Microsoft Releases Computational Cluster Technical Preview Toolkit

John Casu casuj at
Wed Feb 14 01:22:22 PST 2001

> On Tue, 13 Feb 2001, John Casu wrote:
> >There is one relevant advantage that Windows has over Linux, namely the
> >code generation quality of VC++ 6.0/VTune(Proton) vs. GCC/PGI for pentium III and beyond.
> >
> >While development will probably be painful, it's very possible that your app
> >will run significantly faster under a windows cluster, than under a Linux cluster.
> Are there any recent benchmarks proving this for C and/or Fortran code?
> I mean scientific code of course. For what I've seen on the spec web site
> ( benchmarks on Intel CPUs always use a windows variant,
> however they use Intel compilers rather than M$. High time Portland or
> Absoft put in some benchmarks to spec under Linux or Solaris.

My empirically unrigorous experience with some real world apps supported both on Windows 
and Linux, has been that for self contained code that doesn't use much of the OS, VC++ 
will generate better integer code than any other compiler for x86 (with the exception of 
KAP, which I've never tried), and Proton generates the best FP code but weaker integer code.

I'm waiting for Proton on Linux to come out so that I can try out a collection 
of benchmarks including Atlas, FFTW, sprng, and a bunch of others.

If Proton can "drop in" for gcc, then I shouldn't have too much work to do.

My hope is that the targetability of Proton to Pentium III and Pentium IV,
will result in some significant performance improvements.  At the very least,
we'll have another reasonably decent Fortran compiler for Linux.

It's not clear, right now, exactly how much of the KAP code is in Proton.
My guess is very little, but I'd love to be wrong.

> Groeten, David.
> ________________________________________________________________________
> Dr. David van der Spoel		Biomedical center, Dept. of Biochemistry
> s-mail:	Husargatan 3, Box 576,  75123 Uppsala, Sweden
> e-mail: spoel at	www:
> phone:	46 18 471 4205		fax: 46 18 511 755
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

More information about the Beowulf mailing list