AMD market share if they could sell all they make?

  • Thread starter Thread starter Neil Maxwell
  • Start date Start date
N

Neil Maxwell

What would AMD's market share be if they could sell all the CPUs they
make right now, taking away Intel sales in the process? Anybody have
this kind of data?

Just curious about their short-term market share potential (that is,
without new fabs or fab conversions).


Neil Maxwell - I don't speak for my employer
 
What would AMD's market share be if they could sell all the CPUs they
make right now, taking away Intel sales in the process? Anybody have
this kind of data?

Just curious about their short-term market share potential (that is,
without new fabs or fab conversions).


Neil Maxwell - I don't speak for my employer

Good thing! Tell you're employer they've f'd up their business
big time! ;-)
 
What would AMD's market share be if they could sell all the CPUs they
make right now, taking away Intel sales in the process? Anybody have
this kind of data?

Uhh, if you're e-mail address is to be believed, then you work for a
company that probably knows much better than nearly anyone in this
newsgroup! I'm SURE that Intel has someone with their eyes on AMD who
has done all kinds of calculations on just how much of the market
share they could take.
Just curious about their short-term market share potential (that is,
without new fabs or fab conversions).

Rough guess, 20-30%. It depends on a lot of factors though, not the
least of which simply being how big the market is. The market this
year looks like it should be a good 10-15% larger than it was last
year, but AMD's fab capacity has not increased at all. What's more,
they're now making a much larger processor die (Opteron/Athlon64), so
they can produce fewer chips.
 
Tony Hill said:
year, but AMD's fab capacity has not increased at all. What's more,
they're now making a much larger processor die (Opteron/Athlon64), so
they can produce fewer chips.

Have the 512k cache Clawhammers actually come online as their own die (as
opposed to a 1M cache with half disabled or nonfunctional)? Shouldn't that
cut down the die size a bit?
 
Nate said:
Have the 512k cache Clawhammers actually come online as their own die
(as opposed to a 1M cache with half disabled or nonfunctional)?

As far as I understand,

Clawhammer = 1 MB L2 cache. Half is disabled in some Athlons to
improve yield.

I think you meant Newcastle (512 KB L2 cache @ 130 nm) and Winchester
(512 KB L2 cache @ 90 nm).

Time for back of the envelope guesstimates.

Clawhammer
1 MB L2 cache
105.9 million transistors
193 mm^2 @ 130 nm

512 KB cache =~ 25 million transistors

Newcastle
512 KB L2 cache
80 million transistors
150 mm^2 @ 130 nm

Winchester (just a die shrink ??)
512 KB L2 cache
80 million transistors
85 mm^2 @ 90 nm

Winchester should support SSE3.
 
Uhh, if you're e-mail address is to be believed, then you work for a
company that probably knows much better than nearly anyone in this
newsgroup! I'm SURE that Intel has someone with their eyes on AMD who
has done all kinds of calculations on just how much of the market
share they could take.

Yes, you can believe it, and yes, I'm also sure there are people here
who know the answers to such things. If you know anyone in the
industry, you know that this type of market analysis data is not
routinely distributed to the rank-and-file in much of any company I've
worked at (AMD included). I'm a simple fab manufacturing engineer in
a deep and narrow niche (and flash to boot).

Hence, my question (an unbiased one, IMO) in a newsgroup full of
industry-watchers. I'll admit to a vested interest in the long-term
results, but I have no insider knowledge, and no desire for biased
spin on either side.
Rough guess, 20-30%. It depends on a lot of factors though, not the
least of which simply being how big the market is. The market this
year looks like it should be a good 10-15% larger than it was last
year, but AMD's fab capacity has not increased at all. What's more,
they're now making a much larger processor die (Opteron/Athlon64), so
they can produce fewer chips.

