Exploring the use of mixed precision in NEMO
Document typeConference report
PublisherBarcelona Supercomputing Center
Rights accessOpen Access
It has been a widely extended practice in scientific computing to use 64-bit to represent data without even considering which level of precision is really needed. In many applications, 32-bit should provide enough accuracy, and in other cases 64-bit is not enough. In climate science, the inherent difficulties collecting data imply a considerable level of uncertainty, which suggest that the general use of 64-bit to represent the data may be a waste of resources, while on the other hand, some specific algorithms could benefit from an increment of the precision used. These factors suggest that in the future more attention has to be paid to the precision used in scientific software, to use the resources wisely and also avoid losing accuracy. In this work we question whether the precision used in the oceanic model NEMO is necessary and sufficient, and the potential benefits of adjusting this precision.
CitationTintó Prims, O.; Castrillo, M. Exploring the use of mixed precision in NEMO. A: BSC Severo Ochoa International Doctoral Symposium (4th: 2017: Barcelona). "Book of abstracts". Barcelona: Barcelona Supercomputing Center, 2017, p. 52-54.