Device Communication Efficiency

The Device Communication Efficiency (DevCommE) is defined as the ratio between the maximum useful computation time across all accelerator devices and the maximum (also across all accelerator devices) combined useful computation time and time waiting for data.

\[DeviceCommE = \frac{Max(DeviceUseful)}{Max(DeviceUseful + DeviceWaitForData)}\]

This metric represents inefficiency due to data movement, which is impacted by waiting for data from the host, sending data to the host, and transferring data between accelerators.

In order to fully understand the formulas, you may also visit the glossary of the metrics terms.

Related patterns: Avoidable transfers between host and GPU for MPI communication (GPU-Unaware MPI) ·