A sociologist's take on Intel monopoly

[YKhan]

A Japanese sociologist explored the legal and psychological basis of
Intel's monopolistic practices. Some interesting quotes:

By allowing a competitor into the market, the exit cost for the
incumbent is small and may, in fact, prevent the entry of a stronger
firm. This is a strategy that Intel appeared to have adopted with
regards to laptops: we would suggest that the appearance of Transmeta
Crusoe machines in the Sony lineup was not disputed by Intel in the
hope of preventing the stronger competitor - AMD - from gaining a
foothold there. That, however, is speculation.
Relating these theories of monopoly back to the real world
requires a little more thought. Intel clearly conceded some product
space to AMD to allow it to enter the market with its low-end
processors such as the K5 and K6, back in the day. The weak firm,
however, has grown strong enough to compete by expanding its product
line to include high-end processors such as the Athlon, and, today,
cutting-edge technology like the Opteron. As AMD entered the high-end,
Intel monopolised rather than conceded and saturated the market with
offerings for every conceivable type of product. Arai suggests that it
is especially important to watch for monopolisation by any company with
multiple differentiated product lines, since this offers ample
opportunity for such anti-competitive acts as predatory pricing. That
would appear to be born out by the facts of this case.

Intel's monopolistic practices explored
http://www.theinquirer.net/?article=21716

 

[Robert Redelmeier]

A Japanese sociologist explored the legal and
psychological basis of Intel's monopolistic practices.

Hunh? Wouldn't the economics be paramount?

Relating these theories of monopoly back to the real
world requires a little more thought. Intel clearly conceded
some product space to AMD to allow it to enter the market with
its low-end processors such as the K5 and K6, back in the day.

low-end? IIRC, both the AMD K5 & K6 blew away their Intel
competition (486 & Pentium) on both technical and especially
cost-effectiveness. Intel was forced to react with proprietary
Sockets & Slots. The [NexGen?] K6 is so good that it lasted to
compete against PentiumPro and PentiumII and the essence of this
core is still apparent in the K7 & K8.

The shrink should stick to people. It's not nice to make up
facts to make theories work.

-- Robert

 

[Yousuf Khan]

Hunh? Wouldn't the economics be paramount?

Well, it is an institute devoted to both.

low-end? IIRC, both the AMD K5 & K6 blew away their Intel
competition (486 & Pentium) on both technical and especially
cost-effectiveness. Intel was forced to react with proprietary
Sockets & Slots. The [NexGen?] K6 is so good that it lasted to
compete against PentiumPro and PentiumII and the essence of this
core is still apparent in the K7 & K8.

Well, I do have few quibbles myself on his description of the
processors. Towards the end K6 was strictly a Celeron competitor until
AMD got the Athlon ready.

The shrink should stick to people. It's not nice to make up
facts to make theories work.

Actually, I think this guy was probably commissioned by the JFTC
considering that he must've gotten advanced copies of the JFTC findings
-- his report was filed Feb 17th apparently, because the filename is
"Arai_Feb_17.pdf", but the charges didn't get filed against Intel until
late last week. This guy's work is likely to find its way into the court
case that will happen against Intel.

Yousuf Khan

 

[chrisv]

low-end? IIRC, both the AMD K5 & K6 blew away their Intel
competition (486 & Pentium) on both technical and especially
cost-effectiveness.

You recall wrong. The K5 was weak, compared to the Pentiums of the
time, and the K6 BARELY edged-out (by about 4%) the Pentium in integer
performance, despite the benefit of having twice the L1 cache. Plus,
the Pentium kicked the K6 around the block where it really mattered,
floating-point performance.

In short, was the K6 a better value than the Pentium? For many
people, sure. Was it technically superior? Nope.

Intel was forced to react with proprietary
Sockets & Slots.

There were good reasons for the slot. AMD did a slot too, you know.

The [NexGen?] K6 is so good that it lasted to
compete against PentiumPro and PentiumII and the essence of this
core is still apparent in the K7 & K8.

Sounds like someone loves the AMD Corporation.

 

[Robert Redelmeier]

You recall wrong. The K5 was weak, compared to the
Pentiums of the time

Certainly. But the K5 wasn';t aimed at the original
Pentiums. It was targetted against the 486 which Intel
had to kill prematurely as a result.

There were good reasons for the slot. AMD did a slot
too, you know.

Yes, but slots are no more, and it wasn't about slots that
I was talking, but about proprietary CPU/mobo interfaces.

Sounds like someone loves the AMD Corporation.

