An exclusive raffle opportunity for active members like you! Complete your profile, answer questions and get your first accepted badge to enter the raffle.
Someone could tell me please what is Hybrid message passing of the Hypermesh CFD solver and in what cases should I use it? Also if it could help to reduce computational running times.
Let's say you have a compute node / machine with 20 physical cores - and you specify to run with 20 processors/cores.
By default AcuSolve will query the machine and determine it has 20 cores - and set 'number of threads' = 20. This would be pure openmp - where it's a single process with 20 threads.
Pure mpi would be where you specify 'number of threads' = 1. There you would have 20 processes, with 1 thread each.
Hybrid would be a mix. So let's say you specify 'number of threads' = 5. There you would have 4 mpi processes, with 5 threads each using openmp (to get to the same total of 20 cores used).
It's explained a little on the following page - if you look for num_threads
Let's say you have a compute node / machine with 20 physical cores - and you specify to run with 20 processors/cores.
By default AcuSolve will query the machine and determine it has 20 cores - and set 'number of threads' = 20. This would be pure openmp - where it's a single process with 20 threads.
Pure mpi would be where you specify 'number of threads' = 1. There you would have 20 processes, with 1 thread each.
Hybrid would be a mix. So let's say you specify 'number of threads' = 5. There you would have 4 mpi processes, with 5 threads each using openmp (to get to the same total of 20 cores used).
It's explained a little on the following page - if you look for num_threads
https://help.altair.com/hwcfdsolvers/acusolve/topics/acusolve/solver_programs_acurun.htm