The great leveling

Robert Myers wrote:
Well, Power and Sparc both have large existing software bases, so
whether they are hobby chips or not, at least they are useful hobby
chips.

I don't know how much the installed software base really has to do
with what's left of the target markets. IBM wants to protect its
juicy slice at the high end and others would like to grab a slice of
it. I don't _think_ x86 is plausible for that slot unless Intel
decides it's going to try to make it plausible (and, at that, I think
it won't be easy). So the question is, what chip(s) will play in that
market, say, five years from now.

Well, the last I heard Sparc/Solaris had something like 10,000 applications
ported to it. In some cases, Sparc was the only platform those applications
were ever ported to, not even a port for x86 anywhere in sight.

Don't know about IBM Power, but between AIX, OS/400 and all of the other
OSes that run on it (probably includes MacOS too), it's probably also got a
similar number of applications for it.

Then you have to factor in all of the countless custom inhouse apps that
people must've written for them over the years. Then you have to factor in
all of those custom inhouse apps for which people have lost the source code
for, and you now have instant processor lock-in.

Power is a nice chip, but it's a money-loser for IBM. Whether Power
goes or stays depends entirely on how critical IBM thinks it is to
protecting its high-end franchise.

Well, IBM obviously thinks it's pretty damn important to them to keep that
franchise alive. The processor is only a small cost in those high-end
systems, it makes more money selling the systems and their related services
than they lose in the processor.

Sparc's hold on life seems even more tenuous. You'll tell me why I'm
being silly for saying so, I'm sure, but I don't understand why Sun
doesn't see Solaris on Itanium as at least as attractive as its
capitulation to Microsoft.

Sun may be ready to kill off its own implementation, UltraSparc, but it's
now ready to consolidate the market with the Fujitsu version, Sparc64. All
of the people with the lost source code will have to keep buying Sparcs for
a long long time.

At some point, eventually Opteron may account for more than 50% of Sun's
base, but until then Sparc has to be supported, it has no choice but to
support it. Sun can continue to support it through Sparc64 rather than
UltraSparc though.

That leaves Itanium, heavy baggage and all. Intel knows how to make
lots of chips cheap (or at least it did at one time). The story of
the industry for the last two decades has been lots of chips cheap. The
Sparc/Power market is the last holdout.

Well, one of the theories about why Intel is experiencing an inventory glut
right now is because it has an overcapacity right now. Intel has ten chip
plants, five of which are 300mm plants! One or two 300mm plants should be
enough to supply the world with all of the Pentiums and Xeons that they
need. So it looks like it may have built the three extra plants to supply
the world with its previously anticipated $30bn worth of Itaniums. Obviously
since that $30bn worth of Itanium orders hasn't materialized, it's now busy
trying to fill up the capacity of those extra plants with everything from
WiFi chipsets to HDTV chips. It's trying to create enough products to fill
up the capacity of its plants. I don't think even Intel has enough engineers
to spread around to such diverse engineering projects.

In fact, I think Intel is overlooking one of the commodities that it has
plenty of -- chip plants. It should start renting itself out to other
chipmakers to produce their chips, much like TSMC, UMC, Chartered, SMIC,
etc. I doubt they will do this, but it would fill up their plants.

Yousuf Khan

On Tue, 21 Sep 2004 00:10:12 GMT, Robert Myers
wrote:

Yousuf Khan wrote:

Robert Myers wrote:

As to analysts, the sales forecasts having been so useless in the
past, why would anyone take to believing them now?


Comedic relief?


[dropped comp.sys.intel so I don't get pulled over by the OT police]

I think analyst's estimates are a form of intellectual reinsurance
("Well, of course it's BS, but what else do you want me to rely on--my
own BS? My dart board?").

What do _you_ think Yousuf? Will Itanium die while Power and Sparc
survive? All three chips look more like expensive hobbies than
realistic business propositions (and a dart board looks really
attractive for predicting the future of that market segment).

My personal dart board has both Itanium and SPARC dying out in the
not-too-distant future, at least for the high-end of things (legacy
support will of course continue for quite some time). Really I only
see a bright future for three processor ISAs: x86, PowerPC and ARM.

While Power at the high-end might not be a good money-making venture,
the ISA in general has some pretty good support throughout various
market segments. In particular it seems to be doing very well in the
console market and the high-end of the embedded market. At the
top-end with the Power servers it might not make much money, but it
helps push the development for the more profitable chips further down
the line.

