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.
Mike,
I guess that's it for Athlon XP.
I have to say that I am not too thrilled with availability of 90nm product and Semprons (of all shapes, and speeds). It seems that Newcastle and a few Opterons is all that is for sale.
Joe
Keith,
I wonder if any of the mobo makers does a 939 board with nForce 4 Pro, giving it 2 full 16 wide lanes for SLI.
I know it would mostly be a gimmick, but some people want the ultimate...
BTW, I am surprised they included Creative Labs SB Live on the board. Considering my long history of enduring buggy and late drivers, I would rather not go back to their products (but, I like the company. I made a lot of money on their stock in the past).
Joe
NaS,
You must be assuming that with increasing number of the service pack the product becomes more stable. I think SP2 is a mixed bag. Maybe 2 steps forward, 1 step back. While there may be a lot of improvements in security, from my limited experience, it introduced a number of quirks that I have been dealing with.
BTW, I agree with chipdesigner about XP (vs. W2K) for a gameing box.
Joe
Smallpops,
Maybe IDC counts the Sun grid as one system.
Joe
chipguy,
The main factor that has slowed IPF acceptance is the
slow pace that any new ISA acquires a software base
I believe that Esperanto faced the same challenge.
Montecito will give IPF OEMs the ability to take on and
defeat POWERx in performance and cost/performance if
they do their part but those OEMs need to reduce their
margins and costs on low end systems to better compete
against x86 for the 2-4P server segment if they want to
seriously drive up unit sales.
One thing that may be escaping you is that the OEMs work for themselves, not Intel. What's the point of selling Itanium at low margin and low volume with high R&D expenses, if they can sell x86 at low margin, high volume with low R&D expenses.
And, BTW, why doesn't Intel lead the in cutting margins to encourage volume by cutting prices of Itaniums to say $100? At full prices, 4 way Itanium costs almost as much as a complete Opteron 4 way system (including decent OEM margin).
Or even better, to encourage volume even more, Intel could just give the Itanium processors away for free...
Why
wbmw,
A little more realistic is to picture IPF in a "back then" vs "is now" perspective. In the early to mid-90s, Intel was being threatened by RISC vendors (ironically, PowerPC being one of them), and IPF made sense as a winning horse in the battle of architectures. The idea was probably that *IF* x86 were to fail, then the winning architecture had better be something from Intel.
I agree, but in addition to the "defense", a bigger consideration, IMO, was the "offense", which is to eliminate the competition in the mainstream market by monopolizing the instruction set.
Fast forward 5-10 years, however, and the scene changes dramatically. Intel won the battle for the volume server market by changing the landscape to compete on micro-architecture, rather than architecture (which probably came as a surprise to the Itanium folk). In other words, they used x86 with advanced micro-architecture techniques to not only win back share in the existing server market, but also expand the size of the server market (TAM) by several fold.
I would replace the word "win back" to "win". There was not much of a volume market in the servers, since there were either minis/ mainframe, and the Unix boxes that were very expensive and the volume was low. Intel did not really participate.
Novell and later Microsoft Server (cheap) software for the cheap PC processors, which created the market, that was bound to explode.
Having worked with early PC server products, it was clear to me that it would kill the high end eventually, leveraging huge installed base of hardware and software, especially ease of software development. In the 1990s, for this not to be clear to Intel would be most puzzling.
As you know, I am an advocate of the IPF architecture, and I think it's the one thing today that continues to save performance from being behind the minimum.
I am not sure what you mean by this.
However, with a better micro-architecture made from technology as bleeding-edge as today's x86 processors, IPF would be a real screamer. I think Montecito will help here by adding multithreading and advanced power management, two things that IPF processors have lacked over today's x86.
I think Itanium's problems are mostly software related as in lack of, difficulty create it, expense of creating it, barriers to entry for the developers.
The second problem is the vicious cycle of low volume and high prices.
While I agree that Mentocity looks good on paper, there are other things going on that will keep it from breaking the cycle.
I posted this before that there is going to be a tremendous gain in the performance of x86 systems with adoption of 64 bit OSs and apps. Opteron is going to put pressure on pricing by extending the low margin pricing to 4 way, even 8 way. Now add dual core...
The end result of 64 bit software and dual core is that the performance per socket is going to double. 1 socket system will match the performance of 2 socket, etc. The price of this 1 socket system is going to only marginally change from 2 socket.
So if Itanium (according to Otellini) is not very cost competitive in 4 way on less systems, the market where it is competitive > 4 way, (say 40% system revenue, 5% unit wise) will shrink to say 20% revenue-wise and 2% unit wise over time, which will make it so much harder to get any kind of volume going for Itanium.
