Hi Drew,
My advice would be to try FiPy in parallel on an existing device and
do some speed tests to ensure you're happy. However, there's nothing
about a dual CPU workstation that should be an issue. To use FiPy in
parallel, you will need to have Trilinos installed.
On Sun, Feb 10, 2019 at 4:24 PM Drew Davidson wrote:
>
> Hello,
>
> Does anybody know how easy it is to get fipy installed such that it makes use
> of both CPUs in a dual CPU workstation? The OS would be Ubuntu or OpenSUSE.
>
> Will fipy use both CPUs out of the box? Or would I need some sophistication
> to get it to work? I know nothing about Trilinos etc.
See below.
> I don't currently have the workstation. I was somewhat interested in buying
> something like the used workstation on the ithaca ny craigslist (specs given
> below) since it seems like it might be able to do fipy phase field
> calculations for larger problems because of what seems like relatively
> numerous cores and threads, but I would not want to buy a such a device, if
> there was a chance I would be unable to configure fipy to make full use of it.
>
> I have already observed that with my single CPU system (i5-3550p), fipy seems
> to max out its four cores in ubuntu and opensuse without any additional setup.
That doesn't mean that FiPy is being efficient. Most likely, the
solvers are running in threaded mode and FiPy is not really running in
parallel.
To ensure that FiPy is actually running in parallel, try running the
test in parallel,
https://www.ctcms.nist.gov/fipy/documentation/USAGE.html#parallel-tests.
Also try running this script,
https://github.com/usnistgov/fipy/blob/develop/examples/parallel.py
using
mpirun -np 4 python parallel.p
to ensure that you have Trilinos installed correctly.
Finally, once you have those working, then try testing your specific
example using PySparse on a single node and then using Trilinos on
multiple cores to ensure that you're getting a speed up.
When running with a single core, you should only see one core being
used. If all are being used then you may have threading turned on.
This needs to be switched off when compiling Trilinos.
Cheers,
Daniel
--
Daniel Wheeler
___
fipy mailing list
fipy@nist.gov
http://www.ctcms.nist.gov/fipy
[ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
