Hi,
I am simulating the heat transfer across a rod bundle. The mesh is more than 500,000 cells. I takes overnight to run 40 seconds unsteady flow on my AMD Turion X2 computer, longer than needed by a famous commercial CFD software. Is it the right computing time for Code_Saturne? Please advice.
Regards,
Bintoro
computing time
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
Marc Sakiz
Re: computing time
Hi,
you can of course tune the time step, but I guess you did that already and your comparison with the other code is done with the same parameters.
When comparing Code_Saturne with other "famous CFD softwares", there is indeed a major difference. By default, Code_Saturne uses a 10-8 precision for solving the linear systems, while other codes usually use 10-5, sometimes even more.
Actually, this value of 10-8 for Code_Saturne comes from the early stages of development, when we wanted to study the precision of the algorithms and not have the results polluted by errors on linear solvers. Then it became a habit ... and we never took the time to study the effect of an increase in this value ... until now!
In the current validation campaign for version 2.0 we are indeed making comprehensive tests on using a value of 10-5, and so far we get similar results, for a much faster CPU time (20% to 80% gain depending on the configuration). It will probably become the default value. So it seems you can safely change the value to 10-5 in your tests.
To change the value for the solver precision, go to "Numerical Parameters -> Equation parameters" in the GUI and change "Solver precision" for all the variables. Alternatively, you can change it in the routine usini1.f90, the parameter for variable IVAR is EPSILO(IVAR).
Best regards
--Marc--
you can of course tune the time step, but I guess you did that already and your comparison with the other code is done with the same parameters.
When comparing Code_Saturne with other "famous CFD softwares", there is indeed a major difference. By default, Code_Saturne uses a 10-8 precision for solving the linear systems, while other codes usually use 10-5, sometimes even more.
Actually, this value of 10-8 for Code_Saturne comes from the early stages of development, when we wanted to study the precision of the algorithms and not have the results polluted by errors on linear solvers. Then it became a habit ... and we never took the time to study the effect of an increase in this value ... until now!
In the current validation campaign for version 2.0 we are indeed making comprehensive tests on using a value of 10-5, and so far we get similar results, for a much faster CPU time (20% to 80% gain depending on the configuration). It will probably become the default value. So it seems you can safely change the value to 10-5 in your tests.
To change the value for the solver precision, go to "Numerical Parameters -> Equation parameters" in the GUI and change "Solver precision" for all the variables. Alternatively, you can change it in the routine usini1.f90, the parameter for variable IVAR is EPSILO(IVAR).
Best regards
--Marc--