Re: [starpu-devel] [StarPU-MPI: no dedicated thread for communication; function to set MPI_THREAD_MULTIPLE]

[Philippe SWARTVAGHER <philippe.swartvagher@inria.fr>]

Le 12/09/2024 à 14:00, Maxim Abalenkov a écrit :
> > But, IMO, StarPU+MPI on Mac will give you poor performance anyway, so 
> > it doesn't really matter if you have +/- 1 core used for computations 
> > (for prototyping, for instance)…
>
> Do you know, what is the reason for that? Why does StarPU-MPI perform 
> badly on a Mac?

Because a laptop is not done for HPC :) You will have a lot of system 
noise adding too much variability to your results.

> I have my code in StarPU for shared-memory and StarPU-MPI. I would like 
> to run a simple performance comparison test between these two versions. 
> Can you please confirm my logic? To run StarPU-MPI on a single node 
> using only this node’s cores, I set the number of StarPU CPUs equal to 
> the number of MPI. For example to have four MPI processes each running 
> on a dedicated CPU core I will run my code as
>
>    STARPU_NCUDA=0 STARPU_NOPENCL=0 STARPU_NCPU=4 mpiexec -np 4 ./code
>
> Is that right? Thank you and have a great day ahead!

This command will spawn 4 MPI processes, each using 4 CPUs for computations.

If you want 4 MPI processes with 1 core per MPI process, use something like

STARPU_NCUDA=0 STARPU_NOPENCL=0 STARPU_NCPU=1 mpiexec -np 4 ./code


--
Philippe SWARTVAGHER
Assistant Professor @ ENSEIRB-MATMECA
Research team Topal @ Inria Bordeaux-Sud Ouest