In weather and climate models values of relevant physical parameters are often uncertain by more than 100%. Still, numerical operations are typically calculated in double precision with 15 significant decimal digits.If we reduce numerical precision, we can reduce power consumption and increase computational performance significantly. If savings in computing power are reinvested, this will allow an increase in resolution in weather and climate models and an improvement of weather and climate predictions.
I will discuss different approaches to reduce numerical precision beyond single precision in HPC and in particular in weather and climate modelling (including FPGAs, half precision GPUs and reduced precision CPUs). I will present results that show that precision can be reduced significantly in Earth System modelling with no decrease in model quality and that potential savings are huge (including simulations with a specral dynamical core and a cloud resolving limited area model). Finally, I will discuss how to reduce precision in weather and climate models most efficiently, how rounding errors will impact on model dynamics and predictability, and I will also outline implications for data assimilation and data storage.