Data displayed in the following table are available in csv format in the github repository of HEPiX-Forum

Starting with HEP Benchmark Suite version 3, sites can enrich benchmark results by including electric power consumption measurements collected during the benchmark execution.
For details on how to enable this, see the official script documentation.

Power consumption is measured using the command: ipmitool dcmi power reading

Important note:
The values reported by ipmitool may differ from those obtained via Power Distribution Units (PDUs).
Accuracy depends on the server’s hardware and firmware implementation. Therefore, these results should be considered experimental and approximate.

Metrics description

Benchmark result tables include several metrics.
Only those that are not immediately self-explanatory are described below:

• # Meas : The number of independent benchmark measurements performed.

• Score/Ncores : The performance score normalized by the number of available cores (physical or logical).

• sem (Standard Error of the Mean) : A measure of the statistical uncertainty of the score, calculated from multiple runs.

• Spread : A measure of variability across the benchmark runs, calculated as the relative difference between the 95th and 5th percentiles, normalized by the median.

• electric_power_mean , electric_power_std: For each measurement time series collected during the benchmark run, the 85th percentile is determined. Mean and std are then evaluated across all measurement series.

• HS23/W = Score / electric_power_mean