memory (RAM) estimation for a DNS modelling
 

Dear all,

I would like to perform "Direct Numerical Simulation of the Dynamics of Internal waves". I am not sure the need for supercomputer resources in my simulation which will include 7.5E7 grid points. I want to show how much RAM is needed for this simulation and if it is more than 10 GB, I would need a supercomputer. Do you have any idea?

Thank you in advance,

Abbas

Re: memory (RAM) estimation for a DNS modelling
 

Easily! =) i.e. I use my own method for incompressible DNS/LES on tetrahedral/polyhedral mesh. Each tetra is 98bytes in RAM. Say, you have 7.5E7tetras. It means that you'll need 6,845220923424 Gbytes of RAM only to store the mesh. And (!) how long will your PC work? Even for a 1-st order flux splitting scheme and 1-st order GMRES implicit time marching scheme you'll need about 5 min. for each time step!!!

To define the size of a tetrahedral, I use something like "double size=sizeof(TETRA[i])", where TETRA[i] is a pointer to a class of tetras, and "i" is the number of a tetrahedral.