Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
I hope AMD's Q2 results tomorrow won't put you over the edge.
*Put* him over the edge? Have you seen some of the stuff he
has written here and at SI? (Intel tortures people, sends
anthrax in the mail etc). In polite terms, his posts are
often crazier than a large rodent who prefers to reside in
an outhouse.
While one may question the ethics of Intel's use of extortion and intentionally misleading claims to coerce buyers into paying high prices for inferior products, one can't help but notice that they're extremely good at what they do!
Translation:
Damn! Those black Intel thought control helicoptors are hard to
shoot down. Better go two ply on the aluminum foil.
Future caution aside, Intel did very well this quarter with Centrino. I am amazed at how much of a premium those systems get, even above P4 mobiles! I suspect that the prices will come down as the year progresses, but they certainly will continue to be very profitable in Q3.
Intel mentioned that margins were ~50% in 2Q but they expect
them to rise to 54% in 3Q and be 54% for 2003 overall. Unless
IPF/XeonMP is selling in greater numbers than anyone suspects
that means that Banias prices won't be dropping a whole lot.
FYI: Intel sold over a million Banias (Pentium-M) processors
in 2Q and expects to sell over 2m in 3Q. There is serious
money to be made in mobile - if you have the right product.
Can anyone confirm the inquirer's claim that in the
conference call it was mentioned that 1m Banias sold
in 2Q and Intel expects to sell 2m in 3Q?
chipguy, P4's target audience when it was introduced didn't know what SPEC is, much less care about it or ever run it,
Wrong. The P4 was introduced with only the 850 dual Rambus chipset
and it was a pricey bit of work. The primary buyers were ISVs and
technical computing types looking for Alpha FP performance at a much
lower price to run their codes. Other than the odd trust fund gamer,
mainstream use didn't start until the initial (SDRAM) version of the
845 came out in the fall of 2001.
I was just listening to Bloomberg and analyst said that
Centrino margins where 70% I have no idea where he got
that number he could have pulled it out of thin air for
All I know,
That all? Intel's overall margin is over 50%. Banias is so
small in a mature process and sells for so much I wouldn't
be surprised if Banias didn't given Intel margins the likes
of it hasn't seen since the numeric coprocessor days.*
*Numeric coprocessors sold for $500 to $600. To get an idea
of the NC margins, Intel took the 387 core, stuck it in the
corner of one of the versions of the i960Kx series embedded
control RISC as an FPU and sold the entire 960 device (about
4x bigger than a 387 core on its own) for under $100.
chipguy, that's true but not all. Can you explain please why some vendors shipped Athlon motherboards in grey boxes without clear origin marker neither on the board nor in documentation?
They were probably embarassed to associate their brand name
with the poor quality and robustnes of the VIA K7 chipsets
they ended up having to use.
Now, when P4 was first introduced, was not in the top ranks.
Perhaps not when running legacy PC apps but it was when
running appropriately optimized code:
SPECint_base2/SPECfp_base2k circa Dec 2000.
P4/1500 524 / 549
EV6/833 518 / 590
P3/1000 454 / 329
K7/1200 - / 304
The P4 beat out then mighty Alpha in SPECint_base2k and
was only 7% behind Alpha in SPECfp_base2k.
People believe that INTEL is doing a lot of things to deny competitors access to markets
I for one believe it. These things include:
- offering a wide range of generally excellent products at
reasonable prices and continuously replacing them with even
better ones made in the latest processes.
- works hard to ensure that all necessary pieces of the puzzle
are in place in a timely manner (chipsets, reference designs,
optimized applications, first class compilers etc).
- offer reassurance as the world's largest semico with multiple
fabs distributed around the world a secure supply of chips
in practically arbitrarily large quantities free from single
point disruption from political upheaval, terrorist attack,
or natural disaster.
- spends hundreds of millions to globally brand its products
thus providing OEMs with a valuable head start for sales.
Also subsidizes OEM advertising for systems based on its
processors.
Pretty sneaky and underhanded of Intel to do all that and
deny markets to its competitors that can't match it in some
or most of these areas.
The 1322 SPECint number was acually run on an rx2600, which is a 2-way Madison server. It seems very peculiar that a 2-way server with 1 CPU would outperform a single-CPU workstation.
It has to do with the memory system capacity and isn't all that
unusual. The highest SPEC scores for the Alpha EV6x family
of processors came from the 4-way ES40 and ES45 system for
example.
Here is a detailed explanation by John McCalpin of STREAM
fame:
http://www.aceshardware.com/forum?read=105023094
Rather than focussing on how the P4 performs relative to clock, try focussing on how it performs relative to what Intel
could have done.
