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.
Smooth, look at it this way: Centrino offered something AMD couldn't, i.e. low power. Wireless was getting all too standard and the vast majority of AMD notebooks were at the time that Centrino got introduced nicely outfitted with the standard (.11b/g) as well so it was not nearly as relevant a differentiator as low power (.11b/g either as build in or optional). Also .11b/g PCMCIA cards were aplenty as well. Either way it was available en masse. The reason I know is because I bought a notebook back then and I checked virtually all brands that carried AMD DTRs. Low power good performing T&L notebooks, a space ONLY adressed by Centrino. For years, up till Turion.
Now Viiv offers ABOLUTELY NOTHING that AMD can't match, and X2 well outperforms DC P4 (and at lower power). See the difference?
Regards,
Rink
Smooth, Viiv will start with DC Netburst P4 as cpu, I think at least till Conroe. Sure it can still be popular. Just needs to be subsidized a bit heavier than any Intel x86 cpu ever before. Not inconceivable.
Just to put it in perspective: There was e.g. an expensive marketing campaign to get Centrino started, but at least it offered one very compelling feature in low power (besides nice performance). This was especially compelling ofcourse because of lack of competition regarding this feature. Now DC Netburst P4 has no such feature. It will perform significantly below X2 for most of the year (until at least Conroe). This doesn't mean that it won't sell, just that it's unlikely to be a hype or live up to become Centrino-like brand name.
Don't you think that it's logical that this lack of differentiation will show in the sales numbers?
Regards,
Rink
Keith, TX, I bookmarked it (eom)
Keith, (EDITED) no I think I read it. Can't search this board. Didn't want to upset you. I'll find it back somehow.
EDIT: I think I found it back via SI: http://www.tecchannel.de/imgserver/bdb/346000/346016/original.jpg
Regards,
Rink
Keith, (EDITED) Anandtech slides refered to SpecIntRate/Watt. Not just SpecIntRate by itself. Big difference. See Y-axe tags on slides 3 and 5 here: http://www.ANANDTECH.com/tradeshows/showdoc.aspx?i=2503&p=2
But you probably have a different source of information, right? I remember you saying something like that in the past. I know we talked about this before, but can't find the conclusion back (which was that SpecIntRate projections were impressive). Could you please remind me, I mean if it doesn't cost you too much time?
Regards,
Rink
Keith, re: I agree that Conroe/Woodcrest should come in at clock speeds that your proposed range covers [2.8-3.6GHz].
Why exactly do you agree with that?
Regards,
Rink
Pete, re: There are A64 X2 3800+s (and higher) with TDPmax less than 35W in the field bought by real customers from real retail shops in the real world. ... You can get 939 pin DC Opterons and A64 X2s in laptops using less than 35W TDPmax right now.
Can you please provide a link or some other evidence. I'm not seeing what you're talking about anywhere.
Regards,
Rink
Tx Pete, Can you give us a quote / link perhaps? IPF proponents will probably rightfully claim that revenue went up because #processors/server went up. I'd like to know how much exactly so if you'd have a quote/number/link especially for QoQ IPF revenue growth I'd highly appreciate that.
Regards,
Rink
Chipguy, re: 8W more for FBDIMM
I clearly remember reading exactly that for one specific unit but have to admit I can't find it back. I also remember seeing the FBDIMM test case I read about used active cooling (fan) for the DIMMs (I posted about that at SI; will look it up if you'd like that).
Anyway closest I found is this which mentions typical > 5W power consumption for the AMB chip (i.e. less than the 8W I mentioned having seen): http://www10.edacafe.com/nbc/articles/view_article.php?section=ICNews&articleid=213237
Regards,
Rink
Dacaw, re: This has to be the most twisted piece of non-logic I have seen in a long time.
It appears that way I agree. AMD has sufficient stock so it is not likely to lack on the supply side. AMD chipsets are not reported to be short so they might be OK as well. I think AMD is taking share all over the place this quarter, specifically in server and mobile.
Regards,
Rink
Keith, (EDITED) re: Do you expect AMD to release a lower voltage X2 part for the "X2 DTR market"
Yes. Imo it's a technically possible opportunity and it is therefore reasonably likely to happen.