I don't expect Montecito to be a cure-all, though, but rather a much needed boost. As for the architecture's usefulness, now that Intel has made Xeon as successful as it can be, they still need a way to stand against Power (once again, the ironic challenger both then and now).
If Power did not have IBM name and IBM customer base behind it, it would go the way of PA-RISC, Sparc, Alpha, MIPS, etc, given enough time. But with the high end customers, and a low end angle, Power is still a presence, but, somehow I doubt that Itanium is the right weapon to challenge it.
The positioning problem is to make Xeon look good against Opteron so that Intel keeps their x86 share of the market, while growing TAM by replacing RISC with IPF. It's actually a good strategy, but they need to stop making IPF look good by selling Xeon short, and start marketing IPF on its own merit (which I think they are now doing).
Good point. IMO, in absence of Opteron, INTC could have just put a cap on Xeon, but that is not an option now. The ironic thing is the need to make Xeon systems more competitive, performance-wise and price-wise, it makes it even harder for Itanium to succeed.
Joe
This STMicro 2 bit per cell NOR, is this like Intel's multilevel or like AMD's MirrorBit?
Joe
mas,
'But in the late 1990's Intel's marketing department was faced with managing the introduction of Intel's own 64-bit RISC-like processor to eventually replace x86 after they spent ten years telling their customers they didn't need to replace their x86 machines with competing RISC-based computers. To solve this dilemma they came up with the term EPIC or Explicitly Parallel Instruction Computing to explain that not only is the IA-64 architecture not only better than RISC but deserving of an entirely new category.'
It looks like Intel wants Itanium just jumped on the RISC bandwagon: <g>
http://www.siliconinvestor.com/readmsg.aspx?msgid=21045187
http://www.intel.com/products/roadmap/server.htm
Joe
sgolds,
It is interesting how the Itanium cheerleaders are justifying Itanium failure by fabricating an excuse that Intel never planned for Itanium to be a success, that Intel actually planned for Itanium to be a failure (low volume high end niche product).
One thing that needs to be conceded to Itanium is that it succeeded in killing several RISC CPU architectures. Not with the actual working silicon of the CPU, just with the press releases about it.
Joe
aleph0,
One thing you may want to do is to just install 64 bit windows on a new hard drive and keep dual boot for a while, while you migrate the old software. I am sure that there will be some odd 16 bit or dos utility or something that I will realize I need for something, when I move my computer to 64 bit
Joe
Chris,
What should have been a sure thing for Intel--tons of power for massive computing--has become a commercial disaster. With market-research companies drastically downgrading their sales forecasts and Itanium co-developer Hewlett-Packard bowing out, some experts predict that Intel will discontinue the processor.
What? The Esperanto processor is turning out to be a commercial disaster?
Joe
aleph,
There are 2 issues here: how you purchase the product and how you install it. As far as purchasing, Microsoft will apparently let you exchange your existing copy of Windows XP / Server 2003 for 64 bit equivalent, either for free or for nominal charge. If your current version is one of these, you are in luck. If you have a previous version, you have to buy a new one (or get an upgrade version to 32 bit and then exchange for 64 bit).
As far as installing there is no upgrade of an existing system to 64. It has to be a full install from scratch.
Joe
decaw,
Love the URL
LOL.
NaS,
Notice that A64 3800 beats 3.2 GHz Northwood hands down in power consumption. Imagine how it would do compared to 3.8 GHz Prescott.
Joe
Mike,
That one looks like Socket 940 Athlon FX.
Joe
T64,
I know I must be putting you to sleep but don't you think it odd that a stock goes from $15 to $25 and back to $15 in such a short period of time
Must be a conspiracy...
Joe
Klaus,
there will likely be 130nm versions of X52 (for older platforms), while there can are 90nm flavours as well.
That sounds interesting and would start to make the incomprehensible article on The Inquirer, well, even more incomprehensible:
http://www.theinquirer.net/?article=20252
Joe
Somewhat OT:
The US economy - GDP - grew at 4.4% in 2004, .2% less than mid year estimate of 4.6% (which would have been a 20 year record), but it is the very respectable, and the best in last 5 years.
http://www.smartmoney.com/bn/ON/index.cfm?story=ON-20050128-000562-0841
So much for the doom and gloom merchants at New York Times, and the merchandise they were so desperately trying to sell with their sales team (oops - reporters) and their advertisement (oops - articles): John Kerry.
