We have recently extended the libtensor library for tensor contractions to enable single precision operations. The initial benchmarks have illustrated that double precision is not necessary in most correlated calculations. Using single-precision results in a factor of 2 speedup and reduced memory footprint, while introducing negligible errors in energies and gradients. We will present recent results on further applications of single- and mixed-precision calculations as well as performance gains on various architectures.

Further explorations of single-precision algorithms within coupled-cluster and equation-of-motion coupled-cluster framework

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.