ARM, of course, is probably the top selling ISA out there these days
(with the possible exception of some really low-end/low-cost stuff
like 6805 or 8051), and it will probably continue doing well on the
really low-power front.

x86, meanwhile, seems likely to continue dominating the desktop and
workstation market for the foreseeable future while constantly taking
away server marketshare from the traditional big-iron machines.

At least.. that's how I see things going. Now, if you don't mind
stepping aside, I have a few more darts to throw :

-------------
Tony Hill
hilla underscore 20 at yahoo dot ca

Robert Myers wrote:

Power is a nice chip, but it's a money-loser for IBM. Whether Power
goes or stays depends entirely on how critical IBM thinks it is to
protecting its high-end franchise.

Is it really losing money, when you factor-in the spin-offs to the
embedded world, such as video-game consoles?

Looks to me like IBM is in a great position with Power, to take up
most of the 64-bit market (after X86-64 takes it's piece of the pie,
of course).

Sparc's hold on life seems even more tenuous. You'll tell me why I'm
being silly for saying so, I'm sure, but I don't understand why Sun
doesn't see Solaris on Itanium as at least as attractive as its
capitulation to Microsoft.

Sparc's days are numbered, no doubt in my mind.

That leaves Itanium, heavy baggage and all. Intel knows how to make
lots of chips cheap (or at least it did at one time). The story of the
industry for the last two decades has been lots of chips cheap. The
Sparc/Power market is the last holdout.

Doesn't look too good for Itanic right now. I suppose if Intel
continues to infuse money into the program, they can keep it alive
long enough to evolve it into something competitive. We'll see.

"Yousuf Khan" wrote:

Well, one of the theories about why Intel is experiencing an inventory glut
right now is because it has an overcapacity right now. Intel has ten chip
plants, five of which are 300mm plants! One or two 300mm plants should be
enough to supply the world with all of the Pentiums and Xeons that they
need. So it looks like it may have built the three extra plants to supply
the world with its previously anticipated $30bn worth of Itaniums.

I doubt that theory. They've known for years (possibly all along)
that Itanic was not going to take the world by storm. Certainly,
since the Rambus debacle, they have known that they can't force
whatever they want onto a resistant market.

Over- and under-capacity situations are simply life as usual in the
semiconductor industry, where you have to plan-for and build fabs,
years before expected demand for your product. A lot can happen in
those years.

Obviously
since that $30bn worth of Itanium orders hasn't materialized, it's now busy
trying to fill up the capacity of those extra plants with everything from
WiFi chipsets to HDTV chips. It's trying to create enough products to fill
up the capacity of its plants. I don't think even Intel has enough engineers
to spread around to such diverse engineering projects.

Of course Intel would like to use some of their CPU bounty to expand
into other markets and grow the business. I'm not convinced that
their product proliferation implies that they are desperate to fill
excess manufacturing capacity.

In fact, I think Intel is overlooking one of the commodities that it has
plenty of -- chip plants. It should start renting itself out to other
chipmakers to produce their chips, much like TSMC, UMC, Chartered, SMIC,
etc. I doubt they will do this, but it would fill up their plants.

I'm certain it's crossed their minds, and if they're not doing it,
it's for a good reason (like the profit margins aren't high enough).

George Macdonald wrote:

On Tue, 21 Sep 2004 00:10:12 GMT, Robert Myers
wrote:


Yousuf Khan wrote:


Robert Myers wrote:


As to analysts, the sales forecasts having been so useless in the
past, why would anyone take to believing them now?


Comedic relief?


[dropped comp.sys.intel so I don't get pulled over by the OT police]

I think analyst's estimates are a form of intellectual reinsurance
("Well, of course it's BS, but what else do you want me to rely on--my
own BS? My dart board?").


Who is that talking in quotes though?:-)... the corporate system buyer?...
the stock speculator?... the system mfr? Intel has certainly gotten a lot
of mileage, PR-wise out of the wild forecasts by analysts... and there's
never a day of final reckoning where facts have to be faced down and denied
or accepted. Nobody ever says: "**** IDC - they're useless... ignore
them". Conspiracy?... err, I hope not!


I don't think you need a conspiracy. Why would an analyst with no
working engineers or scientists want to try to outguess Intel? And
there is the fact that, aside from being an investment vehicle itself,
Intel is an investor, and, like all investors, likes to be told what it
wants to hear.

