Lahey Licensing of Fortran compiler for Linux - in detail ;-)
ctierney at hpti.com
Tue Jan 21 12:01:52 PST 2003
On Tue, Jan 21, 2003 at 09:30:15AM -0800, Jim Lux wrote:
> At 11:06 AM 1/21/2003 -0500, John Burton wrote:
> >But the question remains. In a world where other vendors are charging for
> >the product (fortran compiler) Lahey is charging for running the programs
> >produced by that product. I wrote the FORTRAN source code. I bought the
> >hardware and software for the cluster. I'm paying for the cooling and
> >power to run the cluster. I'm paying for the systems administration. Why
> >should I pay someone else to *run* my code on my cluster? What is the
> >value they add?
> They spent time and money writing and supporting the compiler in the first
> place, that's the value they add. You could write your own code in
> assembler, but that is a)likely to cost you a lot more time in the first
> place and b) unlikely to run as fast in any case; so your total cost "to
> get the job done" is substantially lower (presumably) using their compiler
> than not using their compiler.
> As far as other vendors go: they have other goals and cost recovery/profit
> motivations. For example, Intel wants people to use their processors. It's
> in their interest to "give away" the compiler, especially if it's tied just
> to their processor (and doesn't generate code compatible with VIA or AMD).
The Intel compiler works fine on AMD chips. The Intel compiler was generating
faster code on AMD chips than Portland Group was. This was about 9 months ago,
but the Intel compiler does work. I suspect that the latest Portland Group compiler
performs better than when I last tested it. I don't see why the chipset would be a
problem either, but I have never explicitly tried on a VIA chipset, or are you
talking about another VIA.
> One might ask why Intel charges anything at all (if they do..), and in
> fact, why they don't actually pay people to use their compilers, if only to
> get more Intel processors/products visibility. (Well, actually, they DO pay
> people.. they subsidize educational instutions with free/low priced
> hardware, for instance)
Maybe Intel is just trying to gobble up marketshare. Or maybe Intel understands
that the only way to get the best performance out of their chips is to write
the fastest compilers possible. They do have an unfair advantage that they
have access to the engineers and scientists that designed the chip. In the
end, what matters to me is what is fastest. The Intel Fortran compiler produces
very fast code.
Value add seems to be the the key here. What value add is Lahey providing me
over the products from Intel or Portland Group? It isn't cost as they expect
me to pay for where I run the code, not where I compile. OpenMP support is
nice, but the other compilers provide that as well. I looked on their web site
and I didn't find any mention of SSE2 support. If you are only compiling for
Xeon (or P4) then SSE2 support is a must have. That is what makes these
inexpensive boxes so fast.
I had to evaluate all Fortran options for Linux about 12 months ago for a cluster.
I did not bother trying Lahey because of their licensing policy and that they
did not support SSE2.
Where is the value add?
Craig Tierney (ctierney at hpti.com)
More information about the Beowulf