Timings of global ocean general circulation model OM3 on various platforms at MCS.

diagonal dashed lines indicate linear scaling
 

Note on Chiba City timings:  "2 proc/node" means, for example, 16 nodes were requested
for a model with 32 MPI processes.  So the 2 nodes were both competing for access to the switch.
In 1 proc/node timings, # of nodes = # of MPI processes and  the extra processor on each node is idle.
 

Below is same plot, same legend without the log scale