I haven't heard AMD talking about it though.
EDIT: re: So would it make sense to somehow start pushing a Mobile/DTR S939
all of a sudden, for a market that is close to non-existent? My answer would be a clear "no".
You have a point there too. Guess we'll have to see.
Regards,
Rink
Keith, re: Now that´s a stretch. Not only that there aren´t any DTR X2s from manufacturers that I´ve heard of before (and I have never seen a single one in stores), and now underclocking and undervolting one is supposed to help AMD´s market position, but probably only in the DTR segment.. Thanks for explaining your reasoning, but this really sounds like grasping at straws instead of a relevant view from a market perspective.
I think it's well within technical possibilities (not likely a stretch in this department as current mobile A64s are only one speedbin down at most). I think that a Mobile A64 X2 therefore has a reasonable chance of happening (say a slightly over 50% chance).
So where exactly in my reasoning do you think this is 'grasping at straws'? Do you think AMD will let this technical possibility unharvested? Or do you think it's technically impossible? If it's the latter I surely don't see what makes it technically impossible, so could you please explain?
Regards,
Rink
Keith, re: there is no "Mobile Athlon 64 X2 4400+".
I did call it 'slightly hypothetical' for a reason (reading your reply I think I should have expressed myself more clearly): I talked about clocking down and volting down a DESKTOP A64 X2 4400+ to perform like a hypothetical MOBILE A64 X2 3800+. That's useful in the DTR segment.
Regards,
Rink
DC, Dempsey's FBDIMM >1000 pin northbridge chip consumes a HECK OF A LOT more power than 20W. The FBDIMMS themselves add something in the order of an additional 8W per DIMM and need active cooling.
Regards,
Rink
Keith, re: And your hypothetical undervolted desktop parts don´t help AMD´s market position.
A64 MOBILE processors (not Turions) are lower voltage desktop parts. So it does help with AMD's market position, but probably only in the DTR segment. Still useful.
Regards,
Rink
Keith, yeah, not everyone has the same flaws but we sure all have some. Tx for providing a bit of contra weight. OTOH the A64 X2 3800+ is a desktop voltage part. A only slightly hypothetical 4400+ that is volted down might help quite nicely to get close to par with this particular version of Yonah measured by Anand. So Pete/Petz's points might actually not be entirely out of wack. Lastly although it's not useful now DDR2 might help AMD based systems towards slightly lower power when it's actually released onto the market.
Anyways, I thought power consumption of the 3800+ and Yonah were quite close at idle, and that the difference between idle and load was quite close for Yonah.
Regards,
Rink
Re: does Intel break out revenue from Itanium chips separately?
Are you dreaming? Ofcourse not!
Regards,
Rink
Chipguy, re: Time period and source please [for Rink's statement that High Performance revenue has been declining in a growing market].
Anual report SGI. YoY. High Perf. rev. includes both old MIPS and new IPF. Their anual report does not split it out afaict, but it's hard to imagine it either in any other line.
Rink wrote: If Itanium sales are so great at HP why doesn't HP show it's growth more than compensates for the combined decline in their PA-RISC, MIPS, and Alpha systems?
Chipguy wrote: In its latest quarter [Q4] it said its BCS sales grew
17% sequentially compared to 8% for ISS sales. You
do know which HP server families fall under BCS and
which under ISS? Kind of ironic given all the hype of
HP's Opteron sales on this forum.
Yep. But also: On a year-over-year basis, industry-standard revenue increased 12%, networked storage revenue grew 17% and business-critical systems revenue declined 1%. See here: http://www.hp.com/hpinfo/investor/financials/quarters/2005/q4.pdf
Both are important. ISS is only showing signs of healthy growth, while BCS shows signs of fatique besides the more healthy ones you pointed out. Also keep in mind that Q3 BCS results showed a 10% decline QoQ, but a 8% increase YoY which puts your exclusive focus on QoQ growth for Q4 (last quarter reported) in a different light.
