[Gate-users] Parallel simulations and hardware specifications

[Matthieu Dondey]

​​

Dear Gaters,



We expect to soon have some funds in our lab to buy a simulation
workstation and one of its purposes will be Gate Monte Carlo simulations.

I’ve tried to gather what the important hardware specifications are but I’d
like to have advice from people who are already performing parallel
simulations on multi-thread workstation or clusters.

My first question is about methodology:

To split a simulation into several jobs when using a multi-core computer,
is the process the same as the one described in the Users guide - Parallel
Computing -> How to use Gate on a cluster?

Next, is it really beneficial to use GPU + CPU instead of CPUs alone? Is
the Nvidia CUDA technology well adapted to Gate simulations and what are
its benefits and downsides?

Lastly, I’ve read in the Users Guide:

*In planning simulation time, it is important to be aware that Gate
simulations currently seem to benefit only from the addition of physical
CPUs. A computer with 8 hyper-threaded physical CPU cores (16 logical CPUs)
will have the same computational efficiency if 8 processes are run
simultaneously as it would with 16 simultaneous processes.*



Can someone confirm and explain why it is impossible to use logical threads
with Gate (“seem to benefit” doesn’t seem realy categorical)? Is it an
issue which will be fixed in next versions of GATE? Is it a Geant4 specific
problem? Or is it possible to exploit logical threads on Geant4?


Any help will be greatly appreciated and I'd be very grateful about any
other advice on hardware specifications (RAM usage, important spec of
processors, benefit of a SSD, etc.)


Cheers,


Matthieu
-------------- section suivante --------------
Une pi�ce jointe HTML a �t� nettoy�e...
URL: <http://lists.opengatecollaboration.org/mailman/private/gate-users/attachments/20150410/45a0928d/attachment.html>