Intel follows lead of AMD, introduces model numbers

[Robert Myers]

|> |> > On Sun, 14 Mar 2004 21:57:29 -0600, Rob Stow <[email protected]>
|> > wrote:
|> > >They are far and away Intel's best *desktop* cpu.
|> >
|> > The only available published SpecCFP2000 numbers I am aware of, ...
|>
|> This depends on a common definition of "best". If Rob meant _per Watt_
|> performance, I think the PM outperforms the P4 by a large margin...

And per cubic metre, per square metre, per kilogram, per rack,
per desk, ....

Currently, it doesn't do it per dollar, but Intel could change
THAT at the stroke of a pen[*]!

[*] A logical pen.

All of those are important criteria for potential server, cluster, and
HPC applications, but the claim was that the Pentium-M is Intel's best
desktop chip, and it seems unlikely that Intel could sell it as a
desktop chip on the basis of power consumption.

I can just see it now. People will be buying energy efficient
desktops to save the earth...with an SUV sitting in the driveway.

RM

 

[Nick Maclaren]

|>
|> All of those are important criteria for potential server, cluster, and
|> HPC applications, but the claim was that the Pentium-M is Intel's best
|> desktop chip, and it seems unlikely that Intel could sell it as a
|> desktop chip on the basis of power consumption.

Not even with the rapid moves towards using fliptops on the desktop?
A Prescott is a pretty hot chip to put in even a full-size 'laptop'.
Oh, yes, it can be DONE, but the things necessary to prevent trouble
remove most of its advantages.

|> I can just see it now. People will be buying energy efficient
|> desktops to save the earth...with an SUV sitting in the driveway.

Not quite

In my earlier posting, I pointed out that the Pentium-M's advantages
were in all of the areas where the profit lies, whether on the
desktop or elsewhere. From Intel's point of view, that makes it
Intel's best chip!


Regards,
Nick Maclaren.

 

[chrisv]

I don't know quite what the problem is, but Intel definitely seems to
be having some SERIOUS troubles making 90nm chips. What Intel claimed
was going to be their fastest ramp of a new process technology ever is
turning out to be a VERY slow process.

"Fastest ramp ever". Yeah, that's what JC was saying when the P4 was
about to come out. 8) I think they say that every generation. Easy
for the suits to say, harder for the poor engineers...

 

[chrisv]

They definitely have a tricky proposition here. They do want to
market their P4 chips well and especially their Pentium-M chips
better, but they can't go entirely based on performance or it will
become painfully obvious that the Celerons are trash.

Intel is expert at setting prices to match performance expectations to
extract maximum money from the market. It's a fundamental of their
business, and quite necessary particularly in the CPU business, with
the constant changes in the product line-up.

Expect their "performance ratings" to be purely market-driven,
producing at least the illusion of the classic Intel price curve, with
each step up the performance ladder costing you somewhat more than the
next-lowest step.

 

[Rob Stow]

The only available published SpecCFP2000 numbers I am aware of, 547
base, 552 peak, Hewlett-Packard Company, ProLiant BL10e G2 (1.0GHz,
ULV Intel Pentium M), do not support that assertion, even if you
assumed that results would scale linearly with frequency without
changing the Front Side Bus, which they would not. Compare that to
Intel Corporation Intel D875PBZ motherboard (3.4 GHz, Pentium 4
processor with HT Technology Extreme Edition): 1548 base, 1561 peak.

I suspect that media benchmarks would look even more unfavorable for
the Pentium-M. SpecInt2000 numbers for the systems already cited 673
base, 687 peak for the Pentium-M and 1342 base, 1393 peak for the P4
systems already cited, respectively, make it more of a horse race, but
let's not get carried away.

The benchmarks aren't everything.

The P4-EE might be Intel's benchmark winner, but it is their
*worst* desktop processor - it is a power hog that requires
lots of noisy cooling and has terrible bang/buck and bang/watt
performance. A 3.x GHz P4 NorthWood is better in bang/buck,
but not much better on bang/watt.

A 1.4 GHz or 1.6 GHz Pentium-M is far more than enough horsepower
for the overwhelming majority of office workers. If it wasn't
for the price and the lack of motherboards, what would make it
the ideal processor for office workers is that it can be used with
passive cooling and it has excellent energy efficiency.

