Topcluster listing

Erich Strohmaier erich at cs.utk.edu
Thu Sep 21 13:47:56 EDT 2000


In principal I would agree - a 64bit flop is a 64bit flop no mater how I
compute it. 

In practice however it seems to me to be much harder to utilizes the SIMD
units on current micro-processors than to use the "normal" FPU.
On traditional vector systems (Cray and such) you have full support of the
vector units by vectorizing compilers. I am not aware of anything
close to this on micros. Quoting SIMD flops on a wide basis would raise a
lot of expectations and cause a lot of disappointments. Rather than
inflating "Macho-flops" I would rather be more realistic in my
presentations.

Even besides this argument, I will have to swith to a CONSISTENT way of
representation regardless what it is. Quoting one system with SIMD and the
next one without does not make sense.

Erich


On Thu, 21 Sep 2000, Mattson, Timothy G wrote:

> Erich,
> 
> I disagree with your suggestion of breaking out regular 64 bit FLOPS from
> SIMD FLOPS.  A FLOP is a FLOP and you shouldn't start passing judgement on
> what kind of FLOPs are regular or irregular.  For example, are the FLOPS on
> a vector processor SIMD FLOPS or regular FLOPS or a new category of
> Vector-FLOPS?
> 
> Look, for this list to have long-term value, it needs to cut across
> architectures and be valid over a long time period.  So you don't want to
> assume a particular architecture in defining the categories on the list.  
> 
> I strongly suggest that you stick to Peak IEEE 64 bit FLOPS and LINPACK 1000
> FLOPS.  We all know the shortcomings of the LINPACK benchmark, but since it
> is a CPU bound benchmark, its a good measure of the best you can realize
> from a particular chip.
> 
> --Tim Mattson,
> Intel Corp
> 
> 
> 
> Disclaimer: My employer doesn't even know my opinions are, 
>             so please, don't hold Intel responsible for them.
> 
> 
> 
> 
>  
> -----Original Message-----
> From: Erich Strohmaier [mailto:erich at cs.utk.edu]
> Sent: Thursday, September 21, 2000 7:32 AM
> To: Brian Haymore
> Cc: Rajkumar Buyya; tfcc-l at bucknell.edu; beowulf at beowulf.org
> Subject: Re: Topcluster listing
> 
> 
> Good point, I will change 'peak' to 'regular 64 bit peak' and add a
> field 'SIMD peak' - Any better names
> 
> Erich
> 
> 
> On Wed, 20 Sep 2000, Brian Haymore wrote:
> 
> > A concern about the data that is submitted...  I can see from some of
> > the numbers people have submitted that peak performance is not
> > consistently being represented.  Not that the data is invalid, but that
> > some have represented their peak performance based on standard flops
> > while others have represented them using newer SIMD performance numbers
> > for both SSE and 3DNow!.  It would be beneficial to have things
> > consistent so that people viewing the site can find continuity in the
> > information shown.  I don't know if having the SIMD performance values
> > in or not is the right way to go, but having a standard would be better
> > I think.....  I'm sure others will also have opinions on this and I'd
> > like to hear what others think...
> > 
> > Rajkumar Buyya wrote:
> > > 
> > > Hello Erich, (and colleagues)
> > > 
> > > Great job. I agree with your concerns. I think, some of our TFCC
> volunteers
> > > will be able to help you out in verifying validity of submissions.
> > > 
> > > The web page (http://www.TopClusters.org) has been updated as per
> > > comments and suggestions from Prof. Dongarra. It appears like we are
> really
> > > heading towards making all happen.
> > > 
> > > ----------- The main update made to the site is --------------
> > > BENCHMARKS
> > > 
> > > We are starting to collect and rank the clusters on the bases of the
> LINPACK performance results and
> > > will later include benchmarks for communication and I/O.
> > > 
> > > Linpack HPC benchmark url: http://www.netlib.org/benchmark/hpl/
> > > -----------
> > > 
> > > I am sure many members of the list either have their own clusters
> (mostly built by
> > > themselves) or have access to them. It will be great to see all of them
> actively
> > > making submissions of their clusters to the list. Also, inform about
> TopClusters.org
> > > to their friends and colleagues.
> > > 
> > > Regards
> > > Raj
> > > 
> > > Erich Strohmaier wrote:
> > > >
> > > > Hi everybody,
> > > >
> > > > Yes, I set up the prototype in February and announced it on beowulf
