Is Itanium the first 64-bit casualty?

In article
,
MitchAlsup wrote:
As one who has architected, designed and built RISCs from two (or 3)
major RISC architectures, and x86-64 high end processors: It is my
considered opinion, that the overhead of x86 is no where near as big
as you are making it out to be. I consider x86-64 with SSE... to be
EASIER to implement than SPARC-V9 with VIS--and in this case I have
actual experience at both. Now there may be some RISC somewhere that
is easier to implement than the SPARC, but your contention is that
that architecture is so much easier that the pure mass of volume in
x86-land simply can never make up the difference. This is patently
FALSE.

Looks like Sparc is the first to die, not Itanic as this thread claims.

Sun Micro kills 'Rock' high-end chip project--NYT
http://www.reuters.com/article/marke...0090615?rpc=44

Of course when Itanic dies no one will "officially" notice, there just
wont be another upgrade for HPs big servers. Core2 chips will take over.

Fujitsu will still be selling a Sparc chip long after Itanic dies, so it
all depends on how you classify death.

Brett

On Jun 21, 12:40*pm, Robert Redelmeier wrote:
Robert Myers wrote in part:

On Jun 17, 10:26*am, daytripper wrote:
And yet you, without hesitation, characterize who I am,
as if you knew me. *

I do from your posts. *I'm unconvinced you know how you
appear to others. *Or perhaps you seek negative feedback.

At my age, I have a pretty good idea of how people react to me. With
the help of paid assistance, I even have some insight as to why.

As with anyone you will ever speak to, you have no idea of the whole
story. If you did know the whole story, perhaps you would be more
tolerant--and perhaps not. There are lots of people in technical
disciplines who have all kinds of problems with others, and many are
very, very bright. None are deserving of the abuse they get from
those who should be capable of better. If you've ever found a serious
bug in a computer program, you know how badly some apparently
unimportant little thing can screw things up.

You have also "explained" to me the meaning of words uttered

No. *I have explained to you the meaning I take from
those words. *Your interpretation is not the only one.

That's true of all words ever uttered, and it's possible that I don't
understand the words you wanted to explain to me, but I *have*
actually spoken directly to the person in question about them.

by someone whom I know personally and you don't.

I'm shocked you would verbally abuse someone you
actually personally knew. *My estimation of you has
dropped even further.

I am *truly* sorry that you feel abused. My estimation of your
defenses was that your defense in depth puts you beyond the
possibility of verbal abuse. If I was wrong, please accept my humble
apologies.

You know *everything* with absolute certainty--except
yourself.

You have it reversed. *I know myself with reasonable
certainty. *Everything else is doubtful.

That's solipsism, and you in particular know yourself not at all, so
thoroughly are you surrounded by what a psychiatrist would call
defenses.

There is no perhaps about it. *If you don't like my style
of discussion, don't talk to me. *The idea that you would
try to "discipline" a total stranger is revolting.

You reject the basis of all society, large and small scale?

Of course not. The basis of society is power and a complex set of
psychological interactions that allow someone to obtain one or more
allies. The more allies you have, the more you can claim you are in
the right. That is the basis of society, and it makes little
difference whether I reject it or not. It is the way people and most
particularly mobs and gangs behave. I've stood up to lots tougher
than you--and *lots* smarter.

Someone on a religious list asked me where I would find someone pure
enough for my taste. I did not give the answer that, for the context,
would have been immune to dispute. For purposes of this discussion,
the same answer that I gave there will serve:

http://en.wikipedia.org/wiki/Diogenes_of_Sinope

When I get near hardware, just like when I get near other
people's theories, I am sometimes able to point out things
the hardware builders haven't understood because they figured
they'd just "look it up in the back of Hildebrand" when
they needed to. *That's my place in the world. *People who
think they know what they're doing do all *kinds* of things
they don't understand, sometimes with catastrophic results.

Everyone _loves_ a critic. *NOT! *
Your attitude even undermines your own effectiveness.
I have seen many more effective critics.

Where did I say that everyone loves a critic? That's ridiculous.
People hate criticism, and they should. Most criticism, as in the
case of your criticism of me, is counter-productive.