Instead of inventing strawman arguments and hypotheticals, let's
have a look at how P4 "performed" in a metric much more relevent
to this board:
AMD margin (from SEC filings)
2000 46% (P4 intro'd Nov, 2000)
2001 33%
2002 22%
Over the same period as P4 became the primary core across all
of Intel's product line its margins slowly rose from just under
50% to the 52 - 53% range in the face of a severe tech turndown.
So the P4 made/is making loads-a-dough for Intel and all the
while sticking it to AMD bigtime. We could also look at what
the P4 did to AMD's market share and ASP but that would be
piling on. So wow!, what a "performance" by the P4!!! :-P
chipguy, if you replace 80% with 90% it will look even
worse for you
I am not interested in "what looks worse". Use the most
accurate figures available (including real SPEC scores,
not SPEC_rate scores) and let the numbers work themselves
out:
I2/1.3/3.0 - 875
I2/1.5/6.0 - 1077
Scale down I2/1.5/6.0 to estimate I2/1.3/6.0
x * ((1.5/1.3 - 1)*0.94 + 1) = 1077
x = 941
The real figure of 875 is 7% less than 941 and so this 7%
represents my estimate for performance loss from cutting
the 6.0 MB cache in half.
Ok, let compare SGI Altix 8P spec_int scores:
1. Itanium 1.3 Ghz 4M score 79.4
2. Itanium 1.5 Ghz 6M score 98.3
Ok let's do. But why did you show *SPEC_rate* numbers?
Do you know the difference between speed and throughput?
BTW, the 1.3 GHz Madison has 3 MB of L3, not 4 MB, and
performance scaling with frequency is over 90%.
Come back when you have a clue.
Releasing a low IPC design at souped up clock rates was a cynical play to public perceptions, but it wound up teaching the public that clock speed is certainly not everything.
No kidding? I think it taught the public to see through double
talk and excuses from a one time performance leader falling
further behind and with no way to respond on the product front.
Maybe the public remembered how AMD claimed reaching
1 GHz clock rates was like landing on the moon.
Here's a sign how the public accepted AMD's claims:
AMD margin:
2000 46%
2001 33%
2002 22%
The P4 was introduced in late 2000 and ended AMD's free ride
over coppermine. AMD's fortunes have been falling ever since.
More Yamhill rumours:
http://www.theinquirer.net/?article=10470
Also, SMT, CMP Prescott, MS-Intel spats, Tejas not taped out.
Also reiterates the Inq's opinion of Prescott frequencies and launch dates.
AMD supporters seem to have a love/hate relationship with the
notion of yamhill. On one hand they claim to fear yamhill
because they know Intel could steamroller AMD64 with their
own 64 bit x86 scheme, compatible or not. The dogma holds that
IPF is an ongoing disaster for Intel that diverts them from
the true course.
Yet OTOH there seems to be a subconscious desire for yamhill
to be true, for affirmation of AMD's course from big brother
Intel. The deepest darkest fear of Jerry's kids is that IPF
uber alles really is Intel's intention and that it is big and
strong enough and has enough OEM and ISV support to drive
it home. AMD would be left stuck at an evolutionary dead end
in the great tree of computing history facing a much more
formidable barrier to cloning IPF than it did with x86, MMX,
SSE, and SSE2.
Yep, the PIII always was a good core. Small, power efficient, somewhat limited, but responds well to additional cache.
P4 is a terrible core, but now that Intel's getting back to the PIII, they're doing better.
Yeah right. P4 was introduced in late 2000, a year when AMD's
margin was 46% competing against coppermine PIII. Competing
against P4, AMD's margin fell to 33% in 2001 and 22% in 2002.
Nah, I expect him to come up with something along the
lines of "for Intel you use the typical power consumption,
but if it's AMD use the maximum. That's the only way to keep
it fair
Not at all. I say we compare power on the basis of what maximum
thermal power each processor vendor tells system designers to
accomodate. For Opteron that is 85 W and for P4 it is 82 W.
Feel free to download the respective data sheets from the AMD
and Intel web sites and check yourself.
Umm, yeah, true. But there is a practical current per MOSFET pair that you can drive. In addition, the
current load on the inductors is a factor also.
LOL, If you think 52A is the hairy edge of solid state power regulation then you better not investigate
the power conversion electronics in electric cars, your head may explode. And your concern about
the inductor limitations is even more over the top.
So the specification of 52A is an absolute max, we want to make sure that the VRM doesn't
explode if the processor fries type of spec. Which is why Fairchild labels it a 42A (typ.) regulator
and not a 52A. Here, check a PR http://www.businesswire.com/webbox/bw.110602/223100309.htm
But I am assuming you understand the difference between typical and maximum. Maybe I am mistaken in this...
