On 12/08/16 17:04, Diego Zuccato wrote: > Il 12/08/2016 01:42, Christopher Samuel ha scritto: > >> > The CR_ONE_TASK_PER_CORE is a hold over from then, it means that if >> > you've got HT/SMT enabled you'll get one MPI rank (Slurm task) per >> > physical core and that can then use threads to utilise the hardware >> > thread units. > > And IIUC that should give the max performance.
Depends on your application. The advice you often see for floating point performance is to disable it. But testing yourself is the only way to really be sure. >> > Without it you'll get a rank per thread, which may not be >> > useful to your code. > Can two threads of a single core get allocated to different users' > tasks? That would kill performance... > I'm trying to make things simple for my users, but that can be quite > hard in an heterogeneous cluster like ours... No, I'm pretty sure Slurm will only allocate at the physical core level, not the thread unit level. cheers, Chris -- Christopher Samuel Senior Systems Administrator VLSCI - Victorian Life Sciences Computation Initiative Email: [email protected] Phone: +61 (0)3 903 55545 http://www.vlsci.org.au/ http://twitter.com/vlsci
