[mdlug] opensuse 64 bit vrs 32 bit advantages speed etc.

[Dan Pritts]

On Fri, Nov 02, 2007 at 11:01:31AM -0400, Raymond McLaughlin wrote:
[aaron kulkis quoted too]

> > 4) buy 64-bit hardware, install 64-bit OS and run
> >     both 32-bit and 64-bit code.

> Option 4 is a available on *some* Intel 64 bit processors.

Any 64-bit intel CPU you'd buy in a "PC" would qualify.  You'd have
to go way out of your way and pay a lot more money to buy an Itanium.

Core 2 Duos are 64-bit; "Core" (what you might think of as "core 1")
CPUs are not.

There are 64-bit versions of the pentium 4 & celeron, as well as
Current xeons are, I think, core2-based.  some of the older p4-based
Xeons.  Current xeons are, I think, core2-based.

The relevant intel marketing word is "EM64T".

> My off the cuff recollection is that when Intel first entered the 64 bit
> market their product, codenamed Itanic^h_um was a fresh new architecture
> with no backwards compatibility, and thus unencumbered with concerns
> about supporting legacy code.

It is not compatible with x86 but it might be with HP PA-RISC.  Not sure;
but the architecture comes out of work done originally at HP and it was
certainly the market successor to PA-RISC.

A relevant intel marketing word for Itanium is "IA64".

> > Basically, the 64-bit platform is not yet ready for
> > general purpose computing -- it fulfills certain
> > niche areas (massive finite element analysis,
> > extremely large database operations, etc) but has
> > not yet been completed in many areas.
> > 
> > Even on AMD-64 chips, which can run 32-bit code
> > in a 64-bit platform (the only CPUs I know of which
> > are capable of doing this), there are currently
> > a lot of problems still.
> > 
> > Unless you have some bleeding-edge NEED for
> > 64-bit code, my advise is to wait until this
> > mess is all sorted out -- it's probably going
> > to take another 12 - 24 months.  Personally, I
> > don't need the frustration, and NOTHING that I
> > do would benefit significantly from 64-bit install,
> > but a lot would be degraded significantly if I
> > were to go that route at this time.
> 
> I disagree with this assessment. I would have had more validity 18
> months ago but the software has matured in that time, and of late you
> would really need to go out of your way to buy a 32 bit machine, eg
> closeouts and refurbs. Aguablu good enough for now, but also closer to
> the end of its life cycle and support cycle.
> 
> And on a x86_64 machine you could install 32 bit everything and be
> happy, or provided your distro carefully (I can vouch for openSuSE) you
> can install 64 bit and be just as happy and arguably better off for
> having made the transition.

Red Hat supposedly does a good job of including 32-bit compatibility
libraries too (I know I install lots of compatibility libs on my
servers but i generally run native 64-bit code).  I think windows
does OK too but I'm sure there are exceptions.

There are a few other enhancements that were made in the AMD64
architecture that can help beyond the extra address space.  The one
that comes to mind is that they doubled the number of CPU registers;
i've read that x86 is "generally felt to be a register-starved
architecture" so this could be a help.  Certainly for many benchmarks
it doesn't seem to matter.  However, this supposedly helps Java
performance a lot, and maybe better compilers will help in teh future?

Personally, I'm with Aaron - for the desktop I wouldn't bother with
a 64-bit OS today, unless i had need & budget for huge amounts of
memory (photoshop or video editing come to mind).  

But of course Raymond is right, there's not much choice other than
to buy 64-bit CPUs.

danno
--
dan pritts
danno at umich.edu
734-929-9770