No, I like Intel too. I like to recognize quality
wherever it is found. Intel has _great_ process.
But the P7 (Pentium4) was a horrible stop-gap made
necessary by the failure of IA64 (Itanium).

-- Robert

 

[YKhan]

Certainly. But the K5 wasn';t aimed at the original
Pentiums. It was targetted against the 486 which Intel
had to kill prematurely as a result.

No, the AMD 5x86 was aimed against the 486's, it fit into the 486
sockets. The K5 was aimed against the Pentium Classic, it fit into its
socket. AMD never got any decent ramp up of this processor, and
therefore it brought the K6 up quickly. The K6 was the processor
previously being designed by NexGen Technologies before they were
bought by AMD.

The K6 was actually a very good processor, but it was designed to
compete against Pentium Classic and Pentium MMX, at a time when Intel
was migrating over to the P6 architecture, 6th generation architecture.
K6 was at best a 5.5th generation architecture, so it was behind the
P6's.

The K7 Athlon actually played a leapfrog on Intel's P6 architecture
(which at that point was represented by Pentium 3), and it was still
able to give Intel's 7th generation Pentium 4 a huge headache for most
of its lifecycle. But then again, we now know that even Intel's own 6th
generation (now represented by Pentium-M) is also still able to give
its own 7th generation a beating.

Yousuf Khan

 

[keith]

No, the AMD 5x86 was aimed against the 486's, it fit into the 486
sockets. The K5 was aimed against the Pentium Classic, it fit into its
socket. AMD never got any decent ramp up of this processor, and
therefore it brought the K6 up quickly. The K6 was the processor
previously being designed by NexGen Technologies before they were
bought by AMD.

The K6 was reworked extensively from the Nextgen processor. The Nextgen
processor was quite nice, but it suffered from not being socket-5/7
compatable. That was a fatal difference at the time, so it took some time
for AMD to get it to fit.

The K6 was actually a very good processor, but it was designed to
compete against Pentium Classic and Pentium MMX, at a time when Intel
was migrating over to the P6 architecture, 6th generation architecture.
K6 was at best a 5.5th generation architecture, so it was behind the
P6's.

As soon as the K6s got an internal L2 they did quite well against
the slotted PIIs. They ran out of gas being limited by the 100MHz SS7
socket and when Intel went to the integrated L2 and back to the socket it
was all over for the K6. The slot actually sucked from the bilge. What a
horrible package.

The K7 Athlon actually played a leapfrog on Intel's P6 architecture
(which at that point was represented by Pentium 3), and it was still
able to give Intel's 7th generation Pentium 4 a huge headache for most
of its lifecycle. But then again, we now know that even Intel's own 6th
generation (now represented by Pentium-M) is also still able to give its
own 7th generation a beating.

;-)

Says a lot for P4, eh?

 

[Tony Hill]

Relating these theories of monopoly back to the real
world requires a little more thought. Intel clearly conceded
some product space to AMD to allow it to enter the market with
its low-end processors such as the K5 and K6, back in the day.

low-end? IIRC, both the AMD K5 & K6 blew away their Intel
competition (486 & Pentium) on both technical and especially
cost-effectiveness. Intel was forced to react with proprietary
Sockets & Slots. The [NexGen?] K6 is so good that it lasted to
compete against PentiumPro and PentiumII and the essence of this
core is still apparent in the K7 & K8.

The K6 did very well against the PentiumMMX, but it was released
almost exactly one month to the day before Intel brought out the
Pentium II (K6 was released in April of '97, the PII was released in
May of '97). From that point on, AMD was relegated to competing on
the low-end with the K6 and it's follow-ups.

The same was largely true with the K5. While it competed very well
with the low-end Pentiums, by the time it actually started shipping in
PR75 and PR90 forms (early to mid '96), Intel was selling Pentium 150
and 166MHz chips. Intel hit 200MHz before AMD managed to release a
PR100 chip.

While AMD's chips did compete well at the low-end due to their price,
neither the K5 or the K6 was very competitive on the high-end when
compared to Intel's offerings of the day. It wasn't until the Athlon
came out that AMD really had a high-end competitor.

 

[Tony Hill]

Certainly. But the K5 wasn';t aimed at the original
Pentiums. It was targetted against the 486 which Intel
had to kill prematurely as a result.

Check your time table! The K5 didn't see light of day until *THREE
YEARS* after Intel had released the Pentium. Even the PPro was
shipping before the K5 made it to market. The 486 was long-dead by
that time!

