[rescue] SunBlade 2000 vs SunBlade 2500 vs Ultra 45

[Jonathan Patschke]

On Sun, 2 Aug 2009, Phil Stracchino wrote:

> Itanium, realistically, sucked from day one.  It cannot reasonably be
> described as having ever been a successful processor.

Say whaaaat?

The 1.5GHz Itanium 2 CPUs (released 2003) in my rx2600 are just barely
beaten by 2.6GHz CPUs (released 2006) in my Mac Pro.  That is, running
Mathematica's benchmark on a Mac Pro ranks it just 20% faster than a
rx2600 per core.  I'm not entirely sure, but I would assume the benchmark
does not use a large enough dataset to thrash the memory bus.

> It managed to find a small number of customers in a few niches (mostly
> not on actual technical merit, but because the customers drank Intel's
> kool-aid) and that's about the best that can be said of it IMHO.  Long
> on complexity, short on bang-for-the-buck.

Early on, this was true because the compilers for the platform had not
matured.

> (And I'm told by people in a position to know that it's a nightmare to
> write a compiler for it.)

Yes, and this is by-design.  The IA64 architecture was designed such that
a lot of the speculative complexity that happens at runtime could be
precomputed by the compiler.  Why burn tons of silicon trying to predict
something the compiler could've told you to begin with?

However, when the compiler gets it right, whoa--IA64 really hauls in terms
of MIPS/clock.

That said, I wish the box ran a bit cooler.  It's only 2U but can heat a
reasonably-sized bedroom in about an hour.

Also, I wish there were a reasonable Unix[0] that ran well on it.  HP-UX
is not my favorite OS, but it is, IMO, at least better than Linux--barely.


[0] Yes, it wants to run VMS, and it runs VMS very well.  However, I don't
     know the VMS-specific APIs well enough to do interesting things with
     the OS.
-- 
Jonathan Patschke < "There is great satisfaction in building good tools
Elgin, TX          > for other people to use."
USA               <                                     --Freeman Dyson