The goal of the question is to look at supply and demand issues (and
related income/margin issues) in perspective, prompted by an earlier
post on short-term market share. In the past, AMD has hit 20%+ market
share, though they appear to be in the 15% range currently based on Q1
data. It appears they have dropped market share but gained
profitability by being less aggressive on pricing cuts (which is good
for both companies, IMO, and may be related to Sanders' lessening
control). Without profits, they have no long-term expansion
capability.

According to reports, AMD's ASP is rising, but it's not clear to me
whether that's because they're selling a larger percentage of high-end
CPUs or discounting the entire line less. I'm sure someone here knows
more detail.

The thought is that, regardless of Intel's perceived screwups and
AMD's performance benefits, AMD's upside and market share impact is
somewhat limited by their mid-term capacity, while Intel's large
manufacturing base and ongoing transition to increased 300mm capacity
allows them extra production flexibility and continuing margin
pressures. As I see it, it's physically impossible for AMD to pick up
more than 10-15% of Intel's CPU business, and if they picked this up
in the lower-end CPUs, it would hurt them more than it would hurt
Intel.

I'm aware that this is holy war material, and I don't care to go
there. I'm just looking for facts and educated guesses to base my own
long-term plans on.

Personally, I'm all for a healthy AMD; as I said, I think it's good
for both companies.


Neil Maxwell - I don't speak for my employer
 
Grumble said:
As far as I understand,
Clawhammer = 1 MB L2 cache. Half is disabled in some Athlons to
improve yield.

I think you meant Newcastle (512 KB L2 cache @ 130 nm)

My misunderstanding; I guess I meant Newcastle -- what I meant was the "512k
cache Athlon 64 where it wasn't just disabling half of the 1mb cache on the
older die."

I thought that was ClawHammer and that that was the distinction in cores
between the 1mb and 512k cache ones. Because I thought that the 3200+/3400+
were the same core as the Opterons/A64FX -- just with a socket 754 carrier
and as such gimpage of the extra memory channel.
Winchester (just a die shrink ??)
512 KB L2 cache
80 million transistors
85 mm^2 @ 90 nm

Winchester should support SSE3.

Assuming everything goes well with the process shrink, that should help push
the A64 prices down, right? If I understand correctly, half the die size ==
more than 2X the yield, all other things being equal...
 
The goal of the question is to look at supply and demand issues (and
related income/margin issues) in perspective, prompted by an earlier
post on short-term market share. In the past, AMD has hit 20%+ market
share, though they appear to be in the 15% range currently based on Q1
data. It appears they have dropped market share but gained
profitability by being less aggressive on pricing cuts (which is good
for both companies, IMO, and may be related to Sanders' lessening
control). Without profits, they have no long-term expansion
capability.

AMD's market share figures have been reasonably stable for quite about
5 years now. They've varied a bit, but most been in the 15-20% range.
According to reports, AMD's ASP is rising, but it's not clear to me
whether that's because they're selling a larger percentage of high-end
CPUs or discounting the entire line less. I'm sure someone here knows
more detail.

From what I've seen it's mainly that they are selling a larger
percentage of high-end CPUs. They still sell a lot of dirt-cheap
AthlonXP and even Duron chips. Just have a look at the prices that
AMD themselves list for an AthlonXP 2500+, only $79, and you KNOW that
HPaq is getting them for a lot less. On the other hand, the AthlonXP
3000+ and 3200+ seem to still be selling reasonably well and the
Athlon64/Opteron line has been fairly successful, and those are all
high-margin parts.

Where AMD has been hurting a bit, from what I've seen, is in the
mobile space. The AthlonXP-M is actually a rather competitive
product. It's performance and power consumption are well within
striking distance of the Pentium-M, but the Centrino marketing
campaign at Intel has been VERY effective. As a result AMD has had to
compete with the Mobile Celeron chips instead. The AthlonXP-M is a
MUCH better mobile processor than the Mobile Celeron (and even better
than the new Celeron-M), but AMD has still had to compete on price to
get anywhere in this market.
The thought is that, regardless of Intel's perceived screwups and
AMD's performance benefits, AMD's upside and market share impact is
somewhat limited by their mid-term capacity,

Somewhat limited maybe. But I don't think that is the primary
limiting factor. The biggest factor working against AMD right now
seems to be Intel's marketing department, who seem able to more than
make up for any faults in the products themselves.
while Intel's large
manufacturing base and ongoing transition to increased 300mm capacity
allows them extra production flexibility and continuing margin
pressures. As I see it, it's physically impossible for AMD to pick up
more than 10-15% of Intel's CPU business,

That's probably fairly accurate. Note though that AMD does have a
fairly large new fab coming on-line in 2006, so if you're looking at
this for a somewhat more long-term deal than AMD could end up with a
much larger chunk of the market.
and if they picked this up
in the lower-end CPUs, it would hurt them more than it would hurt
Intel.

I doubt it. The economics of computer processors tend to emphasize
volume over almost everything else. Even if they're only selling
AthlonXP chips for $35 or $40 than AMD is probably still making a
profit on a per-chip basis. The real costs are all in R&D, and those
are mostly sunk costs when it comes to the low-end chips.

The thing that might hurt AMD would be to take away the low-end of the
market from Intel but lose the high-end in the process. They might
end up with a larger chunk of market share in the process, but they
wouldn't be making the high margins on the new processors where R&D is
still an on-going expense.
I'm aware that this is holy war material, and I don't care to go
there. I'm just looking for facts and educated guesses to base my own
long-term plans on.

If you're talking about long-term plans in terms of stocks and such, I
personally wouldn't want to get too much into either company. However
if you've got some kind of stock options for Intel, they aren't going
anywhere any time soon. Despite any advantages that AMD has shown,
Intel still has pretty much a strangle-hold on the business PC market.
Dell sells Intel chips exclusively and even HPaq sells only a very
small number of AMD-based business systems. Same goes for IBM and
Toshiba in the business laptop market.
Personally, I'm all for a healthy AMD; as I said, I think it's good
for both companies.

That much is certainly true.
 
Neil Maxwell said:
What would AMD's market share be if they could sell all the CPUs they
make right now, taking away Intel sales in the process? Anybody have
this kind of data?

Just curious about their short-term market share potential (that is,
without new fabs or fab conversions).

Last I looked, and that was eras ago in this business, was in the 98-2000
frame during AMD's last successful push through Intel's territory (i.e. K7
Athlon days). At that time, they were just transitioning from 250nm Athlons
to 180nm Athlons and Durons; the Athlons with 256KB L2 were about 105
sq.mm., and the Durons with 64KB L2 were about a paltry 85 sq.mm. Based on
those dimensions it was estimated that AMD could supply about 30-40mln
processors to the market per year, depending on the mix of processors.
Assuming a 100mln x86 processor market, that would be 30-40%. They pushed up
past 20% and were approaching 25% before they were stopped by Intel. So they
still had some extra capacity to spare if they could've moved on without any
resistance.

Now they've been pushed back to 15% marketshare. At 130nm, their newer 512KB
Athlon XPs I think are about 120 sq.mm, but their 256KB Athlon XPs are about
95 sq.mm. Depending on the mix, I think these are only good for supplying
between 20-25mln processors. But now with the mix-in of the 64-bit 1MB L2
cores, that calculation may have gone down one more time. I'm not sure how
much the 64-bit cores size is, is it 160 sq.mm. or 190 sq.mm? Again, I'd say
depending on the mix, the upper limit is still 25 mln processors (assuming
only 256KB XPs are sold), but the lower limit might have gone down to 18 mln
from 20 mln (assuming only 64-bit cores are sold). AMD is also working on a
512KB L2 version of the 64-bit core which I assume will be only 140 sq.mm.
But then again they also have plans for a dual-core 64-bit core with 2MB L2
cache. This is making my head hurt.

I think once they transition to 90 nm, the acreage on each chip will be
halved.

Yousuf Khan
 
Tony Hill said:
Where AMD has been hurting a bit, from what I've seen, is in the
mobile space. The AthlonXP-M is actually a rather competitive
product. It's performance and power consumption are well within
striking distance of the Pentium-M, but the Centrino marketing
campaign at Intel has been VERY effective. As a result AMD has had to
compete with the Mobile Celeron chips instead. The AthlonXP-M is a
MUCH better mobile processor than the Mobile Celeron (and even better
than the new Celeron-M), but AMD has still had to compete on price to
get anywhere in this market.

Yeah, that's for sure. I just bought a cheap little XP-M ultralight notebook
for a relative recently. I couldn't believe all of the features it had
built-in: 3-USB 2.0 ports, 1-100 Mbps Ethernet, 1-56K dialup, and 1-11 Mbps
WiFi-B. All for about Cdn$1500 which works out to US$1100. Which is Celeron
notebook territory. In fact there were Celerons right next to it with less
features and a heavier footprint.

Yousuf Khan
 
On Sat, 15 May 2004 14:23:54 -0400, Tony Hill

<snip>....

Thanks for the input. More food for thought... It's not easy to get
objective analysis on such issues, as so many people attach lots of
emotional baggage to these corporate competitions.


Neil Maxwell - I don't speak for my employer
 
On Sat, 15 May 2004 14:23:54 -0400, Tony Hill

<snip>....

Thanks for the input. More food for thought... It's not easy to get
objective analysis on such issues, as so many people attach lots of
emotional baggage to these corporate competitions.


Neil Maxwell - I don't speak for my employer

Here is one thing nobody mentioned - outsourcing. If there ever would
be the sort of demand for AMD chips that could translate into real
buck$, AMD could easily sign up some foundries to fab the
Athlons/Opterons. The first name that comes to mind is IBM. Since
IBM shares with AMD their semiconductor technology, it would be
relatively easy for IBM to ramp up production. Not something
completely new for IBM - they used to make chips for Cyrix. With
IBM's capacity the AMD/IBM share could go up all the way to 100%
(after all, who in their right mind would want any P4 derivative these
days?), but don't worry Neal - your coworkers from marketing dept. and
faithful Dell will never let it happen ;-)
 
Here is one thing nobody mentioned - outsourcing. If there ever would
be the sort of demand for AMD chips that could translate into real
buck$, AMD could easily sign up some foundries to fab the
Athlons/Opterons. The first name that comes to mind is IBM. Since
IBM shares with AMD their semiconductor technology, it would be
relatively easy for IBM to ramp up production. Not something
completely new for IBM - they used to make chips for Cyrix. With
IBM's capacity the AMD/IBM share could go up all the way to 100%
(after all, who in their right mind would want any P4 derivative these
days?), but don't worry Neal - your coworkers from marketing dept. and
faithful Dell will never let it happen ;-)

Actually no, according to a cross-licensing agreement between Intel and AMD,
AMD gets to use the x86 instruction set from Intel for a "low" royalty
payment (possibly even zero), just so long as AMD produces it's chips at its
own plants. AMD can produce its chips at outside fabs, but it has to pay
"high" royalties to Intel for it (maybe something like 20%, I'm not sure).
It can produce chips at outside fabs for low royalty, just so long as AMD
owns a part of those plants; the percentage of low-royalty chips equals the
AMD ownership percentage of the plant. If for example AMD owns 20% of a
plant, then upto 20% of the chips coming out of it can be AMD chips. If AMD
produces more than this amount from this plant, then it has to pay the
higher royalties on the portion above 20%.

Amongst other items in their cross-licensing agreement, if AMD is ever sold
to a third party (i.e. a third party owns more than 50% of AMD), then Intel
can terminate AMD's license including the x86 license! That's why AMD has
resisted going fabless all of this time, producing at an outside fab will
cost it a lot of money in royalties to Intel. So far AMD has given some
manufacturing over to contract fabs for minor chips, like motherboard
chipsets, and the like.

Yousuf Khan
 
On Tue, 18 May 2004 08:55:30 GMT, "Yousuf Khan"

....snip...
Actually no, according to a cross-licensing agreement between Intel and AMD,
AMD gets to use the x86 instruction set from Intel for a "low" royalty
payment (possibly even zero), just so long as AMD produces it's chips at its
own plants. AMD can produce its chips at outside fabs, but it has to pay
"high" royalties to Intel for it (maybe something like 20%, I'm not sure).
It can produce chips at outside fabs for low royalty, just so long as AMD
owns a part of those plants; the percentage of low-royalty chips equals the
AMD ownership percentage of the plant. If for example AMD owns 20% of a
plant, then upto 20% of the chips coming out of it can be AMD chips. If AMD
produces more than this amount from this plant, then it has to pay the
higher royalties on the portion above 20%.

Amongst other items in their cross-licensing agreement, if AMD is ever sold
to a third party (i.e. a third party owns more than 50% of AMD), then Intel
can terminate AMD's license including the x86 license! That's why AMD has
resisted going fabless all of this time, producing at an outside fab will
cost it a lot of money in royalties to Intel. So far AMD has given some
manufacturing over to contract fabs for minor chips, like motherboard
chipsets, and the like.

Yousuf Khan
Correct me if I am wrong, but I thought IBM should have their own
cross-licensing agreement with INTC, including x86.
 
AMD ownership percentage of the plant. If for example AMD owns 20% of a
plant, then upto 20% of the chips coming out of it can be AMD chips. If AMD
produces more than this amount from this plant, then it has to pay the
higher royalties on the portion above 20%.

I thought AMD said it was 30%, at least thats the last number I rem.
Ed
 
Correct me if I am wrong, but I thought IBM should have their own
cross-licensing agreement with INTC, including x86.

They might, so if AMD were to use IBM as its fab, IBM's cross-license would
apply rather than AMD's. However, it's not certain how much of the x86
license IBM is granted access to. Maybe it would only cover upto the
technology present with the 386 generation (last time IBM made x86's of its
own), so maybe things like MMX through to SSE won't be covered?

Regardless, even if all x86 extension technologies are covered in IBM's
agreement, they may have to pay a royalty for them to Intel. That is, AMD
might have a lower royalty agreement with Intel than IBM. So Opterons built
at an IBM plant might cost more than ones built in AMD plants simply for the
royalty issues.

IBM could cross license some of its process manufacturing technologies to
Intel in return for reduced royalties on x86 technology, but since IBM
doesn't make x86 chips anymore, there may be no incentive for IBM to get an
updated x86 license. And I think IBM considers its process manufacturing
technologies much more valuable than x86 technology in its competition
against Intel.

AMD and Intel being in the same business sector probably have a lot more
relevant stuff to trade with each other.

Yousuf Khan
 
Here is one thing nobody mentioned - outsourcing. If there ever would
be the sort of demand for AMD chips that could translate into real
buck$, AMD could easily sign up some foundries to fab the
Athlons/Opterons. The first name that comes to mind is IBM. Since
IBM shares with AMD their semiconductor technology, it would be
relatively easy for IBM to ramp up production. Not something
completely new for IBM - they used to make chips for Cyrix. With
IBM's capacity the AMD/IBM share could go up all the way to 100%
(after all, who in their right mind would want any P4 derivative these
days?), but don't worry Neal - your coworkers from marketing dept. and
faithful Dell will never let it happen ;-)

Outsourcing production of large complex chips to an external fab is an
extremely non-trivial task, particularly as feature sizes get smaller.
There are so many things that can go wrong - process cliffs everywhere
- and many of these aren't obvious until end of line testing. Even
starting up a new internal fab where you have complete control over
everything is risky business.

In addition, joint ventures/foundry deals are notoriously tricky. IBM
is not out for AMD's best interests, and vice-versa. Most
corporations will pull the rug out from under a partner for short-term
gain - there's a long history of this. AMD was working a joint
venture 300mm fab build with UMC some years back; did anything become
of that? They were going to use this to get to 30% market share by
this year.

