WRF

If I had to guess they are running the WRF model [1][2]. The AI part is post-processing the model output. With a fair amount of reading the manual anybody can run their own WRF. WRF scales from running on a laptop to supercomputers with 1000s of cores. [1] https://www.mmm.ucar.edu/models/wrf [2] https://github.com/wrf-model/WRF

For some context, WRF (https://github.com/wrf-model/WRF), the current state-of-the-art weather modeling system (also developed by NCAR) can (to my limited knowledge) only run on CPU. There have been efforts to run it on GPU (https://wrfg.net)... though it doesn't look like it's been kept up to date. It seems like FastEddy mostly replaces WRF-LES, which is used for high-resolution localized modeling.

Not quite, but at optimization levels set by default WRF configuration for each compiler (definitely not a fair comparison): gfortran: -O2 -ftree-vectorize -funroll-loops ifort: -O3 I don't have the timing results anymore. This was in 2018 on Xeon Platinum 8168. I recently tried replacing -O2 with -Ofast -ffast-math to gfortran settings, which gives about 18% speed up. So still far from Intel. I recently proposed it here [1]. [1] https://github.com/wrf-model/WRF/issues/1254