You seem to be the one with the conceptual gap. The part is specified for normal operation up to
52A because the Opteron core is specified to consume up to 52A. There, was that so difficult? If
the regulator is described as 42A typical then that is probably the typical current that AMD told them
the Opteron would draw.
I agree that the figure is there. But there is some evidence that it is over-stated. I base my
opinion on the fact that most of the voltage controllers for the Opteron just can't deliver 85 watts,
most are around 60 watts, and some are less. Of course, you might be like some and hold the
opinion that AMD just mislead them...
http://www.fairchildsemi.com/news/2002/0211/fan5098.html
If you are going to be sarcastic it would help if you knew what you were talking about in the
first place. This is a regulator device. It doesn't power the Opteron, it controls external power
MOSFETs that drive the current consumed by the Opteron Vdd core supply. The on-chip
DDR interface and HT links need their own separate supplies.
If you look at the datasheet for the Fairchild regulator here:
http://www.fairchildsemi.com/ds/FA/FAN5098.pdf
you will notice some waveforms showing typical operation ripple with load currents of 10A,
42A, and 52A. You will also see a lot of the device's electrical specs, like droop and current
mismatch, are specified for a load current of 52A.
Here's the amazing coincidence: If you look at the first sheet of the Opteron datasheet you
will see that power supplies include VDD(core): 1.55V at 52 A (max), VDDIO: 2.5V at
2.9A (max), and VLDT (HT links): 1.2V at 1.5 A. Add the power drawn by all these supplies
up and you get 89.7 W (MDR reported Opteron max power as 89W, this is probably how
they got that figure). Of course we all know from previous P4 discussions that a processor
cannot sustain max core power for a thermally significant period and this is why AMD rated
the max thermal power of Opteron at a lower figure, 84.7 W.
chipguy, how many transistors the EV8 core has, not counting cache?
Excluding the 3M L2 cache, about 30m in the CPU and about 30m in the memory
controller and router. The Opteron is roughly 25m in the CPU and about 15m in the
memory controller, router, and HT links.
As I see Opteron size and speed, it's core is not that much bigger than Athlon.
Cache is not drawing power. The only difference is memory controller and 1-3 HT
links. Can you claim 3HT+memcontroller >= 35 W?
How do you figure that? The Athlon TDP at 1.833 GHz is 68 W. The Opteron has a
20% longer pipeline, more logic in the front end to raise IPC and support AMD64,
and greater leakage power due to SOI and larger cache (A cache in 130 nm bulk
CMOS might leak 1 or 2 W per MB, an SOI cache perhaps more). Factor in the 3HT
links instead of an EV6 bus and a 128 bit DDR memory interface with controller
and a 85 W TDP is quite reasonable.
You know pretty much yourself that there is no technical way for Opteron
at 1.8 Ghz to dissipate 85 W or even 60 W
No kidding? Then tell me why the Alpha EV8 expected power dissipation
at 1.8 GHz in a 130 nm SOI process was 150 W at 1.1V?
Does physics work differently on the west coast than the east coast?
Can you tell me what you are, personally, doing on this board?
Trying to save unaware people that wander into here from droid BS like the
stuff you spout.
What do people think? Is SGI a buy here, or are they about to do a tactical bankruptcy and
start over with the old employees, old technology, new shareholders?
Here is some news that may give further clues to the answers you search for:
http://biz.yahoo.com/prnews/030710/sfth092_1.html
Silicon Graphics, Inc. (NYSE:
SGI - News) announced today that it has agreed to sublease its Amphitheatre
Technology Center campus in Mountain View, California to Google Inc. SGI will relocate
its headquarters to its nearby Crittenden Technology Center campus, where it will lease
additional space. The lease transactions are expected to result in a net reduction in
SGI's facilities occupancy costs of $14 to $17 million per year beginning in July 2004.
Can you think of any reason why AMD would exaggerate this maximum power figure?
Of course, can't you?
Sorry I couldn't hear you over the crinkling noise from your aluminum foil hat.
What did you say?
chipguy, first let use 55 W for Opteron, second let add 15 more watts to I2 for northbridge/memory controller:
AMD says their chip has a maximum power dissipation of 85W. It is right there in the data
sheet, the document that hardware developers refer to. If you want to engage in weird fan
boy masturbatory fantasies don't bother to reply to my posts.
I took the 85 Watt figure from the front page of AMD's data sheet for 1.4, 1.6, and 1.8 GHz
Opteron. Can you think of any reason why AMD would exaggerate this maximum power
figure?
Of course, Opteron doesn't draw anywhere near the 85 watts you have in your table.
Are you basing this statement on the amusing attempts of some enthusiast web sites
to measure Opteron power consumption? Do you know the difference between typical
power and maximum power? I would be shocked if they measured power anywhere
within 20% of 85 Watts.
No doubt the Madison doesn't draw anywhere near 107 Watts running most codes
either.
chipguy, I recall some of you Intel folks facing the argument that Itanium has bad
power/performance numbers claimed that power is not important. I think it was you.
Well, read this:
I think you folks often mix imagination with truth, to make Intel looking better. I wonder
what the purpose is?
Let's see who is twisting the truth: Let's have a look at some power/performance
numbers:
SPECfp_base2k / TDP (Watts), ratio (SPECfp_base2k/Watt)
130 nm:
I2/1500 - 2119 / 107 = 19.8
P4/3200 - 1252 / 82 = 15.3
K8/1800 - 1122 / 85 = 13.2
PPC970/2000 - 1120*/ 90 = 12.4 (*est. with AIX compilers)
K7/2200 - 873 / 76 = 11.5
180 nm:
I2/1000 - 1431 / 100 = 14.3
EV7/1150 - 1124 / 125 = 9.0
K7/1733 - 613 / 70 = 8.8
P4/2000 - 735 / 92 = 8.0
So what is all this talk about IPF having "bad power/performance numbers"? I think
you are either ignorant or twisting the truth as you accuse others of to make Intel
look bad. The numbers above speak for themselves. ;^)
Intel has never publically positioned the IPF family as anything other than a high end server
processor family
Do you still think so, given the link that was dug up saying they were going to make workstations with it?
Yes you are right, I should have explicitly included workstations. IPF was targeted at RISC
processors which are generally considered server chips but every RISC vendor also had
at least one line of workstations based on the same chips. My point was that there was no
talk about targeting the desktop.
wbmw, in financial industry we definitely do.
If Sun will make SparcStation notebooks, we will buy them, because we'd love to have notebooks compatible with our servers. Maybe not so many to justify the new Sun business :)
Companies like Tadpole did make SPARC and Alpha based laptops. They sold poorly.
chipguy, the term 'workstation' is generally used in this industry to indicate a high end
desktop machine with advanced graphics capability. Yes, a workstation is a desktop.
You can make up any definition you want but the computer vendors and the market research
firms that track the industry clearly distinguish between them. IIRC U.S. sales in 2002 were
divided between $68B in desktops and $2.5B in workstations.
All of this, though, is to detract from the main point: Intel was officially talking
about the 64-bit collaboration with HP in 1994, despite your efforts to redefine the
question to when they showed their first official roadmap.
Do I have to have to explain what a roadmap is to you now? In 1994 they didn't even
identify any specific product let alone a roadmap. What's a roadmap without any roads
or even a single city? A blank piece of paper. You yourself said that disclosure of Merced
came later. The first roadmap was in 1998 which identified the processors McKinley,
Madison, Deerfield and their approximate schedule and performance relationship to
Merced and among themselves.
please don't try to be such an expert spinmeister.
Trying to clue you in on the meaning of terms like desktop, workstation, and roadmap
isn't spin control. It's my good deed for the day. :-P
(Note that Merced was promised for 1999 workstations back in 1997.)
Sigh, do you know the difference between a desktop and a workstation? The issue
was your allegation Intel promised IPF based *desktops*.
So, are you still so sure that Intel never publicly talked about the project until 1998
They never showed an IPF roadmap with future processors on it before 1998 or the
relationship of the IPF roadmap to the x86 roadmap or the marketing positioning
of IPF vs x86. If you haven't noticed, Intel also markets x86 processors for the
server and workstation market too.
Remember this when you feel so quick to label other posters as definitively wrong,
as you so frequently do.
Perhaps if you learned the meaning of the terms workstation and desktop and the
difference between them you wouldn't confuse the two. This isn't my distinction
either:
http://www.mdronline.com/publications/tl/intel_2h2000/lof2.html
Your petulance on this issue reminds me of this joke:
Q: What is the difference between a used car salesman and a computer salesman?
A: The used car salesman knows when he is lying.
chipguy, do customers want to extend IPF to desktop? Yes. Is that technically possible? Yes.
WIll it help Intel to make more money? No. Bingo!
Considering IPF product pricing, your first and third point seem contradictory.
Anyway, this has nothing to do with your incorrect characterization of Intel's publically
stated intentions for IPF, past or present.
Also, did anyone catch the reference to "AMD Hammer at 2.5 GHz" in chart 49?