Joe
Mike,
re: ADAFX55DEI5AS
This one has to be 130nm.
D stands for Socket 939
E stands for 1.5 V
I stands for cas temp of 63
5 stands for 1MB
AS stands for CG stepping
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30430.pdf
Joe
chipguy,
You are right but haven't taken this reasoning to its
logical conclusion. Those 64 bit x86 apps are going to
get optimized alright. They will be optimized for by
far the most popular and lucrative 64 bit x86 platform.
And looking forward right now that is Prescott/Nocona.
I am sure Intel compiler will be, but Microsoft's compiler was developed on K8 processors ... Also, the chances are that Merom will not be anything like Prescott / Nocona.
Joe
Tenchu,
Interesting. Then again, if HyperThreading doesn't impact single-threaded performance per core, why not leave it activated?
A scenario where 2 active threads get assigned to the same physical CPU. So you end up with double die size, 3 speed grade reduction, and no gains. If HyperThreading is disabled, you are guaranteed that threads get assigned to separate physical CPUs.
When considering performance of CPU with multiple cores, while each additional core brings less additional performance to the table, it is somewhat predictable. HT + dual core adds an element of unpredictability
Joe
Smallpops,
How high do you think they can go with 90 nm?
A good rule of thumb would be whatever Intel says minus 10%.
Joe
Alan,
When building an app, it makes very little sense in most apps to open a can of worms by compiling for anything other than the lowest common denominator, which is basically 386 or 486.
The lowest common denominator machine for 64 bit computer has a different set of assumption, further 10 or 15 years of advances of CPU architecture, plus you get to use extra registers.
So what is really a real world comparison here? There is none, IMO. Hand assempled code or highly optimized code may not necessarily apply. There is only one real world result, the unoptimized 32 bit executable (based on C code), for which there is no 64 bit counterpart in the x86-secrets comparison.
I think we will get a flood of these comparisons in summer of this year when Microsoft Visual Studio 2005 goes to General Availability.
Joe
subzero,
You recently claimed Wally Maghribi's court testemony about AMD's racist comments was "hearsay".
The case was dismissed, so it does not look like anyone was buying or endorsing what may have been said.
on the other hand, it looks like the Japanese FTC is buying or endorsing the allegation. They raided Intel's offices, so they may have solid evidence in their hands.
Joe
UND,
I wonder if AMD will release a separate power consumption spec for 90nm FX-55 or if there will be just an overall cap. Hopefully, the 90nm Fx-55 will fit with ease in the old 89W spec, so that it can be sold as upgrades to systems such as the Compaq gaming box.
Joe
Interesting interpretation of Japan FTC issue:
http://www.aceshardware.com/forums/read_post.jsp?id=115121542&forumid=1
Joe
Keith,
Since all 90nm CPUs are labeled as such, I guess the new Semprons are 130nm. It seems that AMD has a way to market even the most marginal parts, which is good.
Joe
Mike,
It looks like some 2,000 and change of that 14,000 order got delivered.
Joe
Mike:
90nm FX-55
WOW!
Best news I heard in months. This must be the E stepping.
Joe
<edit>I am kind of excited about it, but it was Ihub that somehow repeated the message 3 times</edit>
wbmw,
Hector Ruiz, up on stage to officially confirm that Sun was AMD's highest volume Opteron customer in Q4 last year.
I guess the question is why is Sun outselling HP. I think the reason is probably this grid Sun just put in place. It seems to have 15,000 Opteron CPUs.
Now add to that the regular run rate that was probably between 15,000 and 25,000 in Q4, that's 30,000 to 40,000 CPUs.
I don't know the total unit sales of Opteron. My guess is 150K to 200K. So Sun would be 20% to 25% of total sales.
Just WAG figures...
Joe
tecate,
I think it's disgusting that articles like this are printed without any evidence, it's all hearsay.
The article is a leak on what Japanese FTC is about to do. When Japanese FTC does issue this warning, it is based on investigation and evidence, not hearsay.
Joe
wbmw,
I agree that AMD will have a smaller average die size in 2005 relative to 2004, but they are also shifting to a more expensive process with a larger number of metal layers and process steps.
Good point. I think the first 90nm stepping (Dx) is using the same number of metal layers. There were some rumors that the next stepping (Ex) will have 1 or 2 extra layers, so it will become an issue shortly.
I wonder if there is makes sense for AMD to keep the low end of the market (that does not need higher clock speeds) at the Dx stepping, or if it is more practical to keep all the cores synchronized.
Joe
Smooth2o
What has changed here?
