurate with chemical accuracy (~1e-4 Ha/atom in ground-state energy, ~1e-4 Ha/Bohr in ionic forces and ~5e-6 Ha/Bohr^3 in cell stresses).

i. OLCF Summit: 6 NVIDIA V100 GPUs and 2 IBM POWER9 CPUs (42 cores total) per node

Strong parallel scaling of wall-time per SCF iteration step on OLCF Summit GPU nodes. Please refer to DFT-FE 1.0 for details on the breakdown of the wall-time into various computational steps.

BCC Mo6x6x6 monvacancy (431 atoms, 6,034 electrons---Gamma point)	BCC Mo8x8x8 monvacancy (1,023 atoms, 14,322 electrons---Gamma point)

Minimum total run wall-time including initialization costs and computation of ionic forces achieved on OLCF Summit GPU nodes at a parallel scaling efficiency of around 40 %