|January 18, 2011, 12:14||
Memory requirement for lattice boltzmann methods (simple BGK)
Join Date: Jan 2011
Posts: 1Rep Power: 0
I wondered if anyone can give an assumption on the memory requirement for a LBM - simulation with simple BGK dynamics.
So far I used only a 128x128x256 lattice, and my quad-core Desktop could run the simulations within 1-3 days.
Now we are increasing the lattice size, possible to even 1000x1000x1000 (or 750^3 is ok too).
The forumla I´ve come up with looks like:
M = N^3 * 19 * 8 * x (in byte)
19 since im using D3Q19, 8 bytes for a double datatype (could scale down to 4 for float), and N^3 for the 3 dimensions.
The big question now is the x, which should basically be 3 if I´m correct (2 velocity and 1 density matrix yes?). I do not use my own code, but use the package http://www.lbmethod.org/palabos/index.html
From the simulations I´ve done so far, it seems <x> is around 1.5 for palabos, so maybe they store the data in a more efficient way. Also, for multiple cores the required memory increases, but with 4 cores on my desktop, this can be estimated too.
I´m new to the subject, so if any1 has more experience a good estimate would help a lot, since it affects the new hardware we will buy.
thanks in advance
|Thread||Thread Starter||Forum||Replies||Last Post|
|Lattice Boltzmann Methods for Shallow Water Flows||phy4me||Main CFD Forum||0||November 16, 2010 11:56|
|Velocity Under-relaxation in SIMPLE type methods||Matt U.||Main CFD Forum||6||July 4, 2005 05:29|
|boundary conditions for lattice Boltzmann methods||Frederik||Main CFD Forum||0||July 22, 2004 13:17|
|Memory requirement for unstructured solver||Praveen||Main CFD Forum||0||July 19, 2004 10:07|
|comments on FDM, FEM, FVM, SM, SEM, DSEM, BEM||kenn||Main CFD Forum||2||July 18, 2004 18:28|