[Wenqiang]

Hi,


I am using SU2 to simulate a simple geometry, an 3D intake with far field boundary. The mesh is fully structured which has 1.5 million points. The simulation is pure aerodynamics without any adjoint simulation. When I was running this steady case, the RAM usage is 11.5GB (from system monitor in Ubuntu). When I tried another mesh with 4 million points, my system directly kills it as I guess it needs more than 32GB RAM.



I also tried a test case from SU2 tutorial ..../compressible_flow/Unsteady_NACA0012, this mesh has 14576 points, and the RAM usage on my desktop is about 77MB. If the RAM is proportional to the number of mesh points, a 1.5million mesh will need 8GB.



So I want to ask if anyone tried the mesh larger than 1million? What is the typical RAM usage? Is 10GB the normal RAM requirement for a 1 million mesh? I can provide my .cfg file if needed, but it is almost the same with that used for Unsteady_NACA0012, nothing special..... Any information or suggestions are welcome, thanks.



all the best,
Wenqiang

[pcg]

Hi,
With a 3D RANS problem on a hex dominant mesh with multigrid, the develop branch uses about 6.5GB per million nodes with the following linear solver options:
LINEAR_SOLVER_PREC= LU_SGS
DEFORM_LINEAR_SOLVER_PREC= JACOBI

If you use the ILU preconditioner this will go up to about 8.2GB per million nodes.
Unstructured grids will use more memory as in general each node will have more neighbors.

Be careful not to specify a large number of linear solver iterations when using FGMRES, each extra iteration costs about 80 MB per million nodes.

Cheers,
Pedro

[Wenqiang]

Dear Pedro,


Thank you for your quick reply. Let me give it a try.


best,
Wenqiang