Whatever way I look at it I simply can't conclude that IPF will go beyond largely replacing the PA-RISC,Alpha,high end MIPS group. IPF isn't exactly moving down the chain, while with 4-8 socket Opteron earlier, X3&Horus&ServerworksChipset now, and the announced 32 socket AMD solution in the future, x86 is rather likely to move up the chain. Power looks healthy to me. Sparc remains vulnerable but will imo keep being chosen for it's software base.
Regards,
Rink
Chipguy, re: 1) Alpha and MIPS sales are now at trivial levels, probably
on the order of a few ten millions of dollars per quarter
combined. HP will not sell *any* Alphas after fall 2006.
SGI's High Performance Revenue [MIPS + IPF} has been declining in a growing market. If MIPS is trivial that means IPF rev at SGI has been declining.
re: 2) HP's HP-UX sales are approximately flat which indicates
that growth of IPF sales for running HP-UX approximately
equals the decline of PA-RISC sales. But more than half
of IPF sales has nothing to do with HP-UX and AFAICT that
proportion is increasing.
Standard servers were up more than ESS.
re: Combine 1) and 2) and it is clear that IPF sales growth is
much more than the decline in those three RISCs. Intel's
internal research shows that IPF market share growth comes
from: 40% friendly RISCs, 40% competing RISCs (POWER,
SPARC), and 20% from x86.
I can't trust internal research info from Intel. Also even if true this is meaningless unless you didn't mean 'market share growth' but simply 'market share' instead. OK let's presume that. x86 servers have simply continued a long growth trend. That trend does not show ANY impact that IPF is stealing share from x86. If it does than apparently x86 is stealing share from IPF,PA-RISC,Alpha,high end MIPS at a similar rate (or more in general, how much of PA-RISC, Power, Sparc, x86, etc.. market share comes from IPF?). Intel can claim whatever they want; imo it is really largely meaningless until anyone can claim solid growth of IPF/PA-RISC,high end MIPS,Alpha higher than the rate that the server market is growing. HP's website shows a large amount of Alpha and MIPS systems that can still be bought. I can't take your word for it that they amount to insignificant sales, and HP doesn't break it out. If Itanium sales are so great at HP why doesn't HP show it's growth more than compensates for the combined decline in their PA-RISC, MIPS, and Alpha systems?
Re: WTF? Nearly every IPF server sold running Windows represents
a sale taken away from x86. MS is a business and its choice
of PowerPC for Xbox 360 clearly shows it will *not* support
Intel products in any segment where it doesn't make sense on
its own merit. MS plans clearly show that IPF is an important
platform for Windows in scale-up and back office enterprise
servers now and going forward.
See above + Windows on I2 is not necessarily a win over x86, it's only an indication. MS clearly supports AMD64 over IPF valuing AMD64 rightfully many many many times more important for their business than IPF. It's one of their upwards growth opportunities that they take serious.
Regards,
Rink
Chipguy, re: That is pure BS. The closest thing to the captive system replacement sales you imply is IPF sales to run HP-UX and VMS. That represented less than half of IPF sales in Q3. The majority of IPF sales are on Windows and Linux which can't be considered captive in any way.
Sorry chipguy, I did not mean to say that. I said IPF is only replacing Alpha, PA-RISC, and SGI/FS-MIPS architectures, and that combined group isn't growing much at all. Sure they take some wins away from Alpha/Power/x86, but Power/Alpha/x86 are taking wins away from IPF/Alpha/PA-RISC/high end MIPS too. The result is that IPF, Alpha, PA-RISC, high end MIPS as a group aren't growing.
LOL, from what I have seen IPF bashers have changed their
tune little from when IPF was selling in the range of $8m
a quarter four years ago to $619m a quarter now. I suspect
in 2 or 3 years when IPF systems are selling at >$2B/qtr
the same idiots would proclaim "itanic is dead" even if
Intel announced it was making over a $1B annual profits
from IPF MPU sales.
