Tench,
One sided? Sure. I read and am pasting your post below for
the readers to scroll down... not addressing point by point
since you don't care to address my points and that is fine..
Funny you should mention "the hope is critical apps". Databases?
Native Unix PA-RISC and UltraSparc III and Alpha tie down
the Unixes as far as Merced is concerned, McKinley is a different
story. CAD? Well even Unigraphics doesn't go away at first :-)
and what we will definitely see from the Alpha side of the house
is loaded $3500-$4000 EV68 boxes 8 months or so before Merced goes
volume. If Merced does indeed ship at a $2000 price point,
MDROnline's estimate of a $9000 Merced workstation is well within
reason. You would pretty much be a little looney to buy a Merced
to run an application (if avail. on Alpha) as the EV68 box will
outperform the Merced and be considerably cheaper. Besides, EV68
has many months of momentum... the high-end 1.2 GHz should be a nice
box compared to Merced and the value box at 933 MHz will do quite
nice too!
So while I agree with your assessment, I think there are considerable
barriers to entry for Merced late 2000 timeframe. According to
Craig Barrett at PCExpo he expects to see Merced in 12 to 18 months.
Maybe he is sandbagging us on that 18 months.. but still late
Y2K seems within reason for Merced boxes.
> Any questions?
Just one. What do you see as Merced's strongest segment and why?
Rob
-----
Tench wrote:
Rob, the whole picture is much more complicated than the one-sided portrait that
you paint for Alpha. In this post, I'll cover only the issue of software migration from
IA-32 to IA-64. Hope you'll forgive me for dodging the other issues because
discussing them could take up several more paragraphs.
The big problem with the migration to IA-64 is inertia. Without the hardware IA-32
translation, moving from IA-32 to IA-64 is no easier than moving, say, from IA-32
to Alpha. And Alpha already has the lead in this case. But because Merced throws
in IA-32 compatibility, the transition will be a lot easier because no one will have to
throw away the existing software base that is out there.
But why should IA-32 compatibility matter, if performance isn't going to be too
hot? After all, if all you want to run is IA-32 programs, what's the point of buying
Merced if we already have a well-established Xeon platform? Well, the hope is that
the critical apps like database management, web servers, engineering CAD tools,
and other CPU-crunching programs will be the first to make the transition to
IA-64. They're the real programs that we all care about when it comes to Spec
benchmark scores. The other apps that aren't performance-critical like e-mail
servers, Samba, network card drivers, etc., won't need to be ported to IA-64
immediately. (Actually, I don't know about the drivers, but I'm sure you get the
point.) Same thing goes with desktop applications like Microsoft Office, Netscape,
and games. In other words, the bulk of the effort is concentrated on porting the real
heavy-duty apps from IA-32 to IA-64. The rest can wait, particularly since IA-32
isn't going to die off any time soon.
How well will this transition work? Hard to say, since supporting both IA-32 and
IA-64 is dependent on the OS. We all know just how SUPERB a job Microsoft
does with their operating systems. At least there'll be other OS's like Solaris, Linux,
and SCO. And how well will IA-32 compatibility be accepted in the market?
That's also hard to tell, but supposedly the Tier 1 guys like Dell, NEC, and even
Compaq like the idea.
Any questions? |
