RE: FDS4 Beta Test
- Subject: RE: FDS4 Beta Test
- From: Charles Bouldin <charles.bouldin@nist.gov>
- Date: Mon, 5 Jan 2004 10:09:19 -0500
- Content-Type: text/plain; charset="us-ascii" ; format="flowed"
- In-Reply-To: <6C350EFBC115314E8EC7CE1CF497198926267C@london-data-02.london.bristol.hoar elea.com>
- References: <6C350EFBC115314E8EC7CE1CF497198926267C@london-data-02.london.bristol.hoar elea.com>
>Sender: "Miller Hannah" <millerhannah@london.hoarelea.com>
>Subject: RE: FDS4 Beta Test
>
>
>Kevin
>
>I am just buying a new computer. Is it worth buying a 64bit computer. Will
>FDS run much faster?
>
>Miller
If I can throw in my $0.02....I have been trying to figure out the
"64 bit" question myself. The first thing to say is that it is too
early to reach final conclusions! However, I think that the 64 bit
desktop machines are all likely to give significant improvements in
FDS speed.
Some hard data: I have a dual processor 2.0 ghz chips G5 Macintosh.
This runs FDS4 at a speed equivalent to a dual processor 4.2 to 4.8
ghz Pentium 4 (if such a thing existed). I have built and installed
MPI on this machine and just completed a parallel processing run that
used both cpu's. It took 59 hours on the dual processor Mac and 107
hours on a dual processor 2.4 ghz Pentium.
I expect, but cannot yet demonstrate, that similar improvements will
be possible with itanium or Athlon64 based machines. Two major
unknowns at this time are the operating system, either Linux64 or a
64-bit version of XP(?), and the existence of good 64 bit F95/C
compilers.
We don't yet have any 64-bit PC hardware that is analogous to the 64
bit Mac. The Mac also benefits from an existing IBM fortran compiler
for the G5 chip. The hardware/OS/compiler troika has to be tested in
its entirety on each platform before we can reach conclusions about
which 64 bit system is "best" for FDS runs.
At this point, a dual processor G5 Mac looks like a very good
platform to run FDS. MAJOR CAVEAT: Smokeview does not yet run on the
G5, but I am working on porting it. You can run FDS on the Mac and
view .smv files on a PC, just as we do with the Linux/PC combination
of calculation/viewing.
I'm giving a presentation at Interflam that will discuss the parallel
processing version of FDS, and will get into issues of Linux cluster,
vs. Windows cluster, vs. PowerMac clusters. Each of these
configurations has some things to recommend it. Which is best for any
particular group/lab/office/problem depends on case specific details.
I would be interested in hearing from anyone who is running FDS on an
itanium or Opteron or Athlon64 system. Probably this is something
that could be discussed at length off-list and then we could post
conclusions back to the list.
Date Index |
Thread Index |
Problems or questions? Contact list-master@nist.gov