If you could use 1.6 GHz P-Ms instead of a 2.6 GHz P4s in the
desktops of office workers, you would suffer no loss in computing
power and between the cpu and the air conditioning you would save
about over one KWh per machine per day.

All that is needed is P-M motherboards and ideally also lower P-M prices.

 

[Robert Myers]

A 1.4 GHz or 1.6 GHz Pentium-M is far more than enough horsepower
for the overwhelming majority of office workers. If it wasn't
for the price and the lack of motherboards, what would make it
the ideal processor for office workers is that it can be used with
passive cooling and it has excellent energy efficiency.

If you could use 1.6 GHz P-Ms instead of a 2.6 GHz P4s in the
desktops of office workers, you would suffer no loss in computing
power and between the cpu and the air conditioning you would save
about over one KWh per machine per day.

There is just no way that Intel can simultaneously argue that people
need increasing power on the desktop and start selling them a chip
that is a step back in power.

One KWh is less than 10 cents on the US average. Compared to $30/hr
fully burdened cost for the lowest imaginable level of office labor, a
slower computer only has to cost 12 sec/day in lost productivity to
eat up the cost savings.

RM

 

[Nick Maclaren]

|>
|> There is just no way that Intel can simultaneously argue that people
|> need increasing power on the desktop and start selling them a chip
|> that is a step back in power.

Why not? That is what the marketing department is there for.
And, if Intel's isn't up to the task, the UK has lots of slightly
used spin doctors for hire at discount rates.

|> One KWh is less than 10 cents on the US average. Compared to $30/hr
|> fully burdened cost for the lowest imaginable level of office labor, a
|> slower computer only has to cost 12 sec/day in lost productivity to
|> eat up the cost savings.

A watt in the CPU is more than a watt in the building, when the
heat has to be removed (a) by fans in the desktop and (b) by fans
and heat exchangers in the building.

And, if the extra fans raise the noise level in the office by
5 dB, or the extra heat raises the temperature by 5 degrees
Fahrenheit, what is the productivity loss?

[ Nothing to me, in either case, but I have 60 dB hearing loss
and actually like hot conditions (and I don't mean by UK standards).
Most of my colleagues differ. ]


Regards,
Nick Maclaren.

 

[Robert Myers]

|>
|> There is just no way that Intel can simultaneously argue that people
|> need increasing power on the desktop and start selling them a chip
|> that is a step back in power.

Why not? That is what the marketing department is there for.
And, if Intel's isn't up to the task, the UK has lots of slightly
used spin doctors for hire at discount rates.

|> One KWh is less than 10 cents on the US average. Compared to $30/hr
|> fully burdened cost for the lowest imaginable level of office labor, a
|> slower computer only has to cost 12 sec/day in lost productivity to
|> eat up the cost savings.

A watt in the CPU is more than a watt in the building, when the
heat has to be removed (a) by fans in the desktop and (b) by fans
and heat exchangers in the building.

And, if the extra fans raise the noise level in the office by
5 dB, or the extra heat raises the temperature by 5 degrees
Fahrenheit, what is the productivity loss?

[ Nothing to me, in either case, but I have 60 dB hearing loss
and actually like hot conditions (and I don't mean by UK standards).
Most of my colleagues differ. ]

Oh, the hell with it. <Dons flameproof set aside for heaven only knows
what reason>.

Intel got off to a really good start with Pentium-M. I don't think a
single-threaded, single core Pentium-M is going to fly as a P4
replacement. That being the case, I hope Intel will do something
really interesting, instead of just producing one more #$*!% me-too
chip.

Who knows, if they really put their minds to it, they could probably
get people to stop talking about their track record with Itanium.
After they got done caving in to popular opinion and replacing the P4
with a really imaginative, really capable multiple-threaded Pentium-M
derivative, they could conceivably even cave in to popular opinion and
let it compete with Itanium.

As for another ten years of arguing about miniscule details of
single-threaded chips, I can live without it.

RM

 

[The little lost angel]

One KWh is less than 10 cents on the US average. Compared to $30/hr
fully burdened cost for the lowest imaginable level of office labor, a
slower computer only has to cost 12 sec/day in lost productivity to
eat up the cost savings.