Yes, but slots are no more, and it wasn't about slots that
I was talking, but about proprietary CPU/mobo interfaces.

True enough, this was an important part of the court decision between
Intel and AMD. That being said, Intel DID have a very valid argument
here. They did do the design of the interface and they patented their
technology and it was a perfectly legitimate patent. AMD was on VERY
shaky legal grounds to begin with by using the Pentium bus, with the
PPro bus they just didn't have a leg to stand on.

Of course, in the end it may have turned out for the best as it forced
AMD to look elsewhere, and the resulting Alpha EV6 bus that they
licensed was a superior architecture anyway. It also laid the
foundation for AMD to develop their own Hypertransport bus, which is a
VERY nice design. Simple, flexible, extremely economical but still
very effective.

No, I like Intel too. I like to recognize quality
wherever it is found. Intel has _great_ process.
But the P7 (Pentium4) was a horrible stop-gap made
necessary by the failure of IA64 (Itanium).

While this may best be a discussion left for another thread, but IMO
the P4's failing was more due to the fact that Intel tried to solve a
problem that didn't exist. It seems like they had some study group
that decided the ONE application everyone was going to be doing in
2000 was a HUGE amount of multimedia streaming. They then went about
designing a chip that would be very good at streaming all sorts of
multimedia content and to hell with everything else because multimedia
streaming was to be the one and only thing that people would care
about.

Of course, the reality is that there is only so much multimedia
streaming that people really do, and the P4 turned out to be somewhat
lackluster in other areas. It's by no means a weak chip, just perhaps
less than ideal for the work that most of us do in the real world.
Intel's little fantasy world of massing streaming media just doesn't
interest most people.

 

[Yousuf Khan]

The K6 was reworked extensively from the Nextgen processor. The Nextgen
processor was quite nice, but it suffered from not being socket-5/7
compatable. That was a fatal difference at the time, so it took some time
for AMD to get it to fit.

Yeah that's right, but I didn't want to overburden with too many details.

As soon as the K6s got an internal L2 they did quite well against
the slotted PIIs. They ran out of gas being limited by the 100MHz SS7
socket and when Intel went to the integrated L2 and back to the socket it
was all over for the K6. The slot actually sucked from the bilge. What a
horrible package.

The 100Mhz Socket 7 was higher than what was available when Intel left
that platform. Intel left it at 66Mhz. It took a fairly long time before
anybody got their FSB's to 133Mhz, I don't think it came until after the
Pentium 3 was introduced. I remember even the original Athlon K7's were
initially running on a double-pumped 100Mhz FSB.

Says a lot for P4, eh?

There's now even talk of Pentium-M based Xeons coming in a few years.
Everyone is going back to the P6 architecture.

Yousuf Khan

 

[Yousuf Khan]

You recall wrong. The K5 was weak, compared to the Pentiums of the
time, and the K6 BARELY edged-out (by about 4%) the Pentium in integer
performance, despite the benefit of having twice the L1 cache. Plus,
the Pentium kicked the K6 around the block where it really mattered,
floating-point performance.

The K6 was hardly kicked around the block in floating point performance,
that was the Cyrix 6x86. The K6 was maybe 70-90% the performance of P5
and P6 at floating point clock-for-clock; which most people wouldn't
even notice except in benchmarks. The K6 overhauled the P5 in floating
point performance once its clocks rose up higher than P5. But the P6 &
K6 kept up with each other in Mhz for the most part, therefore the K6
was never able to overhaul the P6 in floating point.

Yousuf Khan

 

[keith]

Yeah that's right, but I didn't want to overburden with too many details.

Ugly is in the details.

The 100Mhz Socket 7 was higher than what was available when Intel left
that platform. Intel left it at 66Mhz.

True. Cyrix moved it up to 75MHz and 83MHz (wuz there). AMD finally
got serious at 100MHz.

It took a fairly long time before anybody got their FSB's to 133Mhz,

That wasn't an issue after L2 was integrated. The K6-III with a 100MHz
bus was a challenger to Intel's like offerings. Slot-1 was a technical
nightmare, but Intel needed it for differentiation. Score one for the
marketeers.

I don't think it came until after the
Pentium 3 was introduced. I remember even the original Athlon K7's were
initially running on a double-pumped 100Mhz FSB.

133MHz? Nope. It wasn't needed until then.

There's now even talk of Pentium-M based Xeons coming in a few years.
Everyone is going back to the P6 architecture.

