Traditional metrics of computational efficiency such as performance
counters and scaling curves do not tell us enough about real sustained
performance from climate models. They also do not provide a basis for
comparative information across models.
In this talk we introduce a set of metrics that can be used for the
study of computational performance of Earth System models. These
measures do not require specialized software or specific hardware
counters, and should be accessible to anyone. They are independent of
platform, and underlying parallel programming models.
We present results here for these measures for a diverse suite of
models from GFDL, and propose to use these measures as a basis for a
CPMIP, a computational performance MIP.