[mpich-discuss] Help Mpich2

[Jeff Hammond]

hyperthreading is almost universally terrible for HPC.  it's disabled
in every supercomputer i've ever used.

very few codes are sufficiently threaded for MPI+OpenMP to pay off
except when they are memory- or bandwidth-limited.  otherwise Amdahl's
Law prevails and process-based parallelism wins.

jeff

On Mon, Jul 9, 2012 at 8:22 PM, Ricardo Román Brenes
<roman.ricardo at gmail.com> wrote:
> according to my reserches, it all depends on your network and processor
> type. If u got a fast network, really fast not 1GB ethernet... you can
> "sacrifice" the transfer time. If you got a processor with hyperthreading
> its possible that ur threads run faster
>
> On Mon, Jul 9, 2012 at 6:37 PM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>>
>>
>> On 05/26/2012 06:43 PM, angelo pascualetti wrote:
>>>
>>> How many processes (-np) should tell mpich to be faster than openmp?
>>
>>
>> There's no way we can answer this question.  Depending on the application,
>> either MPI-only or MPI+OpenMP can be faster.  MPICH supports both models, so
>> you can experiment with them.
>>
>>  -- Pavan
>>
>> --
>> Pavan Balaji
>> http://www.mcs.anl.gov/~balaji
>> _______________________________________________
>> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
>> To manage subscription options or unsubscribe:
>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>
>
>
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>



-- 
Jeff Hammond
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
http://www.linkedin.com/in/jeffhammond
https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond