so Jobs gets screwed by IBM over game consoles, thus Apple-Intel ?

[keith]

First you identify Dell as a high ASP competitor of Intel and now you are
recommending MS as a role model for purchasers of servers :^)

Dell is *not* in any way a competator of INTC, they are INTC's retail arm.
There is no way an hionest person could come to your absurd conclusion.

And the top 500 supercomputers?? How many were Xeon? AMD? Power?

Over half the systetms are IBM. Go figure.

Rephrase:

[...one basket, i.e. Dell... and AMD has a piece...]

better for you?
OK.

BTW you don't seriously think Dell makes its profits from selling $299. PCs
- their Enterprise Division is a *very* big player in server *and*
networking. Michael Dell was voted Network World's "Network Man of the
Year" for 2003 & 2004 by their readers - made no sense to me but there
ya go.

AMD's processor business *is* healthy & profitable - their huge losses from
Spansion dragged them into the red recently... something which they are
trying to correct by dumping their interest.

That may be true. Let's take a look at their quarterly now that they
have eliminated the failing flash business (that they lost to a
competitor).

Flash is a fickle buisness that has nothing to do with CPUs.

However, they are evolving into a one-product company.

....and INtel isn't? They haven't done woderfully in any market other than
CPUs either.

You may well be right about the server market. However, AMD badly needs
manufacturing capability to take advantage of any opportunities for
growth. That is the basic issue I was referring to. If Intel needs added
capacity, somebody reaches into their back pocket, pulls out $3B and
builds a factory in qne year. AMD can't respond that way. My guess is
that within this year Intel will meet or exceed any real and perceived
advantages that AMD64 now has. At that point their manufacturing
advantages will dominate.

Nonsense. AMD can survive, quite well, on the manufacturing that it has
available with a ramp as they gain acceptance. AMD dosn't need to compete
with themselves.

That is why Jobs went with Intel, rather than AMD. He may have seen IBM
to be technologically dominate at one time. He changed his mind. He
could change it again (more easily). But there is nothing known today
that would argue that Intel is not the right choice for Apple.

Utter nonsense.

 

[Robert Myers]

Over half the systetms are IBM. Go figure.

Well, no. Go read the exchange starting with Del's "We're Number 1"
post on comp.arch.

RM

 

[James Arveson]

Dell is *not* in any way a competator of INTC, they are INTC's retail arm.
There is no way an hionest person could come to your absurd conclusion.

Try to follow the conversation if you choose to comment. Mr. Storm explained
that he meant that Dell was a customer of AMD, as well as Intel.

Over half the systetms are IBM. Go figure.

Try to educate yourself on the subject. NONE of the systems are made by
Intel. I wasn't commenting on who made the systems. I was commenting on
which microprocessors were used in them. Over half of the systems (including
many of those sold by IBM) use Intel microprocessors.

Rephrase:
[...one basket, i.e. Dell... and AMD has a piece...]
better for you?
OK.

BTW you don't seriously think Dell makes its profits from selling $299. PCs
- their Enterprise Division is a *very* big player in server *and*
networking. Michael Dell was voted Network World's "Network Man of the
Year" for 2003 & 2004 by their readers - made no sense to me but there
ya go.

AMD's processor business *is* healthy & profitable - their huge losses from
Spansion dragged them into the red recently... something which they are
trying to correct by dumping their interest.

That may be true. Let's take a look at their quarterly now that they
have eliminated the failing flash business (that they lost to a
competitor).

Flash is a fickle buisness that has nothing to do with CPUs.

Another irrelevant and inaccurate comment. We were discussing the viability
of AMD as a business. The flash memory market was a part of that business.
It no longer is, reducing their product mix.

...and INtel isn't? They haven't done woderfully in any market other than
CPUs either.

They seem to be doing OK in the flash memory business. A look at their Q1
report also indicates expanding businesses in wireless and mobile
components. Nevertheless they, too, are dominated by one market segment. The
difficulty AMD has is that they are shrinking their product mix, not
expanding it.

Nonsense. AMD can survive, quite well, on the manufacturing that it has
available with a ramp as they gain acceptance. AMD dosn't need to compete
with themselves.

I don't think so. They can respond to a modest ramp, but they can't take
advantage of a significant market opportunity. If they are to exploit their
present advantage with AMD64, they have to provide product to new customers.
If they don't they will lose the opportunity.

Utter nonsense.

To you. Not to Jobs. That is why you are not a CEO of a large corporation
and he is.

James

 

[keith]

Try to follow the conversation if you choose to comment. Mr. Storm explained
that he meant that Dell was a customer of AMD, as well as Intel.

Bullshit. They may sell some AMDs under the table, but please don't lie
like this. Dell is nothing but an Intel retail outlet.

Try to educate yourself on the subject. NONE of the systems are made by
Intel. I wasn't commenting on who made the systems. I was commenting on
which microprocessors were used in them. Over half of the systems
(including many of those sold by IBM) use Intel microprocessors.

The systems are made by?

 

[James Arveson]

Bullshit. They may sell some AMDs under the table, but please don't lie
like this. Dell is nothing but an Intel retail outlet.

Your issue is with Mr. Storm, not me. It was his claim. You seem to have a
problem following this conversation. it seems to over your head. Why not
drop out?

The systems are made by?

A lot of companies - go look for yourself.
http://www.top500.org/lists/plists.php?Y=2005&M=06

" A total of 333 systems are now using Intel processors. Six months ago
there were 320 Intel-based systems on the list and one year ago only 287.
The second most-commonly used processors are the IBM Power processors (77
systems), ahead of Hewlett-Packard's PA Risc processors (36) and AMD
processors (25). "
http://www.top500.org/news/articles/article_68.php

 

[Tony Hill]

Try to educate yourself on the subject. NONE of the systems are made by
Intel.

For the anal-retentive out there, there actually is one system built
by Intel. ASCI Red at #139 on the list. Built in 1999 it is one of
the oldest systems still on the current list, and it consists of just
shy of 10,000 Pentium Pro processors.

I wasn't commenting on who made the systems. I was commenting on
which microprocessors were used in them. Over half of the systems (including
many of those sold by IBM) use Intel microprocessors.

For those that are curious, the most common processors used are as
follows:

Intel x86 (Xeon/P4/PPro) - 254 systems / 50.8%
Intel Itanium - 79 systems / 15.8%
PowerPC (Power4, Power5, PPC 440, etc.) - 77 systems / 15.4%
HP PA-RISC - 36 systems / 7.2%
AMD Opteron - 25 systems / 5.0%
All others - 29 systems / 5.8%


Of course, if you break it down by percentage of Rmax you get a
slightly different set of numbers:

PowerPC - 36.54%
Intel x86 - 32.69%
Intel Itanium - 14.07%
AMD Opteron - 6.30%
HP PA-RISC - 2.74%
All others - 7.66%


The small number of BlueGene/L systems really skews performance
results in IBM's favor. AMD and our "other" category are the only
other two chip types that delivery system performance above the
others. For AMD this is mainly a case of newer systems (ie there
simply are no 4 or 5 year-old Opteron systems to bring down the
average, unlike eg. PA-RISC), while for the "other" category it is
mostly skewed by a few big NEC SX6 processors, a la Earth Simulator.


Of course, the actual relevance of all of these numbers leaves
something to be desired. Even this list is by no means complete, as
discussed recently in another thread about BMW/Sauber's supercomputer.
That system used a good chunk of 2.2GHz Opteron processors and
probably would have slotted in somewhere around 300-350, but they
never bothered to run the Linpack test and submit the results. There
are MANY more systems out there that are more powerful than a lot of
the systems on this Top500 list. In fact, other than the top 15 - 20
or so, it's really not a very accurate list of the most powerful
supercomputers in the world, even if the only thing you are counting
is Linpack performance (which in itself is not the end-all, be-all
benchmark for HPC stuff, it's just an easy test to run in a controlled
manner which has a fair degree of relevance to many supercomputing
applications).

Another irrelevant and inaccurate comment. We were discussing the viability
of AMD as a business. The flash memory market was a part of that business.
It no longer is, reducing their product mix.

Err, AMD is still in the Flash business at the moment. They are
working on plans to spin off Spansion as an independent company, but
for the time being they are still the majority owner.

They seem to be doing OK in the flash memory business. A look at their Q1
report also indicates expanding businesses in wireless and mobile
components. Nevertheless they, too, are dominated by one market segment. The
difficulty AMD has is that they are shrinking their product mix, not
expanding it.

Intel would probably do well to continue shrinking their product mix,
they've done rather poorly in pretty much all their businesses. Their
flash revenues are less than 1/10th those from processors. To the
best of my knowledge they have never really made any money from that
division, about the best they've managed is roughly break-even.

I don't think so. They can respond to a modest ramp, but they can't take
advantage of a significant market opportunity. If they are to exploit their
present advantage with AMD64, they have to provide product to new customers.
If they don't they will lose the opportunity.

With AMD's current fab they could support about 20-25% of the ~175M
world market for x86 processors (there's a fair degree of variation
depending on just what processors they're selling, dual-core chips
with 2MB of cache need much more fab space than a single-core chip
with 128KB of cache), assuming they were pumping the chips out full
tilt. Right now AMD has between 16 and 17% of the world market. So,
even without any new fabs they do still have some room to grow.

They also seem to be progressing on schedule for volume shipments from
their new fab starting in about a years time. They also have some
agreements with chip foundries to produce processors if demand
requires. Long story short, they can satisfy a fairly large increase
in business.

To you. Not to Jobs. That is why you are not a CEO of a large corporation
and he is.

I'm rather certain that AMD played a major role in Steve Job's choice
of Intel processors. AMD gives them a really solid second-source for
processors should Intel falter. This is one of their major complaints
about their current deal with IBM. After Motorola kind of fell off
the map they were left with ONLY IBM to supply them with PowerPC
chips. While the announcement may have only mentioned Intel
processors and initial systems will probably all use Intel chips, I'm
quite certain that Apple will keep in contact with AMD.

 

[James Arveson]

For the anal-retentive out there, there actually is one system built
by Intel. ASCI Red at #139 on the list. Built in 1999 it is one of
the oldest systems still on the current list, and it consists of just
shy of 10,000 Pentium Pro processors.


For those that are curious, the most common processors used are as
follows:

Intel x86 (Xeon/P4/PPro) - 254 systems / 50.8%
Intel Itanium - 79 systems / 15.8%
PowerPC (Power4, Power5, PPC 440, etc.) - 77 systems / 15.4%
HP PA-RISC - 36 systems / 7.2%
AMD Opteron - 25 systems / 5.0%
All others - 29 systems / 5.8%


Of course, if you break it down by percentage of Rmax you get a
slightly different set of numbers:

PowerPC - 36.54%
Intel x86 - 32.69%
Intel Itanium - 14.07%
AMD Opteron - 6.30%
HP PA-RISC - 2.74%
All others - 7.66%


The small number of BlueGene/L systems really skews performance
results in IBM's favor. AMD and our "other" category are the only
other two chip types that delivery system performance above the
others. For AMD this is mainly a case of newer systems (ie there
simply are no 4 or 5 year-old Opteron systems to bring down the
average, unlike eg. PA-RISC), while for the "other" category it is
mostly skewed by a few big NEC SX6 processors, a la Earth Simulator.


Of course, the actual relevance of all of these numbers leaves
something to be desired. Even this list is by no means complete, as
discussed recently in another thread about BMW/Sauber's supercomputer.
That system used a good chunk of 2.2GHz Opteron processors and
probably would have slotted in somewhere around 300-350, but they
never bothered to run the Linpack test and submit the results. There
are MANY more systems out there that are more powerful than a lot of
the systems on this Top500 list. In fact, other than the top 15 - 20
or so, it's really not a very accurate list of the most powerful
supercomputers in the world, even if the only thing you are counting
is Linpack performance (which in itself is not the end-all, be-all
benchmark for HPC stuff, it's just an easy test to run in a controlled
manner which has a fair degree of relevance to many supercomputing
applications).

There has been some criticism of Linpack as a relevant benchmark for
anything except highly threaded scientific problems. Other types of problems
may be better solved on systems with fewer, more powerful processors than
are used in the Blue Genes machines, for example. But that is always the
problem with benchmarks, I guess.

I am currently working with a company that put together their own Xeon-based
machine for analyzing seismic data for oil exploration. They "claim" their
computer outperforms most of the Top 500, but have no interest in
benchmarking the computer. I expect that there are many other computers like
that.
We were discussing the viability

Err, AMD is still in the Flash business at the moment. They are
working on plans to spin off Spansion as an independent company, but
for the time being they are still the majority owner.

Thanks for the correction.

Intel would probably do well to continue shrinking their product mix,
they've done rather poorly in pretty much all their businesses. Their
flash revenues are less than 1/10th those from processors. To the
best of my knowledge they have never really made any money from that
division, about the best they've managed is roughly break-even.

It's difficult to extract the flash segment of the profits as they are
buried in the Mobility group. But you are probably not far off. As I said,
they too are effectively a "one product" company. From Apple's perspective
that is good news. The one product happens to be the product that Apple is
interested in.

to new customers.

With AMD's current fab they could support about 20-25% of the ~175M
world market for x86 processors (there's a fair degree of variation
depending on just what processors they're selling, dual-core chips
with 2MB of cache need much more fab space than a single-core chip
with 128KB of cache), assuming they were pumping the chips out full
tilt. Right now AMD has between 16 and 17% of the world market. So,
even without any new fabs they do still have some room to grow.

That capacity of 40M chips/yr is consistent with the numbers I have seen -
about 1M/ Athlons/wk from 5K 200 mm wafers coming out of fab 30. That
converts to about 60% yield for a 100 mm die -not too impressive. My numbers
are 2 yrs old and, hopefully, they have improved by now.

The assumption was that AMD would exploit their opportunity in the high end
part of the business. I would anticipate that a ramp of their dual core
processors would put a big hole in any capacity they might have. If they are
having any yield problems, the impact would be even greater. I could do a
yield analysis, but it would likely give a number that would be
prohibitively low.

tel, rather than AMD. He may have seen IBM

I'm rather certain that AMD played a major role in Steve Job's choice
of Intel processors. AMD gives them a really solid second-source for
processors should Intel falter. This is one of their major complaints
about their current deal with IBM. After Motorola kind of fell off
the map they were left with ONLY IBM to supply them with PowerPC
chips. While the announcement may have only mentioned Intel
processors and initial systems will probably all use Intel chips, I'm
quite certain that Apple will keep in contact with AMD.

I agree. Now Apple finally has two potential suppliers who really are in the
microprocessor business. I would also expect that the business arrangement
involved a commitment on the part of both parties - Intel to deliver and
Apple to buy. If Intel falters, AMD would have a shot. I would be surprised
if he spread his resources over two suppliers, however. He had enough
headaches dealing with IBM/Moto.

There is another factor that may have influenced Jobs. He was likely highly
impressed by IBM's research capability and, to a lesser extent, that of
Motorola. Intel's research efforts, in the areas of interest to Jobs, at
least match their capability and are all under one corporate roof. In
contrast, AMD does not have a highly visible research effort.

James

 

[George Macdonald]

First you identify Dell as a high ASP competitor of Intel and now you are
recommending MS as a role model for purchasers of servers :^)

No, you misread or mispunctuated my comment - see below. As for M$ as a
role model, they *do* have a lot of err, customers who have "chosen" their
OS & software and have not been quiet about their enthusiasm for AMD64.
Nobody can say "Gee I dunno if it's gonna work". Check the numbers on the
MSN farm: reduction from 250 computers to 25 IIRC.

And the top 500 supercomputers?? How many were Xeon? AMD? Power?

What does it matter?... numbers of available systems doesn't really count.
When an outfit like DALCO can put together a MP HPC system for Sauber, from
racks, bits & pieces, the whole equation changes. The fact is that this is
a market where AMD did not have that high ASP presence... now it does and
in a *big* way.

Rephrase:

[...one basket, i.e. Dell... and AMD has a piece...]

better for you?
OK.

BTW you don't seriously think Dell makes its profits from selling $299. PCs
- their Enterprise Division is a *very* big player in server *and*
networking. Michael Dell was voted Network World's "Network Man of the
Year" for 2003 & 2004 by their readers - made no sense to me but there ya
go.

AMD's processor business *is* healthy & profitable - their huge losses from
Spansion dragged them into the red recently... something which they are
trying to correct by dumping their interest.

That may be true. Let's take a look at their quarterly now that they have
eliminated the failing flash business (that they lost to a competitor).

Eliminated? AFAIK it's still on the block and still a "monument sur le
dos".

However, they are evolving into a one-product company.

And Intel is not?

You may well be right about the server market. However, AMD badly needs
manufacturing capability to take advantage of any opportunities for growth.
That is the basic issue I was referring to. If Intel needs added capacity,
somebody reaches into their back pocket, pulls out $3B and builds a factory
in qne year. AMD can't respond that way. My guess is that within this year
Intel will meet or exceed any real and perceived advantages that AMD64 now
has. At that point their manufacturing advantages will dominate.

Again you seem to be short on facts: in the next year, if everything plays
out, AMD will have two additional production facilities: their own Fab 36
is already well into 300mm 65nm prototype production and "going well" by
all accounts; Chartered needs the work badly so is at least highly
motivated. As for Intel catching up technically, that is not going to
happen until P4 is dead... and even then, you think AMD is twiddling their
developmental thumbs?

That is why Jobs went with Intel, rather than AMD. He may have seen IBM to
be technologically dominate at one time. He changed his mind. He could
change it again (more easily). But there is nothing known today that would
argue that Intel is not the right choice for Apple.

Jobs went with Intel because Intel sees opportunities to nix the wayward M$
and re-establish their proprietary architecture niche. It's gonna be a
long road but as long as Apple does what it is told........ it's a *BIG*
maybe.

Oh, AMD said they'd never even talked to Apple and were "too busy"
anyway.:-[]

 

[George Macdonald]

A reward for selling some of the crappiest network switches in
the world (PowerConnect)?

Really?.... Love it!

AMD's processor business *is* healthy & profitable - their huge losses from
Spansion [sic] dragged them into the red recently... something which they are
trying to correct by dumping their interest.

They can dump their interest in my bank account if that will
help them.

Why? The flash market, it would appear, is intrinsically a boom 'n' bust
cyclical play, judging by events of the recent past. When you see the
number of devices, it has to be profitable but............

 

[George Macdonald]

Try to follow the conversation if you choose to comment. Mr. Storm explained
that he meant that Dell was a customer of AMD, as well as Intel.

WTF is Mr. Storm? Nobody has claimed that Dell is a customer of AMD,
though they *will* apparently do an AMD system for you if you ask loudly
but not loudly enough for Intel to hear.

They seem to be doing OK in the flash memory business.

The entire flash business is in the pan - *nobody* is "doing OK" with it.
Intel lost a bundle of money dumping flash. If one assumes that they have
clever people, this was an overt, calculated move to sink Spansion.

A look at their Q1
report also indicates expanding businesses in wireless and mobile
components. Nevertheless they, too, are dominated by one market segment. The
difficulty AMD has is that they are shrinking their product mix, not
expanding it.

Those -- wireless, mobile -- are all "support products" to sell processors
as part of their "platformization" effort. Intel hasn't made any $$
directly from them. Can you say "Centrino"? Oh they've also given up on
their LCoS venture because they obviously could not compete there but it
too was "support".

 

[Robert Myers]

There has been some criticism of Linpack as a relevant benchmark for
anything except highly threaded scientific problems.

I would have thought the FFT was a highly-threaded scientific problem.
Blue Gene won't use about more than 512 processors effectively in doing
an FFT, no matter how many you wire together, if IBM's own results are
to be believed. The problem with linpack is that it places minimal
demands on global communication. It is just about perfect to make a
machine like Blue Gene, which is weak in the global communication
bandwidth department, look good.

I've done my public bitching about both Linpack and Blue Gene.
Linpack, IMHO, measures almost nothing, except that the machines are
connected and you know how to program them. Other than that, you might
as well declare a warehouse at Dell the most powerful supercomputer in
the world from the number of boxes it contains.

Blue Gene is another story. There are important classes of problems
that it cannot do well, but there are enough problems at which it can
do well that IBM is making sales. The fact that global communication
demands are so modest, which is what allows Blue Gene to be marketed
successfully, also means that you might just as well hook up a bunch of
boxes with gigabit ethernet, and that's what many do, as I'm sure you
know.

What Blue Gene really has going for it, other than a nice-looking
enclosure, is low power consumption: a property that I expect will come
to dominate the discussion of supercomputers, or at least it should.

Other types of problems
may be better solved on systems with fewer, more powerful processors than
are used in the Blue Genes machines, for example. But that is always the
problem with benchmarks, I guess.

Blue gene is well-suited to problems that mirror the machine's layout:
problems on a fixed rectilinear grid with limited non-local
communication.

RM

 

[Keith R. Williams]

Your issue is with Mr. Storm, not me. It was his claim. You seem to have a
problem following this conversation. it seems to over your head. Why not
drop out?

A lot of companies - go look for yourself.
http://www.top500.org/lists/plists.php?Y=2005&M=06

" A total of 333 systems are now using Intel processors. Six months ago
there were 320 Intel-based systems on the list and one year ago only 287.
The second most-commonly used processors are the IBM Power processors (77
systems), ahead of Hewlett-Packard's PA Risc processors (36) and AMD
processors (25). "
http://www.top500.org/news/articles/article_68.php

You do like selective quoting, eh? From the same site:

"IBM continues to establish itself as the dominant vendor of
supercomputers with now more than half of the list (51.8 percent)
carrying its label. The Blue Gene architecture helped IBM to gain
a similar standing at the very top of the list, where now six of
the TOP10 systems are from IBM, five of these being Blue Gene
systems."

 

[Del Cecchi]

keith wrote:




Well, no. Go read the exchange starting with Del's "We're Number 1"
post on comp.arch.

RM

Actually 259 of the 500 were IBM systems, according to my information.
58 out of the top 100, and 977 out of 1687 TeraFlops.

The IBM systems include both Power and non-Power (Intel, AMD (I think))
systems and they use a variety of interconnect:Myrinet, Quadrics,
InfiniBand, Gigabit Ethernet, Federation, Colony and BlueGene fabric.

Clearly IBM is not dogmatic.

 

[Robert Myers]

Actually 259 of the 500 were IBM systems, according to my information.
58 out of the top 100, and 977 out of 1687 TeraFlops.

I could have sworn the discussion was concerning microprocessors.

As long as its possible to make money integrating systems, I'm sure
that IBM will be integrating them, and now that super==big==lots of
hardware under one roof, I'm sure that IBM will continue to be a player
in the arena of "super"computers.

If it gets too expensive for IBM to integrate systems with hardworking
US engineers in forbiddingly remote US locations, it will simply move
the work to Bangalore, or some similar location, to the extent that it
hasn't already. Thus, I am confident that IBM will continue to be a
purveyor of high profile clusters to high profile customers.

As to IBM's presence in the microprocessor arena, I am much less
certain, and the fact that IBM has to count the list multiple ways to
downplay the importance of Intel pretty much tells the trend.

That is to say, whatever Keith intended to say about IBM and
microprocessors (the topic at hand), it was either incorrect,
misleading, off-topic, or said the exact opposite of what he intended.
Take your pick.

RM

 

[Del Cecchi]

I could have sworn the discussion was concerning microprocessors.

As long as its possible to make money integrating systems, I'm sure
that IBM will be integrating them, and now that super==big==lots of
hardware under one roof, I'm sure that IBM will continue to be a player
in the arena of "super"computers.

If it gets too expensive for IBM to integrate systems with hardworking
US engineers in forbiddingly remote US locations, it will simply move
the work to Bangalore, or some similar location, to the extent that it
hasn't already. Thus, I am confident that IBM will continue to be a
purveyor of high profile clusters to high profile customers.

As to IBM's presence in the microprocessor arena, I am much less
certain, and the fact that IBM has to count the list multiple ways to
downplay the importance of Intel pretty much tells the trend.

That is to say, whatever Keith intended to say about IBM and
microprocessors (the topic at hand), it was either incorrect,
misleading, off-topic, or said the exact opposite of what he intended.
Take your pick.

RM

Why did you delete my sentence "The IBM systems include both Power and
non-Power (Intel, AMD (I think)) systems"? Was it to get in a crack
about "I could have sworn the discussion was concerning microprocessors."?

And where did the stuff about Bangalore come from? Don't you want folks
in Asia to have good jobs too?

 

[James Arveson]

You do like selective quoting, eh? From the same site:

"IBM continues to establish itself as the dominant vendor of
supercomputers with now more than half of the list (51.8 percent)
carrying its label. The Blue Gene architecture helped IBM to gain
a similar standing at the very top of the list, where now six of
the TOP10 systems are from IBM, five of these being Blue Gene
systems."

You are right. I didn't quote the who article. I quoted the part that
addressed the issue at hand. We all are aware of Blue Gene's status and its
significance. We also are aware of the fact that IBM shipped the majority of
those systems. My point to the other poster was that this fact is consistent
with the fact that about 2/3 of the systems use Intel processors. Perhaps
you missed that part of the discussion.

James

 

[Flint]

Intel's got more pressing issues to deal with now rather than
pipelines. Like how to match AMD's Direct Connect Architecture
(Hypertransport and internal memory controller).

Yousuf Khan

Give that man a cigar!

AMD's on-die memory controller just keeps getting better too with each
revision of Athlon64, especially since they went to the 90nm SOI
"Winnie" cores, and again for the Venice/San Diego cores, and again for
the X2 dual cores. If Apple wants to go x86, they should have gone
with AMD. Like Apple really needs to ruin their beautiful cases
with an "Intel inside" sticker anyway...

-Rick

 

[Yousuf Khan]

That capacity of 40M chips/yr is consistent with the numbers I have seen -
about 1M/ Athlons/wk from 5K 200 mm wafers coming out of fab 30. That
converts to about 60% yield for a 100 mm die -not too impressive. My numbers
are 2 yrs old and, hopefully, they have improved by now.

Well, yield does not equal capacity. About two years ago, AMD's
marketshare had fallen to about 15%, had nothing to do with how much
they could supply, had everything to do with how much demand there was
for their product. Around that time, Intel had successfully gotten all
top tier OEMs to exclude AMD chips. So their main sales went to only
whiteboxes during that time. So it was more than likely that AMD
reduced wafer starts during that time.

Also two years ago, that was at the start of AMD's transition to
K8-series processors from K7-series. They were already at 130nm node,
so the average K7 series was maybe 100 sq.mm., but the new K8's were
approaching 200 sq.mm. Depending on the product mix of K7 and K8, the
average die size was not 100 sq.mm, but a bit more than that, closer to
150 sq.mm, assuming a 50:50 K7:K8 product mix. Then about a year ago
they made the switch to 90nm which halved the size of the K8's back
down to the 100 sq.mm mark. They kept the old K7's at 130nm so they
remained at 100 sq.mm as well, so the 100 sq.mm average is probably
more accurate from a year ago to present than from two years ago.