Someone should write a book, "How to lie with data."
It's pretty easy.

I was written before you were born and became a best-seller: *
"How to Lie with Statistics" *Darrell Huff [Norton]

You really *are* clueless. I know about "How to Lie with Statistics,"
and I was, in fact, making an allusion to it. My point was that you
don't even have to bother with statistics. You can just point to data
and use your imagination. Or, you can just tell a bald-faced lie that
treats the data as little more than a pretty picture. Computer
graphics, in particular, have encouraged that last practice.

As to when I was born, you have no idea who you are talking to, where
I've been or what I've done in life, both good and bad. It's truly
fortunate in this instance that my name is so common. I don't want
you or anyone else digging. My life could make a shelf full of
books. I could drop names endlessly. I took complex variables
sitting next to and as a personal friend of a person who is widely
credited with a contribution without which the Internet would not
function at all. I've been a disappointment to just about everyone,
precisely because of the perverse personality traits to which you
react without fail.

Robert Myers.

Yousuf Khan wrote:
Wilco Dijkstra wrote:
[attribution lost]
Part of the problem might also be that proprietary closed-source code
such as what exists under Windows, and within Windows itself, has
taken so many shortcuts that quick easy translation is not feasible.

Windows itself and most of its components have been ported many times
in the past to various 32-bit and 64-bit architectures, so the source
code is already clean. Compiling for a different target is as simple
as selecting the right switches in VC++.

So why don't we see this more often? If it's as simple as selecting a
different compiler switch, then it shouldn't cost too much money for an
application developer to run the source through a few different switches.

Compiling isn't the big problem; shipping a product for multiple
architectures means you have to do QA and support for each of them, and
that multiplies your costs. Even closed-source vendors for the UNIX
world typically only support a handful of systems because supporting
_all_ of them is simply not feasible.

I also run a 64-bit Linux on all of my machines, and this version is
just as well supported as the 32-bit version. Of course the big
difference being that this is open-source.

There is no big difference between open and proprietary in terms of
porting. Either can be very easy to port or very hard, depending on
how well it was written.

In theory, yes. In practice, more open-source is more easily ported.

.... because a closed-source vendor has to do all the work of porting,
testing, etc., while anyone who wants to run an open-source app on a new
system can make any necessary changes themselves and share them with
others. Also, with the huge body of portable code out there to be
viewed by new open-source coders, better habits develop and new code is
more likely to be portable than an insular closed-source environment
where the code only has to work on one system and that's all the coders see.

S

--
Stephen Sprunk "Stupid people surround themselves with smart
CCIE #3723 people. Smart people surround themselves with
K5SSS smart people who disagree with them." --Isaac Jaffe

Bill Todd wrote in
news:IMadnRLLI60Feq3XnZ2dnUVZ_hmdnZ2d@metrocastcab levision.com:

It took
a lengthy, consistent string of missed schedules and dashed performance
hopes just to slow it enough to give x86-64 and POWER the chance to
demonstrate clear superiority: had Itanic managed to fulfill just one
original promise (e.g., of significantly better efficiency due to
simpler instruction scheduling) that might have been sufficient to make
the world follow it where they so obviously expected to.

I can't argue with that.

I know one senior person who survived the DEC-CPQ-HP transitions.
The surrender to Itanic was not just about the promised gains of
Itanium, it was the 'addition by subtraction' of costs needed to stay
in the workstation and server markets. Not only could DEC get rid
of the burden of its own fabs, it could also get out of needeing to
design, debug and manufacture the various enterprise servers, mid-
range servers, low-end servers and desktops that were all deemed
necessary. They were seeing such poor ROI on those efforts, they
felt whatever meagre cash they could make by selling rebadged Intel
boxes and services for those customers would be a better market than
the ones they then held. I was told in the EV6 server era, DEC was
making more money selling storage units than the servers, truly a
case of the tail wagging the dog, but it was the cash that mattered.
They felt if they could get a high quality server via Intel and/or HP,
then selling storage and service and software would be a pretty strong
business.

