DYAMOND: the DYnamics of the Atmospheric general circulation Modeled On Non-hydrostatic Domains

Progress in Earth and Planetary Science

Table 5 Computational aspects

Model	Nodes	Cores	SDPD	Δ t	Processor (date of launch into market)
ARPEGE-NH	300	7200	2.6	100 s	Intel Xeon “Ivy Bridge” (2013)
FV3	384	13,824	19	4.5 s	Intel Xeon “Broadwell” (2016)
GEOS	512	20,480	6.2	3.75 s	Intel Xeon “Skylake” (2017)
ICON	540	12,960	6.1	4.5 s	Intel Xeon “Haswell” (2014)
IFS	360	12,960	124	240 s	Intel Xeon “Broadwell” (2016)
MPAS	256	9216	3.5	10 s	Intel Xeon “Broadwell” (2016)
NICAM	640	2560	2.6	10 s	NEX SX-ACE vector processor (2015)
SAM	128	4608	6.0	7.5 s	Intel Xeon “Broadwell” (2016)
UM	340	12,240	6.0	90.0 s	Intel Xeon “Broadwell” (2016)

To give a rough idea of the performance of the different models, their throughput is given and contextualized by some basic information about architectures on which they were run. Columns indicate the number of nodes, the total number of cores on which the simulations were run, the simulated days per day, the shortest (usually acoustic) timestep (Δt), and the type of processor