Yousuf Khan

 

[Robert Myers]

Why did you delete my sentence "The IBM systems include both Power and
non-Power (Intel, AMD (I think)) systems"? Was it to get in a crack
about "I could have sworn the discussion was concerning microprocessors."?

Because I was continuing my response to keith's misleading post, rather
than responding to yours, which was clear.

And where did the stuff about Bangalore come from? Don't you want folks
in Asia to have good jobs too?

The "stuff about Bangalore" was to indicate that, almost no matter what
happens to the economics of systems integration, IBM will hang in
there, one way or another, even if it means hanging its U.S. and
European workforce out to dry. And it probably will mean that.

IBM, naturally, will have it both ways, arguing that the U.S. national
security estblishment needs to buy from it to keep American technology
tip-top while shipping jobs overseas.

That is to say, IBM has reasonbly straightforward strategies available
to stay in the system integration business, almost no matter how the
market pressure comes down.

I don't see any similar strategy keeping IBM in the microprocessor
business.

Do I want people in Asia to have good jobs? As long as everyone
understands that the only technical jobs left in the U.S. will be ones
requiring a security clearance.

RM

 

[Tony Hill]

That capacity of 40M chips/yr is consistent with the numbers I have seen -
about 1M/ Athlons/wk from 5K 200 mm wafers coming out of fab 30. That
converts to about 60% yield for a 100 mm die -not too impressive. My numbers
are 2 yrs old and, hopefully, they have improved by now.

The die sizes for AMD's current processors range from about 85mm^2 for
their Sempron chips up to about 199mm^2 for their new dual-core chips.
Currently they're split somewhere in the middle of the transition from
130nm to 90nm production (I believe they're past the half-way point in
this transition). Yield numbers are a rather tightly guarded secret,
though ranges from 60-80% would be typical. I figure that a 35M to
45M would be a decent rough estimate of their total capacity at Fab
30.

The assumption was that AMD would exploit their opportunity in the high end
part of the business. I would anticipate that a ramp of their dual core
processors would put a big hole in any capacity they might have. If they are
having any yield problems, the impact would be even greater. I could do a
yield analysis, but it would likely give a number that would be
prohibitively low.

Unless you've got a lot more info than the rest of us mere mortals,
chances are that any yield analysis would be rather pointless anyway.
As for the dual-core chips, they are indeed fairly large. On the flip
side, so was the original Athlon64/Opteron when it was first released.
The first Athlon64/Opteron chips were all built from the same die,
weighing in at 193mm^2 on a 130nm production line. The new dual-core
Opteron and Athlon64 X2 chips have a 199mm^2 die on a 90nm production
line. Big chips and, not surprisingly, a big price tag to match.

I agree. Now Apple finally has two potential suppliers who really are in the
microprocessor business. I would also expect that the business arrangement
involved a commitment on the part of both parties - Intel to deliver and
Apple to buy. If Intel falters, AMD would have a shot. I would be surprised
if he spread his resources over two suppliers, however. He had enough
headaches dealing with IBM/Moto.

Very true. Apple has even less incentive to use alternative part
suppliers than Dell does. However I would think that they like to
keep AMD around as a sort of safety blanket.

There is another factor that may have influenced Jobs. He was likely highly
impressed by IBM's research capability and, to a lesser extent, that of
Motorola. Intel's research efforts, in the areas of interest to Jobs, at
least match their capability and are all under one corporate roof. In
contrast, AMD does not have a highly visible research effort.

I sure one could argue one way or the other about the visibility of
AMD's research effort, but the fact of the matter is that it most
definitely does exist. In one extremely crude measure of research,
the number of patents issued to each company, AMD has been
consistently leading Intel for several years now. How this translates
into real research is another matter, but it at least shows that AMD
is doing *something*