Clock headroom issues aside, can the Opteron even operate at an odd multiple of 100 MHz?
It has been inferred. Wbmw has made such reference and it is only moore's slaw confirmation.
Who knows when Intel will go 64bit desktop, but I2 looks to be the only candidate. Unless Yamhill?
There is certainly a lot of room to speculate about IPF going forward but that is not the issue I addressed.
YB claimed "Originally Intel promised to make IPF across the range of products, so some customers may
expect that $1,000 Itanium notebooks and $600 Itanium desktops will come and they made those asumptions
part of the IT strategy.". This is clearly wrong and I challenge anyone to provide evidence to support it.
chipguy, never is a long time. I recall when Intel and HP announced their IPF effort in 1995 the
talked about first product in 1998 and desktop versions in 2000. (No, I don't have a handy reference to
post.)
Intel and HP disclosed basically nothing about their intentions for IPF. All the talk about replacing
x86 was breathless speculation by analysts who had column inches to fill. Linley Gwennap of
Microprocessor Report basically wrote an entire article using one sentence worth of basic
facts (Intel and HP joint project, VLIW, 64 bits, late 1990s) and about 5 pages of speculation.
Much of the non-sense IPF hype in the press of that period flows back to that one article.
I also recall the Intel VP in 1998 who said that the in-progress products were not suitable for
desktops. It was quite a significant change in guidance and was much circulated at the time.
No, you are wrong. That was the first public disclosure of Intel's intentions for IPF. Any change
in "guidance" at that time was due to industry analyst speculation being proven groundless.
Originally Intel promised to make IPF across the range of products, so some customers may
expect that $1,000 Itanium notebooks and $600 Itanium desktops will come and they made those
asumptions part of the IT strategy. Then later they will realise that nobody is planning to do $600
Itanium desktops.
This is a pure and utter fabrication.
Intel has never publically positioned the IPF family as anything other than a high end server
processor family created to attack the high end server market occupied by RISC processors.
Intel has also never showed roadmaps where IPF processor development continues and
x86 processor development ends.
Check the old tech journals of the mid-late 1990s. All the talk of IPF uber alles comes from
over excited analysts who didn't listen carefully to what Intel was actually saying and instead
made up hidden motives and agendas and attributed them to Intel. Kind of like what many
AMD enthusiasts often do except the so-called journalists and analysts involved got paid for
it. Someone asked an Intel VP in charge of uP development about Deerfield when the IPF
roadmap was unveiled around 1998 or 1999. He said it wasn't suitable for the desktop, it
was for low cost servers.
Here's a flyer - buying SGI would be a good way
for Sun to grab a foothold in the IPF hardware market
Nope, it's still there: http://www.sun.com/executives/realitycheck/reality-021803.html
True. But SPARC still "isn't there" here and here and here:
http://www.specbench.org/cpu2000/results/
http://www.specbench.org/jbb2000/results/
http://www.tpc.org/tpcc/results/tpccadvanced2.asp
And it will not "get there" any time soon if ever.
In this scenario the
app is a real cache-buster and spends its entire time waiting on small items from memory.
The real figures can be estimated from the graph on the next page. The I2 is actually much slower than the P4!
These two factors don't compute. The I2 has twice the memory bandwidth as a 2 GHz P4. Depending
on which machines they used to benchmark, the I2 would likely also had lower memory latency.
Also, the graph shows McKinley performing at around 1/4 the level of the EV7. Yet the EV7's STREAM
copy score is only 68% higher than McKinley's. I wonder what compilers they used for this exercise.
What do people think? Is SGI a buy here, or are they about to do a tactical bankruptcy
and start over with the old employees, old technology, new shareholders?
Disclosure: Long SGI, but ready to dump it again
IMO it looks like SGI has a killer machine for the HPC market and their long
history in the field with their Origin products gives them the contacts and
reputation to sell it. The company was in bad shape but it looks like Altix
will go a long way to saving it. If you already own the stock I'd say hold it
a little while longer because the company just finished years of broccoli
and is about to move to dessert. Is it a better buy than other things out
there? I dunno.
A takeover target if IBM decide they underestimated I2?
Possibly but not by IBM. Altix competes with large scale POWERx systems in
large scale single image HPC applications. IBM also has its own decent IPF
products on the way (x455). Here's a flyer - buying SGI would be a good way
for Sun to grab a foothold in the IPF hardware market was well as becoming
more credible in the HPC and Linux world.
You can imagine 108 racks forming a loop with a radius of 50 feets and you know that the
packet can reach any of those racks at the same time, no matter how distant it is, just under
5 * 10^6 seconds. Pretty impressive.
5 us yes, 58 days no.