I don't see how switching from a 2.6Ghz P4 to a 1.6Ghz P-M would
reduce the productivity of most office labour involving mostly Word
and spreadsheets.

My laptop does most of my work on web design and coding, as well as
business spreadsheets and stuff. It's only got a 1.7Ghz P4-M that
typically runs at 1.2Ghz because I prefer longer battery hours.

Now if I don't even feel that a 1.2Ghz P4M is cramping my speed (the
4.5k RPM laptop HDD and measly 128MB are the bottlenecks), I doubt a
P-M with better IPC *AND* at higher clockspeed would result in any
loss in productivity for the typical office worker. Provided they give
them decent 7.2K HDD and more than 256MB of RAM if they plan on
running WinXP.

But the savings in power consumption will definitely be there.
--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code

 

[Alex Colvin]

|> There is just no way that Intel can simultaneously argue that people

|> need increasing power on the desktop and start selling them a chip
|> that is a step back in power.

Why not? That is what the marketing department is there for.
And, if Intel's isn't up to the task, the UK has lots of slightly
used spin doctors for hire at discount rates.

Well, as you know, you can't have well-defined spin in two non-orthogonal
bases.

 

[Peter Dickerson]

Well, as you know, you can't have well-defined spin in two non-orthogonal
bases.

don't you mean two *orthogonal* bases. After all, non-orthogonal would
include parallel or anti-parallel, and they can simultaneously have well
defined spins in each base.

Peter

 

[Nick Maclaren]

don't you mean two *orthogonal* bases. After all, non-orthogonal would
include parallel or anti-parallel, and they can simultaneously have well
defined spins in each base.

So much for the mathematics - has anyone told our Elected Leaders
this fact? Or, indeed, a certain CEO of HP, whose name escapes me



Regards,
Nick Maclaren.

 

[Rob Stow]

I don't see how switching from a 2.6Ghz P4 to a 1.6Ghz P-M would
reduce the productivity of most office labour involving mostly Word
and spreadsheets.

My laptop does most of my work on web design and coding, as well as
business spreadsheets and stuff. It's only got a 1.7Ghz P4-M that
typically runs at 1.2Ghz because I prefer longer battery hours.

Now if I don't even feel that a 1.2Ghz P4M is cramping my speed (the
4.5k RPM laptop HDD and measly 128MB are the bottlenecks), I doubt a
P-M with better IPC *AND* at higher clockspeed would result in any
loss in productivity for the typical office worker. Provided they give
them decent 7.2K HDD and more than 256MB of RAM if they plan on
running WinXP.

But the savings in power consumption will definitely be there.

And that's just for one user. During California's last
big energy crisis with the rolling brownouts I tracked down
the number of state and federal employees in California, plus
the number of municipal employees for LA and San Francisco.
I think I came up with something like 210 thousand - I
always felt that number was low and I think I must have missed
some huge groups of employees.

Assuming that half of them need an "ordinary" desktop computer,
I figured out that if they could save 250 W per user (or 2 KWh
per user per 8-hour work day) by using P-Ms and 15" LCDs instead
of P4's and 17" CRTs then they could have almost eliminated
what they were saying they needed for new power generating capacity.
2 KWh per user would be easy to achieve - particularly when every W
not turned into heat by a processor or CRT saves at least one more
W in air conditioning costs.

Perhaps what places like California need is power rationing.
Start telling companies that they are only allowed 1.5 or 2 KWh
per full time office worker per day for lighting, air conditioning,
computers, etc and we'll start to see some major changes in
the way computers are made and marketed.

You might think it odd that I - from frozen Saskatchewan - am
saying Cali this and Cali that, but the fact of the matter is
that even here we feel the ripples from everything the California
does. As a common example, vehicle fuel efficiency and pollution
improvements here were determined mostly by the auto manufacturers
striving to meet standards set by places like California.


And all this reminds me ... why don't they start putting those
humungous computing clusters up north so that they can completely
eliminate the need for air conditioning ? The 10,000 processor
Red Storm at Sandia, for example, apparently needs 3 MW just for
cooling. Send all the scientists who need the machines there too -
it'll be the only time in their lives they'll ever be cool :-D

 

[Rupert Pigott]

Robert Myers wrote:

[SNIP]

Intel got off to a really good start with Pentium-M. I don't think a
single-threaded, single core Pentium-M is going to fly as a P4
replacement. That being the case, I hope Intel will do something
really interesting, instead of just producing one more #$*!% me-too
chip.