As to who would buy analyst projections, I'd think that just about
anybody who needed to plug a related number into a spreadsheet and had
to be able to defend it would be a potential customer. Everybody in the
business has to be (or should be) wondering where the center of gravity
of the industry is headed. If the analysts really did know, their
predictions would be extremely valuable, I would think. They don't
really know, but a guess based on accepted methodology is much better,
or at least safer, than nothing. I thought you were in OR. Your stuff
works better? ;-).


What do _you_ think Yousuf? Will Itanium die while Power and Sparc
survive? All three chips look more like expensive hobbies than
realistic business propositions (and a dart board looks really
attractive for predicting the future of that market segment).


Without inside knowledge it's hard to be sure but IBM has tremendous depth
and scope for using their designs across a range of internal developments
as well as selling merchant chips in several derivative forms...
sustainable?... I dunno...

IBM sold its Power 440 IP and I gather its commitment to being a
supplier to Apple is less than certain. IBM did form the Systems and
Technology Group, apparently giving up on the idea that Microelectronics
could stand on its own profitability. That signals a commitment to
Power and allows them to hide just how much Power is really costing
them. The subtext, though, is that Power, on its own, is never going to
be a money-maker. How long will IBM be willing or able to tough it out?
Right now, the evidence is that IBM has made the right choice and HPaq
the wrong choice. Over the long run? I still think a proprietary chip
is running against the tide. Only time will tell.

... but I think I have at least as good an idea as
any err, analyst. When something like Alpha can turn rotten, anything can
happen.

I'm amazed at the bandwidth that has been consumed on Alpha without much
of anybody facing up to what happened the the chip was too expensive
to be a merchant chip and the software base never fully materialized.
Where is Windows on Power, anyway?

So will we end up with just x86-64 and ARM as *the* computer
architectures to choose from?

Aside from the embedded market, maybe.

The problem (as always, from my limited perspective) is that none of the
revolutions in microprocessor design have really been revolutions in the
sense that they answered questions there was a big payoff for anwering.
Intel thought IA64 was a revolution that answered an important
question (how to get significant parallelism without recoding
everything), but other architectures have been just about as successful
(or unsuccessful) in achieving the same goal.

It's not as if there were no important questions worth asking--latency
tolerance, moving data around as the virtual real estate gets larger,
and, of course, power consumption--come to mind, but the demand drivers
just aren't big enough to drive a real revolution. Maybe if (say)
google succeeds in its plans for world domination and needs a real low
power revolution the way HPC needs a low power revolution.

What will the Chinese do?... do they
matter?shrug

Of course they matter, but not soon enough for any but the most foolish
to speculate how.

RM

Yousuf Khan wrote:

Robert Myers wrote:


snip


Then you have to factor in all of the countless custom inhouse apps that
people must've written for them over the years. Then you have to factor in
all of those custom inhouse apps for which people have lost the source code
for, and you now have instant processor lock-in.


Lock-in is dead. Even IBM admits it. The day of mainframe on Unix is
here to stay. People buy the high end hardware because they want the
extra insurance of reliability built into the hardware. The easiest way
to get that reliability is still to buy IBM, but it's not the only way.

How all this works with Sun? I put Sun's dog and pony show to the
financial sector on while I was doing other things yesterday. I haven't
heard so much upspin since the days of high school pep rallies. If
their apps were so locked in, I don't think they'd sound so eager.

snip


That leaves Itanium, heavy baggage and all. Intel knows how to make
lots of chips cheap (or at least it did at one time). The story of
the industry for the last two decades has been lots of chips cheap. The
Sparc/Power market is the last holdout.


Well, one of the theories about why Intel is experiencing an inventory glut
right now is because it has an overcapacity right now. Intel has ten chip
plants, five of which are 300mm plants! One or two 300mm plants should be
enough to supply the world with all of the Pentiums and Xeons that they
need. So it looks like it may have built the three extra plants to supply
the world with its previously anticipated $30bn worth of Itaniums. Obviously
since that $30bn worth of Itanium orders hasn't materialized, it's now busy
trying to fill up the capacity of those extra plants with everything from
WiFi chipsets to HDTV chips. It's trying to create enough products to fill
up the capacity of its plants. I don't think even Intel has enough engineers
to spread around to such diverse engineering projects.

