These are great questions! > What's the advantage of this is over running, say, at Penguin scalably on Infiniband nodes if you need the compute? POD gives you

These are great questions! > What's the advantage of this is over running, say, at Penguin scalably on Infiniband nodes if you need the compute? POD gives you the nodes and a secure shell and you take it from there. In contrast, we are building a SaaS, so the hardware and the configuration of the model (from compiling the MPI libraries to running the model) are abstracted away. The user provides input (files for now, parameters in the near future), and gets output as a result. We scale in parallel also, but don't use Penguin under the hood. > I wonder how weather researchers work if they don't have access to HPC facilities Many meteorologists only analyze existing model data, and not necessarily run the models. > is there something about WRF that allows people without domain knowledge to get sensible results? Yes, in the sense that the output data represents 3-dimensional future snapshots of the atmosphere. So you don't necessarily need to understand the internals of the model (and definitely not the labor intensive setup) to interpret the results. It really depends on what you're after. If you study atmospheric dynamics, you're likely already a meteorologist. However an energy trader knows how to interpret temperature, wind power, or insolation trends, but does not need to know about how cloud microphysics are implemented internally.