I think that the VIA-ITX form factor & C3 chips show the way. My
mum's eyes lit up when she realised I could give her a fully
functional PC for word processing, web browsing and emailing that
was silent and fitted into a tiny little box.

There are a number of people out there who really don't want a
PC dominating a desktop. The monitor & keyboard they will accept
but that bloody great lump of noisey metal is a PITA for them. I
would argue that a P-M would be a better desktop CPU simply
because it has a chance of actually *fitting* on the average
desktop in the first place.

Cheers,
Rupert

 

[Robert Myers]

And all this reminds me ... why don't they start putting those
humungous computing clusters up north so that they can completely
eliminate the need for air conditioning ? The 10,000 processor
Red Storm at Sandia, for example, apparently needs 3 MW just for
cooling.

Already suggested that. When megabytes instead of megawatts start
flowing south, then we'll know that "utility computing" is more than
just marketing hype.

RM

 

[Iain McClatchie]

Tony> I don't know quite what the problem is, but Intel definitely seems to
Tony> be having some SERIOUS troubles making 90nm chips. What Intel claimed
Tony> was going to be their fastest ramp of a new process technology ever is
Tony> turning out to be a VERY slow process.

I have heard that at least one problem is that Intel's 90nm process leaks
more than expected. I don't know which kind of leakage it is (gate or
subthreshold). From the confusion I hear, I suspect there is more than
one problem.

IIRC, 90nm is Intel's debut of strained silicon. They are actually using
two new techniques. According to an article in [winces in embarrassment]
Science
The NMOS channel is strained in tension by depositing/growing/whatever some
sort of lattice on top of the poly gate layer. The lattice apparently tries
to match the poly's lattice, but doesn't quite fit, so that the lattice grows
in compression and slightly stretches the poly gate and the underlying
channel.

Q: with poly's being taller than they are wide these days, how does the
tensile strain propagate all the way through the poly to the channel?

The PMOS channel is strained in compression by somehow substituting germanium
atoms into the silicon lattice in the source and drain regions. The SiGe
lattice is bulkier than doped Si, and expands. The lateral expansion on
either side compresses the channel.

So: Is it possible that the strain which is enhancing carrier mobility is
also enhancing subthreshold leakage? (More than to scale, that is.)
Alternatively, is the shear across the gate causing more gate leakage in the
NMOS?

Inquiring little busibodies want to know...

 

[Nick Maclaren]

I have heard that at least one problem is that Intel's 90nm process leaks
more than expected. I don't know which kind of leakage it is (gate or
subthreshold). From the confusion I hear, I suspect there is more than
one problem.

I am also getting vibes that this is a lot more complex than many of
the causes of previous delays, but mine are probably more indirect
than yours.

What I haven't heard is that IBM and/or AMD have the same problems.
There are three obvious explanations:

1) They don't have or have solved the problems.
2) They are far enough behind they haven't hit them yet.
3) Their engineers leak less than Intel's do.

My guess is all three, in some proportion

The NMOS channel is strained in tension ...
The PMOS channel is strained in compression ...

This is all getting very Bohring ....

More seriously, is there anywhere that the principles of this are
written up in a fashion that those of us who gave physics up at
17 might understand?


Regards,
Nick Maclaren.

 

[The little lost angel]

This is all getting very Bohring ....

More seriously, is there anywhere that the principles of this are
written up in a fashion that those of us who gave physics up at
17 might understand?

Ditto from the csiphc idiot who gave up physics at 16 pPp

--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code

 

[George Macdonald]

What I haven't heard is that IBM and/or AMD have the same problems.

Maybe not "same problems" but IBM has admitted that they have serious
"yield" problems with their foundry business chips at 90nm, e.g. nVidia.
Figures as low as 5% have been mentioned.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

 

[del cecchi]

Already suggested that. When megabytes instead of megawatts start
flowing south, then we'll know that "utility computing" is more than
just marketing hype.

RM

Ever been to Rochester, MN in January? Plenty of free AC. But you
Canadians can keep the Alberta Clippers, if you don't mind.

del cecchi