> and
> > > > tfcc. I have also filled the DB mainly with information I gathered
> myself
> > > > from (potentially outdated) web-pages. We have 59 entries right now.
> > > >
> > > > Submissions from the web-interface are sent to me. I edit them and
> update
> > > > the DB. I hope I have catched up now:-)
> > > > I hesitate allowing direct update as there is a far amount of 'trash'
> > > > coming in (such as kids typing in their 'macs')
> > > >
> > > > This is a prototype and things are changing rapidly at times. This is
> why
> > > > Markus and the other people in the TOP500 project do not have access
> for
> > > > updates right now. This will change once the DB structure is
> stabilized.
> > > >
> > > > What we need right now is more active submissions ( I had less then 10
> > > > sites submitting their info themselve!) as well as feedback to the
> > > > usefullness of the collected information.
> > > >
> > > > Thank you
> > > >
> > > > Erich
> > > >
> > > > > Erich has set up the database and I guess he also has all the tools
> > > > > and servers to do this.
> > > > > >From what I understand is that a submission is sent to him via
> > > > > email.
> > > > > I just wanted to get the submission going.
> > > > >
> > > > > Markus
> > > > >
> > > > > > > I'm not sure how much good this will do.  I submitted two of our
> clusters
> > > > > > > well over a month ago, and they've never been added to the list.
> So, I'm
> > > > > > > wondering if anyone is actually maintaining the list.
> > > > > >
> > > > > > Hi Roger,
> > > > > >
> > > > > > Markus Fischer is one of the members doing work on TOP500 project
> and I hope he will
> > > > > > maintaining our TopClusters database/list in the future.
> > > > > >
> > > > > > Regards
> > > > > > Raj
> > > > >
> > > > > --
> > > > > ---------------------------------------------------------------
> > > > > Dipl. Inf. Markus Fischer     Tel.: +49 621 181-2718
> > > > > University of Mannheim                Fax:  +49 621 181-2713
> > > > > B6, 26                                email:markus at markus-fischer.de
> > > > > 68159 Mannheim
> > > > >
> > > > >
> > > >
> > > >
> ===========================================================================
> > > > Erich Strohmaier                       email:  erich at top500.org
> > > > Department of Computer Science         phone:  ++ 1 (865) 974 0293
> > > > 1122 Volunteer Blvd, Suite 203         fax  :  ++ 1 (865) 974 8296
> > > > Knoxville TN, 37996-3450 - USA         http://www.cs.utk.edu/~erich/
> > > 
> > > --
> > > Best wishes,
> > > Raj
> > > 
> > > ------------------------------------------------------------------------
> > > Rajkumar Buyya
> > > School of Computer Science and Software Engineering
> > > Monash University, C5.10, Caulfield Campus
> > > Melbourne, VIC 3145, Australia
> > > Phone: +61-3-9903 1969 (office); +61-3-9571 3629 (home)
> > > Fax: +61-3-9903 2863; eFax: +1-801-720-9272
> > > Email: rajkumar at buyya.com, org | rajkumar at csse.monash.edu.au
> > > URL: http://www.buyya.com, org | http://www.csse.monash.edu.au/~rajkumar
> > > ------------------------------------------------------------------------
> > > 
> > > _______________________________________________
> > > Beowulf mailing list
> > > Beowulf at beowulf.org
> > > http://www.beowulf.org/mailman/listinfo/beowulf
> > 
> 
> ===========================================================================
> Erich Strohmaier                       email:  erich at top500.org
> Department of Computer Science         phone:  ++ 1 (865) 974 0293
> 1122 Volunteer Blvd, Suite 203         fax  :  ++ 1 (865) 974 8296
> Knoxville TN, 37996-3450 - USA         http://www.cs.utk.edu/~erich/
> 
> 
> _______________________________________________
> Beowulf mailing list
> Beowulf at beowulf.org
> http://www.beowulf.org/mailman/listinfo/beowulf
> 
> 

===========================================================================
Erich Strohmaier                       email:  erich at top500.org
Department of Computer Science         phone:  ++ 1 (865) 974 0293
1122 Volunteer Blvd, Suite 203         fax  :  ++ 1 (865) 974 8296
Knoxville TN, 37996-3450 - USA         http://www.cs.utk.edu/~erich/



_______________________________________________
Beowulf mailing list
Beowulf at beowulf.org
http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list