[geeks] New Itanium machines from SGI

[Joshua D Boyd]

On Tue, Jan 07, 2003 at 05:41:48PM -0600, Jonathan C. Patschke wrote:

> > My understanding is that supposedly Itanium2s are still hamstrung by
> > poor compilers.  GCC absolutely sucks, and I'm told that Intel's own
> > compiler isn't much better in this case (hence why they are sponsering a
> > new free compiler being developed in China).
> 
> So, SGI is in the process of dumping a lean, elegant, and proven CPU
> platform with lightning-fast compilers that they're "committed" to,
> in favor of a newcomer CPU with approximately zero marketshare that
> has previously flopped in the market and has no decent compilers, with
> the hope that the single largest source of VBScript viruses and spam
> will ride the cavalry home with the silver-bullet of a compiler that
> will make everything okay?  Where to I sign up???!

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 suspect that a few hand optimized libraries would overcome much of the
compiler problems that the Itanium2 suffers.  A good fortran
implementation would also help.  There are numerous semi standard linear
algebra libraries out there, and hand optimizing one or more of them for
the itanium should make it a pretty stellar supercomputer for programs
that can make good use of the chosen library.

I don't anticipate it being very good for other tasks like database
management until the compiler situation is fixed.

-- 
Joshua D. Boyd