MPI & OpenMP

[Reza Keshavarz]

Hello everyone,

I want to model a fire with 10 meshes (each 9m x10m x11m, each 8000 cells) with a desktop computer (12th Gen Intel(R) Core(TM) i7-12700K, 3600 Mhz, 12 Core(s), 20 Logical Processor(s) & 128GB RAM). What's the best adjustment for MPI & OpenMP? 

Thank you in advance,

[Rob]

Hi.

Just use MPI and do one mesh per core, do not use OpenMP at all. If you really have only 8000 per mesh it should go very fast. Out of interest what are you modelling?

But....

Just reading on Intel:

i7-12700K

Number of cores 12

# of Performance-cores 8

# of Efficient-cores 4

# of threads 20

...so...It would be worth investigating if it is faster with 8 meshes, as 8 cores are high Performance, 4 are efficient (I assume slow). And also to investigate if the performance cores are used first,  and if not can you can assign the meshes only to the high performance cores, not the efficient ones. If that is too complicated to do maybe make a few much smaller meshes and get the order right so that they get assigned to efficient cores, and solve quicker, so the performance cores do not wait on the efficient cores, just one idea of the top of my head. Interested to hear what the others say here. 

Rob

[Kevin McGrattan]

10 MPI processes, each with 2 OpenMP threads.

[Reza Keshavarz]

Thank you Rob/Kevin

Rob, I'm modeling a pool fire + pyrolysis in a kiln (used in forestry mills for drying lumbers). I've attached the fds file. If you have any comment, please let me know. 

Thanks again,