I think the Itanic was not only its own failures to deliver, but also
AMD64. The 4-way, 64-bit Opteron servers really undercut everything
but the enterprise class market. Once this became clear, even Intel
stoked up the x86-64 fires and back-burnered the Itanic. Fast-foward
to today where we see Nehalem is shipping long before Tukwila.

-- Lee

On Sun, 5 Jul 2009 23:20:40 +0000 (UTC), Lee Witten wrote:

Bill Todd wrote in
news:IMadnRLLI60Feq3XnZ2dnUVZ_hmdnZ2d@metrocastca blevision.com:

It took
a lengthy, consistent string of missed schedules and dashed performance
hopes just to slow it enough to give x86-64 and POWER the chance to
demonstrate clear superiority: had Itanic managed to fulfill just one
original promise (e.g., of significantly better efficiency due to
simpler instruction scheduling) that might have been sufficient to make
the world follow it where they so obviously expected to.

I can't argue with that.

I know one senior person who survived the DEC-CPQ-HP transitions.
The surrender to Itanic was not just about the promised gains of
Itanium, it was the 'addition by subtraction' of costs needed to stay
in the workstation and server markets. Not only could DEC get rid
of the burden of its own fabs, it could also get out of needeing to
design, debug and manufacture the various enterprise servers, mid-
range servers, low-end servers and desktops that were all deemed
necessary. They were seeing such poor ROI on those efforts, they
felt whatever meagre cash they could make by selling rebadged Intel
boxes and services for those customers would be a better market than
the ones they then held. I was told in the EV6 server era, DEC was
making more money selling storage units than the servers, truly a
case of the tail wagging the dog, but it was the cash that mattered.
They felt if they could get a high quality server via Intel and/or HP,
then selling storage and service and software would be a pretty strong
business.

I think the Itanic was not only its own failures to deliver, but also
AMD64. The 4-way, 64-bit Opteron servers really undercut everything
but the enterprise class market. Once this became clear, even Intel
stoked up the x86-64 fires and back-burnered the Itanic. Fast-foward
to today where we see Nehalem is shipping long before Tukwila.

-- Lee

DEC never sold a single Itanium-based anything...

daytripper wrote in
:

DEC never sold a single Itanium-based anything...

Indeed not.

I may get the dates jumbled a bit, but it was the CPQ regime who decided
to kill the Alpha and go with Itanium. Shortly thereafter CPQ and HP
decided to merge. IIRC there were a few CPQ Merced boxes manufactured,
but clearly no DEC ones.

IIRC the decision to go to Itanium was made when EV7/21264 was in
development. A lot of the CPU and systems designers saw that for
what it was and headed on to greener pastures. Finding new jobs
wasn't too hard expecially if you were willing to relocate, DEC/CPQ
had a long series of pay freezes in place so these folks were
relatively underpaid.

-- Lee

On Jun 15, 7:40*am, Terje Mathisen "terje.mathisen at tmsw.no"
wrote:

I just noticed (at the time) that it seemed to deliver less performance
than it had been promised to do, even when measured in MIPS/GHz, and
this is why I stated that it seemed to be too complex to get to work as
Intel/HP hoped/promised that it would do.

Sorry to pick this up so late. I don't think that the problem with P4
and Itanium was as simple as that they couldn't clock it faster,
although I'm sure that's part of the story. The problem (from my
armchair architect point of view) is that Intel was always forced into
design choices it didn't want to make because of the active transistor/
power budget. I base my conclusion about P4 on comments by Bob
Colwell and comments by a lurking IBM alumnus who shall remain
nameless. I *know* that power consumption was a design problem for
Itanium, and I speculate, based on what I think I know about P4, that
it affected the design and not just the clock rate of Itanium. As
with everything I've ever said in my entire life, I could be wrong,
but that's my stubborn opinion. In the end, the design they had
envisioned in their mind's eyes just couldn't be squeezed into the
transistor budget, and the reason was power consumption. That's the
smoking gun of Itanium, in my best (amateur's) guess.

I doubt if anyone who really knows is ever going to tell us.

Robert.