Hi everybody! I have a case which I solve with coupled implicit solver, about 500,000 cells. I tried to solve on 1 processor first to see how fast it does 1 iteration, and then tried with 8 processors. Surprisingly it was not faster. So my question is how does one efficiently distribute the job between 8 processors?

Thanks a lot.

First thing to look at is the speed and latency of the interconnect, if it is a distributed system. Also, splitting 500 000 cells 8 ways is a bit too much. Try it with 2, 4 and 6 to see if that helps.

Coupled implicit solver requires heavy communication between CPU's (segregated and coupled explicit solvers don't). So it is efficient only when used on shared memory machines, where all processors have acess to all the memory in the same time. When used in a distributed memory environnement (supercomputer or cluster), its peformance degrades very quickly especially on small grids.In my experience, you will have to partition the grid with at least 200.000 elements per partition, so on this case you will have to use 2, max. 3 CPU's !!

Or you could switch to explicit solver, if you really have to solve using coupled solver.

Razvan