Statistical convergence in strong lens modeling on multiple GPU nodes

Strong gravitational lenses provide an independent probe of cosmological parameters. Until now, gravitational lens modeling has mostly been done using CPUs or single GPUs, making modeling a slow and arduous process. We present several gravitational lenses modeled using multiple GPU nodes, with the greater computational power enabling a greater speed and a greater level of statistical rigor. Using the GIGA-Lens package for Bayesian modeling of gravitational lenses, we achieved MCMC convergence in models of six systems, including full forward models of five. These results inform the future modeling of the large number of gravitational lenses to be discovered by new telescopes including Vera Rubin and the Roman. Modeling lenses en masse paves the way towards searching for low-mass dark matter subhalos as a test of the CDM model.