Hi Ciaron,
on our Omnipath network, we encounterd a simmilar problem:
The MPI needs exclusive access to the interconnect.
Cray once provided a workaround, but that was not worth to implement (terrible
efford/gain for us).
Conclusion You might have to live with this limitation.
Kind regards,
Hello,
yesterday we upgrade our cluster from Slurm 20.02.2 to 20.02.5 and recognized
some problems with the usage of gpus and more than one cpu per task.
I could reproduce that problem in a little Docker container, which description
you could find on the following link.
https://github.com/bikerd
Il 04/11/20 19:12, Brian Andrus ha scritto:
> One thing you will start finding in HPC is that, by it's goal,
> hyperthreading is usually a poor fit.
Depends on many factors, but our tests confirm it can do much good!
> If you are properly utilizing your cores, your jobs will actually be
> slowed
On Fri, 2020-11-06 at 13:00 +0100, Diego Zuccato wrote:
> Il 04/11/20 19:12, Brian Andrus ha scritto:
>
> > One thing you will start finding in HPC is that, by it's goal,
> > hyperthreading is usually a poor fit.
> Depends on many factors, but our tests confirm it can do much good!
>
> > If you a
Il 06/11/20 13:43, Valerio Bellizzomi ha scritto:
> Usually hyperthreading will halve the memory-bandwidth available to one
> thread running in one core, the other half being used by the second
> thread.
True. That's one of the "many factors" to consider.
We tested with MPI jobs that are mostly CP
