The increase in large science focused computational frameworks has raised many issues involving the ability to maintain accurate scientific benchmarks throughout the ongoing evolution of the code. These science based tests allow not only developers access to the latest updates, but the science users as well. It is these scientific tests required for geodynamic code benchmarking in a HPC environment that are investigated. The importance of benchmarking in computational science, for both quality assurance and reliability, is discussed and a case study for thermochemical convection modelling is presented. The implementation of automated testing for science units is described with particular attention to the problems arising from science tests compared to traditional computational tests.