[IP] more on Apple to ditch IBM, switch to Intel chips | CNET News.com
Begin forwarded message:
From: Newmedia@xxxxxxx
Date: June 5, 2005 2:47:29 PM EDT
To: dave@xxxxxxxxxx
Cc: shap@xxxxxxxxxxx
Subject: Re: [IP] more on Apple to ditch IBM, switch to Intel chips |
CNET News.com
Jonathan:
> One reason for this is that Sun's attempts to develop high
> performance SPARC chips have failed.
Just as Intel is now switching to dual-core (to be followed by quad,
etc.), 64-bits, encryption (LeGrande), virtualization (Vanderpool),
imbedded management and so one -- they call it *Technology, the rest
of us might simply call it "architecture" -- since the Megahertz Race
has slowed due to heat/power problems at 90/65nm (and Itanium has
been niched at the high-end) . . . so one should expect IBM, Sun, AMD
and others to turn to more sophisticated system design features also.
Since annoucing their division of labor with Fujitsu earlier this
year, Sun's design efforts for SPARC are almost entirely focussed on
CMT (Chip-level Multi-Threading.) The upcoming "Niagra" design --
with parts that are regularly flashed around at Sun meetings -- will
handle 4 threads per core and sport 8 cores per die. 32 threads per
chip clearly isn't intended to win at SpecInt benchmarks. Mips isn't
the goal of this design . . . obviously.
What Sun (and many others like Azul Systems, etc) have begun to
tackle is the fact that many widely differing workloads are starting
to congeal out of the "Network Computing" cloud and that it probably
makes sense to build machines that can specialize on one or another
group of these workloads. Hey, we've got plenty of transitors to
throw around so why not?
Today none of this can be done with AMD64s . . . or any other
"standard" microprocessor. This is also, by the way, one of the
reasons why Sony and Microsoft are using custom designs for their
next-generation game consoles. If you know your workload, then you
probably don't want a standard part -- or so the logic goes.
Will the next generation "K-10" AMD parts -- now being designed by
Charles Moore, ex-IBM architect of the Power 4 -- be able to handle
this level of systems "customization"? Will Andy Bectholsheim be
able to build "structured ASICs" that incorporate AMD64 cores the way
that Apple, Sony and Microsoft can now do this with PowerPC cores?
Will Transitive's emulation technology adequately bridge from SPARC
to x86?
My guess is that analysts and investors -- as well as customers --
will have plenty of exciting new systems technologies to talk about
over the next 4-5 years . . . which is what makes all this
interesting for me.
Best,
Mark Stahlman
New York City
-------------------------------------
You are subscribed as roessler@xxxxxxxxxxxxxxxxxx
To manage your subscription, go to
http://v2.listbox.com/member/?listname=ip
Archives at: http://www.interesting-people.org/archives/interesting-people/