That is pure BS, as was your previous quote above. I NEVER EVER said that Montecito/Montvale were loosers. When McKinley came out I did not say it was a looser either. When Madison came out, idem dito. Now the same with Montecito. My previous post was in line with this as I said that even after the delay and elimination of Foxton that Montecito would still be "pretty good" (but not good enough to unseat Power; Sparc is more a question but customers buy Sparc for it's software and seem to have reasonable trust in SUN's roadmap so I think Montecito might not even win that much sales from SUN despite the performance delta). I did say between Merced and McKinley quite a couple of times that there were strong indications that it could eventually die (and there were; even Intel claimed it was only a development vehicle at the time). Between McKinley and Montecito Itanium has improved but not at an amazing rate. You simply have no basis to call me an 'IPF basher'.
Because your posts reflect a close to blind believe in Intel's Itanium projections they were rather very often overly optimistic on introduction dates, performance estimates, frequencies, features, etc... About everyone here knows that. You can blame Intel for that, but you can't blame me for questioning those past predictions (or future predictures for that matter).
The reason why I reacted to your post is because you really quite often claim very healthy growth rates for Itanium while not considering current growth was largely predestined because of the demise of Alpha, PA-RISC, high end MIPS and resulting agreements to transition those over to Itanium, mostly before it ever existed. The combined Alpha/PA-RISC/high end MIPS isn't growing much at all, and Itanium growth is not amazing ANYONE. It's only ok. It hasn't visibly taken any growth away from high end x86 either (also something I remember you projected would have long time ago have happened on RWT).
Please keep in mind it's only one part of this topic that I disagree with: As stated at least three times before over the years I liked your Opteron performance estimates prior to 2003 much better as they were much more spot on (@RWT). I continue to appreciate your IPF info as you know much more about it (especially but not exclusively the technical bits) than me. Really the only thing I don't like is posts that are imo not complete or overly optimistic when it comes to IPF prospective.
Regards,
Rink
Chipguy, re: IPF is the architecture with by far the fastest growing sales in the server market, deal with it.
Sure. IPF is replacing antiquated architectures from HP/SGI and some from Fujitsu-Siemens. IPF plus the architectures it replaces aren't growing. The reason for that is that IPF so far just doesn't really impress enough. Montecito was a really good prospect until it got delayed and castrated. Now it's only pretty good, i.e. not good enough to substancially grow IPF + the architectures it replaces. Deal with it.
Regards,
Rink
The fact that Niagara is aimed at a small niche of the market is the reason why I haven't invested in SUN yet. Opteron alone doesn't bring SUN enough margin, but it'll keep them alive (Xeon would have done more or less the same). I wonder though how 8 socket Galaxy will perform. Still it will be some years till the Rock design is due.
Regards,
Rink
Keith, that's an overly simplified story as Intel's gross margin won't be affected by the JV because Intel only has a minority share (unlike AMD has now in Spansion, and like AMD will have after Spansion's IPO).
Regards,
Rink
AMD trades in line with others in its sector. As you I think you know Intel is up because of NAND JV with Micron, IM Flash. At the moment Investors are seeing it positively for both Micron and Intel, and as a neutral for AMD.
Regards,
Rink
Dell dismisses talk of AMD PCs as 'rumor': http://www.siliconinvestor.com/readmsg.aspx?msgid=21905131
So maybe not for the near future...
Regards,
Rink
Intel performance suffers again. Dempsey latest victim: http://www.theinquirer.net/?article=27789
Intel better hope the sweetspot is at 3.2GHz or above because performance suffers uhm 'nicely' when at 3.0GHz or lower. Somehow I think that you might agree though that this is too much to hope for.
Regards,
Rink
Chipguy, if any Itanium product was included in ISS wouldn't it be logical if that was restricted to the recently terminated I2 workstations? I2 servers seem to be rather strongly linked to BCS.
Regards,
Rink
AMD shows off Socket F at Supercomputing
And some killer prototypes on view
...Sources said MSI has a Socket F dual board with two HTX slots likely to [be] available in February...
http://www.theinquirer.net/?article=27776
Quite nice if true.
Regards,
Rink
Chipguy, re: A mixture of old news and misrepresentations.
