Hi Foamers I am trying to s
I am trying to solve a BIG CFD problem in a NOT SO BIG computer, so i had this idea, correct me if i am wrong here:
Using the precompiled OpenFOAM single precision or compile it and use it in my problem should roughly be the same as "doubling" the memory capacity, since the float representation requirements in C++ is roughly half the double float representation requirements;
So that shoul be enough to solve a problem with a size double of the ones that i solve in double precision, but there is a catch ( at least i think so )... :
Due to the reduced precision, the effect of round up errors may simply abort convergence of the solvers in OpenFOAM... or increase drastically the number of iterations needed for convergence...
Does anybody has experience in this...?
It's quite possible. CG solver
It's quite possible. CG solvers, for instance, update search directions based on the residual dot-product of the previous iteration. If the dot-product is inaccurate due to round-off, the search direction falters a little bit, and hence the convergence issues.
|All times are GMT -4. The time now is 10:41.|