model does not speed up if ntasks is increase or the nodes

I am running on the linux luster machine at CSU/CU and it does not matter which resolution i use, C96 or C192, only the first 8 nodes are outputting and it appears that although the logs indicate that all 144 processors are in use it takes the same time if sett this to 8 or any combination.  It feels as if there is something hard wired in the code to set this to 8 processors no matter what is set in the SBATCH and in the MPIRUN command?

The model running speed depends on: 1) node speed, number of nodes, and layouts of the nodes, 2) communication speed between nodes, 3) memory of the node, 4) I/O speed. Both SBATCH/MPIRUN command and model configure file (model_configure) need to be set for the corresponding processor number used in the run. You may also need to change the layout in the input.nml.

Thanks,

Linlin