The new generation of heterogeneous CPU/GPU computer systems offer much greater computational performance but are not yet widely used for climate modeling. One reason for this is that traditional climate models were written before GPUs were available and would require an extensive overhaul to run on these new machines. In addition, even conventional “high–resolution’ simulations don’t provide enough parallel work to keep GPUs busy, so the benefits of such overhaul would be limited for the types of simulations climate scientists are accustomed to. The vision of the Simple Cloud-Resolving Energy Exascale Earth System (E3SM) Atmosphere Model (SCREAM) project is to create a global atmospheric model with the architecture to efficiently use GPUs and horizontal resolution sufficient to fully take advantage of GPU parallelism. After 5 yrs of model development, SCREAM is finally ready for use. In this paper, we describe the design of this new code, its performance on both CPU and heterogeneous machines, and its ability to simulate real-world climate via a set of four 40 day simulations covering all 4 seasons of the year.

This paper describes the first implementation of the d x=3.25 km version of the Energy Exascale Earth System Model (E3SM) global atmosphere model and its behavior in a 40 day prescribed-sea-surface-temperature simulation (Jan 20-Feb 28, 2020). This simulation was performed as part of the DYnamics of the Atmospheric general circulation Modeled On Non-hydrostatic Domains (DYAMOND) phase 2 model intercomparison. Effective resolution is found to be $\sim 6x the horizontal grid resolution despite using a coarser grid for physical parameterizations. Despite this new model being in an immature and untuned state, moving to 3.25 km grid spacing solves several long-standing problems with the E3SM model. In particular, Amazon precipitation is much more realistic, the frequency of light and heavy precipitation is improved, agreement between the simulated and observed diurnal cycle of tropical precipitation is excellent, and the vertical structure of tropical convection and coastal stratocumulus look good. In addition, the new model is able to capture the frequency and structure of important weather events (e.g. hurricanes, midlatitude storms including atmospheric rivers, and cold air outbreaks). Interestingly, this model does not get rid of the erroneous southern branch of the intertropical convergence zone nor the tendency for strongest convection to occur over the Maritime Continent rather than the West Pacific, both of which are classic climate model biases. Several other problems with the simulation are identified, underscoring the fact that this model is a work in progress.

One of the largest sources of uncertainty in climate models comes from convective cloudssystems parametrizations, which are necessary at current horizontal resolutions, of about 25km.In order to fully resolve the effects of such systems, horizontal resolutions need to reach the 1-3km range, which entail a huge increase in computational costs. In recent years, the performance of the world largest supercomputers has steadilyimproved, to the point that we can now think of running a climate modelat cloud resolving resolution, with a Simulated Years Per Day (SYPD) throughputthat is suitable for decade or century long simulations.However, the architectures of the newest supercomputers are becoming more andmore heterogeneous, which makes the task of keeping a performant code base moreand more challenging. Here, we present our effort to upgrade the new non-hydrostatic atmosphere dycore ofE3SM to a version that is highly performant on a variety of architectures, includingGPUs, conventional CPUs, and many-core CPUs. When using GPUs, our implementation isone of the fastest, achieving 0.97 SYPD on the NGPPS benchmark, when running on the full Summit supercomputer.