MadGraph5_aMC@NLO

100% CPU usage while showing 10k events with Pythia

Asked by Debarati

Dear Experts,
                         While showering only 1000 or 10k events with Pythia8 (within Madgraph setup) after having LHE file production done with Madgraph for 1000 or 10k events only for any physics process, my machine (with 8 CPUs in total, 4 cores per socket, 2 threads per core, 1 socket) is shutting down forcibly as 100% CPU usage it is having.

Thus I want to know in order to have a physics process showered with reasonable statistics (1M events lets say for pp>jj process at least) what should be the system requirement in terms of total number of CPUs, thread(s) per core, core(s) per socket, number of socket(s) if I need to shower Pythia8 within Madgraph setup ?

Thanks,
Debarati

Question information

Language:
English Edit question
Status:
Answered
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#1

Hi,

The madgraph setup is designed to split your lhef event in order to saturate your cpu.
This means that if you have more thread, madgraph will split your lhef in more segment such that your cpu is still saturated.
Therefore, I can not really answer your question, pythia8 can run correctly on a single thread (this is actually the normal mode for pythia8) but can also use a 64 thread machine. Obviously, the 64 thread machine will allow you to wait less.

Cheers,

Olivier

Revision history for this message
Debarati (droy485) said : 		#2

Sorry the title of the question should be "100% CPU usage while showering 10k events with Pythia8"

Revision history for this message
Debarati (droy485) said : 		#3

Hello,
           Thanks for some input. I have a following query from your feedback.

 Lets say if in Madgraph I generate 100 events LHEF mutiple times (so that my machine does not cross 100% CPU usage to avoid shut down since as I said for 1000 or 10k events my machine is shutting down whenever I am showering those LHEF with Pythia8 within Madgraph setup) and each time I change the "0 = iseed ! rnd seed (0=assigned automatically=default))" to a nonzero different value in run_card.dat.

Will that take me to gain larger statistics but mutually independent events to shower ?

If yes then eventually at what stage after showering each LHEF having 100 events or so and how I can merge all of them ?

Thanks,
Debarati

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#4

Hi,

Your machine should not shut down because you reach 100% cpu. (this can only happen if you have a cpu cooling issue)
Now then if this is the case the best is to set the nb_core parameter to a smaller value than the number of thread available to be sure that the cpu does not reach 100%

Now more likely than you might face some RAM issue, since then the os can kill randomly some process and sometimes do a reboot due to that. Here the solution is likely the same reduce the number of thread in parallel to reduce the pressure on the RAM.

Cheers,

Olivier

> On 16 Jun 2022, at 10:25, Debarati <email address hidden> wrote:
>
> Question #702196 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/702196
>
> Status: Answered => Open
>
> Debarati is still having a problem:
> Hello,
> Thanks for some input. I have a following query from your feedback.
>
> Lets say if in Madgraph I generate 100 events LHEF mutiple times (so
> that my machine does not cross 100% CPU usage to avoid shut down since
> as I said for 1000 or 10k events my machine is shutting down whenever I
> am showering those LHEF with Pythia8 within Madgraph setup) and each
> time I change the "0 = iseed ! rnd seed (0=assigned
> automatically=default))" to a nonzero different value in run_card.dat.
>
> Will that take me to gain larger statistics but mutually independent
> events to shower ?
>
> If yes then eventually at what stage after showering each LHEF having
> 100 events or so and how I can merge all of them ?
>
>
> Thanks,
> Debarati
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Can you help with this problem?

Provide an answer of your own, or ask Debarati for more information if necessary.

To post a message you must log in.