I was talking about pricing (ASPs). I think there will be relative stability in 2005, and the hell will break lose in 2006.
Joe
Keith,
Yes, two thirds of their sales are INTEL based. They waste money on idiotic projects like a BTX SFF for Prescott. The result? The "SFF" got bigger and the noise unbearable. I guess they call that progress.
I think the side effect is the SLI Athlon 64 SFF with the same diminsions.
Actually, I would probably be content with a competent uATX Socket 939 mobo that would fit into the standards based, neat looking Antec SFF case.
Joe
Keith,
I read forum on http://www.sudhian.com/ especially their SFF section, and I see a quality control issues all over the place. Shuttle needs to tighten up this area.
Also, Intel based sales are still a big percentage of their business, and SFF + Prescott are not a happy combination, especially when combined with a power hungry graphics card.
Joe
wbmw,
Let's say AMD's output from Dresden tops out at around 9M parts per quarter.
I have thought of 10M at acceptable yields, but 9M will work fine for the discussion.
Granted, AMD will soon have more parts on 90nm, which lowers die size and potentially increases fab output, but I think this will be counter-balanced by the shift from K7 to K8. In the past several quarters, AMD has still had a sizable mix of K7 Semprons, according to reports. I believe this was to maximize capacity over the past year, so I do not expect the shift to 90nm to improve things that much. Moreover, a shift to dual core parts later in the year will cause AMD to shift back to larger die sizes.
While I don't know the exact plans of AMD, I think that the dual core output is going to be miniscule as a percentage of overall sales, therefore will not affect the die size very much. My thinking is 50% of Opterons, which may be some 150K units, and a smaller number for desktop.
Overall, I pretty much agree. The die sizes will be:
- Low end old: Throughbred 84
- Low end new: Winchester, Palermo 86
- Mid range old: Barton 115, Newcastle 144
- Mid range new: Winchester, Venice 86
- High end old: Newcastle 144, Claw/Sledgehammer 196
- High end new: San Diego / Opteron 115, Dual Opteron ~200
which is overall a slight reduction
Spread out over the entire year, I do not expect more than their 2004 production output, and 16% of 180M is 28.8M for the year. At a maximum, if demand for AMD is great enough, they could manufacture 9M per quarter regardless of demand, which would put their yearly output at 36M.
Sounds reasonable.
Meanwhile, IDC predicts 10% growth next year (and in previous years, their predictions have proven conservative). That would bring demand for the year up to ~200M units. If AMD's yearly output without new capacity is 29-36M, this would put their market share at around 14.5-18.0%. Thus, Intel has a potential to gain market share next year due to AMD's lack of capacity, even though AMD will have a much stronger product line. And in the worst case for Intel, AMD only has headroom for about ~150 more basis points in market share.
Yup, I don't expect much in term of unit market share growth for AMD in 2005 beyond what you outlined. Higher number of units, but not much share gain.
If you combine AMD's lack of capacity with Intel's not so strong position in segments of the market contested at most, the result I see is a fairly benign pricing environment to continue another year. It sould be good year for CPU profits.
The gains that I expect for AMD are mainly in terms of ASPs and moving a bigger chunk of its product offerings from the very low end to slightly more rewarding markets, and gains of more of the very high end market (excluding notebook market).
I think the hell will break lose when AMD has more capacity online in 1H 2006.
Joe
Keith,
Asus will add some AMD CPU-based products to what is currently an all-Intel lineup
About time, I have bought a few Shuttle SFFs, and I am very underwhelmed by the quality of Shuttle products. I definitely prefer Asus products.
Joe
wbmw,
Assuming 180M units for the year, that's roughly 45M CPUs per quarter, on average, or probably 35-40M in the slow quarters and 50-55M in Q4.
Thanks. I have read somewhere a number as high as 200M, but that can be a projection for the future.
50-55M for Q4 would mean (unitwise)
Intel 41.1M to 45.2M
AMD 8.3M to 9.1M
Joe
Alan,
Product cost is a strong function of factory utilization. When utilization is high, cost is low. When building inventory, the product cost as shown in the quarterly report is actually understated.
That's a good point.
But in case of AMD, factory utilization is unlikely to drop as they go to dual core (unless there is some kind of global crash in demand).
Joe
mmoy,
In my JPEG stuff, I did make some changes that were favorable to K8 and had a negative impact on P4. But that's life.
Whatever benefits K8 probably benefits Dothan as well, and since netburst core will probably be gone in 2 years, it is probably a good decision for long term.
Joe