Next: Single core versus dual
Up: NEMO performance
Previous: NEMO performance
We begin by investigating the scaling of the code for grids of equal
dimension, i.e where jpni = jpnj. This restricts us to
a relatively small number of processors counts ranging from 64 (8x8)
to 1024 (32x32). The PathScale compiler is used for processor counts
less than 96. The results are shown in figure 3.
Scaling for equal grid dimensions
Performance of NEMO when for the PGI and PathScale
From figure 3 it is clear that the PGI compiler
performs slightly better (a few percent) than the PathScale compiler.