Hi all, phenix.real_space_refine runs pretty decently with run=minimization_global+adp, and the adp part is already parallelized. Using local_grid_search though, it slows down enormously, particularly for large structures, and I’d really like to try using it to correct poorly fit rotamers. Is there any prospect for parallelizing this function? Naively it seems like something that might be amenable to parallelization - so long as adjacent local grids aren’t searched at the same time, it would seem like one might be able to split up the job to run on multiple threads? These days a lot of folks have GPU workstations for cryoEM, which often (in addition to GPUs) have Xeon or Threadripper CPUs with many cores - I’d love to be able to use that computational horsepower effectively for refining my structures. Cheers Oli