[geeks] New Itanium machines from SGI

[Joshua D Boyd]

On Tue, Jan 07, 2003 at 10:20:56PM -0600, Jonathan C. Patschke wrote:
> On Tue, 7 Jan 2003, Joshua D Boyd wrote:
> 
> > Supposedly in it's current state, the itanium2 is supposed to rival the
> > power of a single r14k chip.  SGI states that you can't pack Itanium2s
> > anywhere near as densely in an air cooled system though.
> 
> I thought when companies jump platforms, they do it because the
> resulting platform is -better- than what's already there (m68k->sparc,
> m68k->PPC, m68k->MIPS), not because it can "rival" the current
> offerings.  How is SGI pulling ahead here?

What can I say.  Just because I don't dislike the Itaniums as much as
some people doesn't mean that I see what it is that makes SGI think this
is a good move.

Still is a kinda cool machine though, despite that.
 
> > I don't anticipate it being very good for other tasks like database
> > management until the compiler situation is fixed.
> 
> And expecting a free compiler to emerge and help fix this problem is
> absolutely the dumbest business plan that I've -ever- heard.

Well, they aren't exactly just sitting around.  My understanding is that
the main hope for a good compiler rests on SGI having taken the MIPSpro
compiler and stripped out the MIPS backend, then donated it as open
source, while both SGI and Intel and paying money to a chinese
university to develop this compiler for Itanium.  

Now, let me try to find a link to support that.

http://ipf-orc.sourceforge.net/ 
http://open64.sourceforge.net/

OK, Open64 is an opensourced SGI compiler.  My recal is that Pro64 is
one of the backends to MIPSpro, and this is Pro64 minus the MIPS
support, but perhaps I'm wrong[0].  The main goal of this project is
currently to get it cleaned up and stabalized and generally usefull.

ORC is the Open Research Compiler.  To my understanding, this is the
main hope for the Itanium.  It is a fork of Open64, with the hope of
eventually being merged back together.  It is where the work is being
done by the chinese on a good optimizing compiler is being done.

There are several interesting things here, to my understanding.  First,
this is a much more modern architecture than GCC, and it should be a
better compiler than GCC for all 64 bit platforms, except it is new,
buggy, and only supports IA-64, so who knows if this would be true.
Second, despite theoretically being better for all 64bit platforms, only
Intel seems to be paying much attention to it.  Not Sun, not HP for
either of their 64bit platforms, and barely SGI.  Third, this is the
only free version of Fortran newer than 77 that I know of.

So, I hope these projects get away from being Intel only.  At the moment
it sounds like they should have a good cross platform code base by
virtue of having been moved from MIPS to IA-64, but with only Intel
sponsering it, from what I can see, I doubt that will last long.

-- 
Joshua D. Boyd

[0] http://moss.csc.ncsu.edu/pact02/tutorials.html suggests I'm right
though. See the bio of Sun C. Chan and Fred Chow.