Re: [femm] high performance computing options for FEM

[Dave Squires <djsquires@xxxxxxxx>]

Greg,
How many nodes and elements in the total mesh?
The 4-6 hours might not be unreasonable.

I work with transistor level IC simulation at millions
of transistors and simulations of overnight or even days
is not uncommon depending on function and simulation
time. The algorithms are similar for matrix solving
and Newton iterations. So if your simulation is big
with a small mesh and a million nodes/elements then
4-6 hours is not that bad.

Regards,
Dave Squires

Greg Quinting wrote:

> Members of the FEMM group ...
>
> I am a new member working on a large, static permanent magnet design
> problem for magnetic resonance. Times to calculate range from 4-6 hours
> at high mesh densities on a 550MHz PIII server with 400Mbytes RAM,
> Windows 2000 Professional. I need to optimize the computational
> efficiency while increasing the mesh densities.
>
> Does anyone have experience with any of these options? :
>
> 1. Single Pentium IV at high clock speeds.
> 2. Dual processor Pentium servers
> 3. Clusters under Windows 2000 Advanced Server or
> Datacenter Server, where one might have 8 - 32 processors.
>
> Under options 2. and 3. would I need to get into the source code to take
> full advantage of either architecture, e.g. libraries specifically
> geared towards multithreading on dual processor and/or clusters? (Early
> indications are to the affirmative.)
>
> Thanks,
>
> Greg
>
> ===
>
> > > Gregory R. Quinting, Ph.D.
> > > Anasazi Instruments, Inc.
> > > 4101 Cashard Ave. #103
> > > Indianapolis, IN 46203
> > > 317-783-4126
> > > 317-783-7083 fax
> > > aiinmr@xxxxxxx
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/