HPC systems in the US are increasingly composed of GPUs from different vendors, each with its own vendor-recommended programming model, as well as a variety of community-supported alternatives. Application developers targeting a new machine need to understand