Regardless, AMD is apparently unable to sell all they have the
capacity to make now, so the possibility of them being able to take
the bulk of the market with outsourcing is somewhat slim.

As for who would want a P4 - to be honest, I don't see a big
difference in end results between the chips, and the public really
doesn't care about anything but how it runs. The benchmarks show each
company leapfrogging the other routinely, and an extra 10 FPS in UT2K4
just doesn't mean anything to me. I'm upgrading to a P4 2.8G on a
game box now, and the difference between the P4 2.8 and the Athlon
2800+ looks like it's about $50. Not a big difference in the total
cost, though applying it to a better video card would make a
difference.

As I see it, the real issue here is critical mass, of which marketing,
Dell deals, whatever you want to blame, are only subsets. Intel's
cash flow allows them to do things that AMD is unable to do, whether
it's marketing, cancelling key products without catastrophic fallout,
development of multiple approaches for the future, building expensive
new fabs, whatever.

This isn't necessarily a bad thing; it's just basic corporate
competition and free-market maneuvering. Do you think AMD would do
the same if their positions were reversed? Of course they would, and
people would rail against them for it.


Neil Maxwell - I don't speak for my employer
 
@news04.bloor.is.net.cable.rogers.com>, news.tally.bbbl67
@spamgourmet.com says...
They might, so if AMD were to use IBM as its fab, IBM's cross-license would
apply rather than AMD's. However, it's not certain how much of the x86
license IBM is granted access to. Maybe it would only cover upto the
technology present with the 386 generation (last time IBM made x86's of its
own), so maybe things like MMX through to SSE won't be covered?

No, IBM produced its own '486s and '586s (though there may be
some argument here ;-); The "Blue-Lightening" chips. And of
course Cyrix built their 6x86 and 6x86MX (complete with MMX) at
IBM, under IBM's cross. I don't see a problem with SSE either.
IBM and Intel are cross-license on much of their technology
(though obviously I don't have access to the particulars).
Regardless, even if all x86 extension technologies are covered in IBM's
agreement, they may have to pay a royalty for them to Intel. That is, AMD
might have a lower royalty agreement with Intel than IBM. So Opterons built
at an IBM plant might cost more than ones built in AMD plants simply for the
royalty issues.

IBM pays few royalties to anyone. that's part of the reason to
have such a dominant IP portfolio. Intel surely uses much IP
that IBM owns.
IBM could cross license some of its process manufacturing technologies to
Intel in return for reduced royalties on x86 technology, but since IBM
doesn't make x86 chips anymore, there may be no incentive for IBM to get an
updated x86 license. And I think IBM considers its process manufacturing
technologies much more valuable than x86 technology in its competition
against Intel.

It's more complicated (and simpler) than that. IP is traded like
baseball trading cards. It's not generally specific to one
product. ...and there is a lot more in there than manufacturing
vs. x86 specifics.
AMD and Intel being in the same business sector probably have a lot more
relevant stuff to trade with each other.

IBM isn't in the "processor" and "semiconductor" sectors?
Perhaps I'd better check around again! ;-)
 
Ed said:
I thought AMD said it was 30%, at least thats the last number I rem.
Ed

I'm not sure what you're talking about, but don't take these numbers as hard
numbers, they are just part of an example.

Yousuf Khan
 
Neil Maxwell said:
In addition, joint ventures/foundry deals are notoriously tricky. IBM
is not out for AMD's best interests, and vice-versa. Most
corporations will pull the rug out from under a partner for short-term
gain - there's a long history of this. AMD was working a joint
venture 300mm fab build with UMC some years back; did anything become
of that? They were going to use this to get to 30% market share by
this year.

Yeah, AMD pulled the rug out from under that partner to partner with IBM
instead. :-)

Yousuf Khan
 
Back
Top