EVERYONE? <flabergasted>. What *are* you smoking? (I'm assuming you're
talking about "micro-architecture", but it still makes no sense).

 

[Tony Hill]

The K6 was hardly kicked around the block in floating point performance,
that was the Cyrix 6x86. The K6 was maybe 70-90% the performance of P5
and P6 at floating point clock-for-clock;

Compared to the P5, the K6 was pretty close, but it tended to get
kicked around pretty badly when compared to the P6. Now, I suppose
one could argue that the 20-50% improvement in floating point
performance for the P6 vs. P5 was as much to do with the memory
subsystem as anything else, the end result was the same.

which most people wouldn't
even notice except in benchmarks. The K6 overhauled the P5 in floating
point performance once its clocks rose up higher than P5. But the P6 &
K6 kept up with each other in Mhz for the most part, therefore the K6
was never able to overhaul the P6 in floating point.

The K6 was never all that close to the P6 in floating point to begin
with. The crappy bus meant that it was never going to do very well
(especially prior to on-chip L2 cache), but it's FPU wasn't
particularly strong even if it could be fed with data. The same
arguments could be made about the Pentium as well, which wasn't really
a strong FPU performer, but it was already a generation behind the
latest and greatest when the K6 was released.

 

[chrisv]

Check your time table! The K5 didn't see light of day until *THREE
YEARS* after Intel had released the Pentium.

The 5V Pentium, you mean. The K5 was certainly out and competing
with the first "real" Pentiums, the P90-P133 series.

 

[chrisv]

The K6 was hardly kicked around the block in floating point performance,
that was the Cyrix 6x86. The K6 was maybe 70-90% the performance of P5
and P6 at floating point clock-for-clock; which most people wouldn't
even notice except in benchmarks.

Arguable, but FOR SURE no one could notice the K6's meager
integer-performance advantage without a benchmark.

I'm just amazed that the "K6 blew-away the Pentium" myth is so
persistent. There was never any evidence to support it.

 

[chrisv]

The K6 did very well against the PentiumMMX, but it was released
almost exactly one month to the day before Intel brought out the
Pentium II (K6 was released in April of '97, the PII was released in
May of '97). From that point on, AMD was relegated to competing on
the low-end with the K6 and it's follow-ups.

The K6-III did pretty well performance-wise, but the price never came
down enough for it to gain wide-spread appeal. A lot of people said
"just as fast as a PII for (a bit) less money", but they were missing
the point, which was that the Celerons with on-chip L2 kept-up with
both the K6-III and the PII, and for A LOT less money. The Celerons
were by FAR the best bang/buck for a while, there...

 

[Tony Hill]

The 5V Pentium, you mean. The K5 was certainly out and competing
with the first "real" Pentiums, the P90-P133 series.

The P90 was released in March of 1994. The K5 was released in March
of 1996, a full two years later. Intel released their Pentium 200MHz
before AMD hit PR100 with the K5.

Face it, the K5 was WAY too little too late!

 

[chrisv]

The P90 was released in March of 1994. The K5 was released in March
of 1996, a full two years later. Intel released their Pentium 200MHz
before AMD hit PR100 with the K5.

Face it, the K5 was WAY too little too late!

No argument there...

 

[Yousuf Khan]

The P90 was released in March of 1994. The K5 was released in March
of 1996, a full two years later. Intel released their Pentium 200MHz
before AMD hit PR100 with the K5.

That's not right, the Pentium was puttering around 133Mhz when the K5
PR100 was released. Then when the Pentium 166 came out, K5 was just
trying to reach out to PR133. So it was maybe a speed level back, but
not as big of a jump back as 100 vs. 200.

Actually, when the Pentium reached 200Mhz, it was greeted with a real
(no-PR) 200Mhz K6. I remember the K6 launch pretty well, the K6 was
launched at 166, 200, and 233Mhz. However that 233Mhz was pretty much a
paper-launch.

Yousuf Khan

 

[Guest]

Compared to the P5, the K6 was pretty close, but it tended to get
kicked around pretty badly when compared to the P6. Now, I suppose
one could argue that the 20-50% improvement in floating point
performance for the P6 vs. P5 was as much to do with the memory
subsystem as anything else, the end result was the same.

It's worth noting that floating point performance was a niche market
until Quake hit the market. Suddenly FP was for the masses. Plus the
"free" FXCH Quake's inner render loop favored the Intel chips. Of
course shortly after, everthing then went OpenGL, and perhaps (but only
perhaps) became a little less FP sensitive.

Dale Pontius