Re: Workstation Computers for FLUKA

From: Nicholas Bolibruch <nboli_at_acanac.net>
Date: Thu, 25 Nov 2010 23:36:02 -0600

Hi Turgay,

Depending on the kind of simulations you are running, you can split up
the jobs into multiple processes, and run them in separate instances to
take advantage of multicore cpus. The performance scales linearly with
the number of cores. I've had good performance increases using this
approach with an Intel Core i7 (4 cores + 4 hyperthread virtual cores),
running 7 processes simultaneously, leaving one of the virtual CPUs
available so I can continue to work on the same computer with reasonable
performance. Just be careful to run the separate instances with an INP
file of a different name, and that any files generated do not
conflict/overwrite with one another. A bash or python script to manage
this can be helpful, particularly with renaming output files for use
with some of the post-processing tools.

I personally would love to see a version of Fluka take advantage of
modern graphics processors that have a large number of cores, and are
now showing good performance for double precision operations. There are
some utilities to convert Fortran 95 code to CUDA, obviously with a lot
of manual work to specify what resources to allocate for particular code
segments, unfortunately I have yet to find anything that can do this for
Fortran 77. Maybe converting from F77 to C then to CUDA, but it would
create some highly convoluted code that still needs to be maintained.
However, if my memory serves me correctly, doing this would be a
violation of the Fluka source code license agreement, as it would be
compiling Fluka for an unsupported architecture.

Hope this helps.

Cheers,

Nicholas Bolibruch
Controls& Instrumentation Development
Canadian Light Source Inc.

On 25/11/2010 2:57 AM, turgaykorkut_at_hotmail.com wrote:
> Dear FLUKA experts,
>
> If I use a workstation PC for FLUKA simulations,can I obtain shorter run time?
>
> Best regards.
>
> Dr.Turgay Korkut
>
>
>
Received on Fri Nov 26 2010 - 10:18:33 CET

This archive was generated by hypermail 2.2.0 : Fri Nov 26 2010 - 10:18:40 CET