In fact, I think Intel is overlooking one of the commodities that it has
plenty of -- chip plants. It should start renting itself out to other
chipmakers to produce their chips, much like TSMC, UMC, Chartered, SMIC,
etc. I doubt they will do this, but it would fill up their plants.


That would all put a strong SELL on Intel, I think. It would be like
betting against the Yankees or the Green Bay Packers in the days of
Vince Lombardi.

RM

chrisv wrote:

Robert Myers wrote:


Power is a nice chip, but it's a money-loser for IBM. Whether Power
goes or stays depends entirely on how critical IBM thinks it is to
protecting its high-end franchise.


Is it really losing money, when you factor-in the spin-offs to the
embedded world, such as video-game consoles?

Looks to me like IBM is in a great position with Power, to take up
most of the 64-bit market (after X86-64 takes it's piece of the pie,
of course).


That assumes that IBM can profitably market Power as a volume chip. I
infer that it can't. Don't expect an audit of IBM to reveal what IBM
really thinks Power costs, but the relationship with Apple seems to be a
troubled one, and IBM has consolidated Microelectronics (which had a
streak of losing money) with other business units.

IBM may not be making the decisions it is making for the reasons I am
inferring, but my guesses do pass the laugh test, at least. Turn the
equation around: IBM has put up some really impressive performance
numbers for Power and Apple has a solid, if limited, franchise. What
would Intel do in a situation like that? It would flog the living
daylights out of the chip, that's what. Not IBM's style? Maybe not,
but they're in business to make money, and, if Power were such a strong
player, IBM would be playing it much more surehandedly.

RM

chrisv wrote:
I doubt that theory. They've known for years (possibly all along)
that Itanic was not going to take the world by storm. Certainly,
since the Rambus debacle, they have known that they can't force
whatever they want onto a resistant market.

Well, knowing about it is one thing, admitting to it is another.

Over- and under-capacity situations are simply life as usual in the
semiconductor industry, where you have to plan-for and build fabs,
years before expected demand for your product. A lot can happen in
those years.

Yes, things such as their current overcapacity.

Obviously
since that $30bn worth of Itanium orders hasn't materialized, it's
now busy trying to fill up the capacity of those extra plants with
everything from WiFi chipsets to HDTV chips. It's trying to create
enough products to fill up the capacity of its plants. I don't think
even Intel has enough engineers to spread around to such diverse
engineering projects.

Of course Intel would like to use some of their CPU bounty to expand
into other markets and grow the business. I'm not convinced that
their product proliferation implies that they are desperate to fill
excess manufacturing capacity.

The WiFi chips I can understand, they have a support relationship to their
existing core business (CPUs). HDTV is another thing altogether.

Intel seems to have a history of flailing about when it comes to selling
products outside of its core business. It starts a business and then just as
abruptly kills them. Remember those home networking gear it used to sell at
one time? Also the webcams?

In fact, I think Intel is overlooking one of the commodities that it
has plenty of -- chip plants. It should start renting itself out to
other chipmakers to produce their chips, much like TSMC, UMC,
Chartered, SMIC, etc. I doubt they will do this, but it would fill
up their plants.

I'm certain it's crossed their minds, and if they're not doing it,
it's for a good reason (like the profit margins aren't high enough).

Or it may be worried that people would think that the reason it is producing
chips for other companies is because it can't fill up those fabs itself.

Yousuf Khan

Robert Myers wrote:
How all this works with Sun? I put Sun's dog and pony show to the
financial sector on while I was doing other things yesterday. I
haven't heard so much upspin since the days of high school pep
rallies. If their apps were so locked in, I don't think they'd sound
so eager.

Well, you can't run your business simply relying on lock-in customers. The
lock-ins will produce a steady state of income, but not enough to run your
business on.

That would all put a strong SELL on Intel, I think. It would be like
betting against the Yankees or the Green Bay Packers in the days of
Vince Lombardi.

My thoughts too. If Intel were to announce that it has the capacity to build
other people's chips, then Wall Street would wonder why it doesn't fill up
all of those chip fabs with its own production?

Yousuf Khan

Yousuf Khan wrote:
Well, you can't run your business simply relying on lock-in
customers. The lock-ins will produce a steady state of
income, but not enough to run your business on.

Really? Do you think Microsoft isn't running profitably?
They currently have a lock-in on PC operating systems.

-- Robert