We'll see. To be honest I'm not that sure if Montecito was cancelled. It was implied in a recent Inq article, which aren't exactly always right.
re: SGI has been hurting but nearly as much some here like
to portray
BTW I think there's a somewhat significant chance that you're right about SGI. Montecito/Montvale delays are rather likely to hurt SGI though. Despite the new products I expect their profits to decline because of this (Montecito/Montvale is just that much more powerful than Madison that significant purchases of these large I2 systems are likely to be delayed and hence SGIs profit margins will erode).
On the low end: HP killed I2 workstations because they didn't sell well enough. There's no reason to expect SGI's I2 workstations to be much more successful than HP's (none that I can see that is).
Concluding I think this trend of lower profitability that I expect can only possibly be reversed Q4 '06 at the earliest (first full quarter when Montecito/Montvale is available). SGI is an innovative company and will do nicely eventually when Intel doesn't screw up again.
Regards,
Rink
Have you seen Presler's performance already (955!!)? http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=zh_en&trurl=http%3a%2f%2fwww.hkepc...
Regards,
Rink
(EDIT: Posted this link to Smooth already in previous post)
Smooth, re: Yeah, what is that anyway, the AMD performance rating value?
Well, that would definitely be on their wish list... Besides it would compare Xeons performance with AMD desktop processors (as Opterons don't have a PR rating, just a number).
BTW have you seen Presler's performance already (955!!)? http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=zh_en&trurl=http%3a%2f%2fwww.hkepc...
Regards,
Rink
Major Chinese Itanium customer dumps HP for IBM http://www.computerpartner.nl/article.php?news=int&id=1924
P5 over Itanium. China's State Administration of Taxation. The deal represents IBM's largest Unix installation in Asia and is in the top 10 of largest Unix installations worldwide.
I know, just an anecdote...
Regards,
Rink
Chipguy, one embarrassing thing about the 100TFLOPS 10K Opteron SUN supercomputer in Japan is that it doesn't come from NEC (like the Earth Simulator), Fujitsu or any other capable Japanese firm. Another is that it comes from SUN that apparently chosen despite the fact that it has to go out of it's league (mid to lower high end systems) to create this system. Also apparently Itanium and Xeon based systems weren't capable enough to ensure a win.
More close to home Montecito is cancelled, the production of Montvale will only start this summer according to the latest industry snippets meaning systems end of Q3 at the earliest, Montvale isn't 65nm anymore (but 90nm), and Montvale won't ever use the Foxton technology you liked so much either. SGI's hurting like hell, while Cray finally found it's line upwards last couple of Q's. Besides that Paxville is really hot and underperforming, while Dempsey is reasonably hot too (also considering the 1000+ pin chipset and actively cooled memory), and while performing reasonably competitively it doesn't scale as well (without X3) as Opteron (without Newisys' chipset help) which is especially significant when it comes to virtualization once implemented by AMD ~Q2 '06.
Regards,
Rink
Smooth, did you see that Number column?
Getting close to 8086...
Regards,
Rink
Doug, re: EPS would be adversely affected in 2006 by $0.66 due to the sale of its Spansion division.
You replied: Considering AMD will report 40% vs. 60% of Spansion's earnings as its own, post-IPO, 'Sumit' apparently expects that 20% of Spansion's earnings amounts to $0.66/sh, meaning s/he thinks Spansion is going to earn $3.30/sh in 2006. AMD's 40% amounts to $1.32/sh in 2006 from Spansion alone. Whoops. Sumit needs a refresher course or two.
EXACTLY my thoughts as well. Noone in their right mind expects spansion to earn $3.30/sh next year. Plus like you point out AMD will not exactly be 'adversely affected' at all when compared to this years disastrous impact Spansion had on AMD.
In fact I'd really hate that if Spansion would be able to churn out that ridiculous $3.30 next year because I don't have access to the IPO. I think ~60c/sh is way more reasonable.
Regards,
Rink
Keith, I misread. Seems our expectations are reasonably in line then. Mine were based on recent notebook market share of 11% and I simply added ~4% that I thought was reasonable considering good competitive position this Q and only somewhat reasonable competitive position next year.
Regards,
Rink
Wbmw, 64b Turion is marketable at lower prices against DC but 32b Yonah. Whatever you think significant amounts of consumers will buy 64b laptops over 32b ones especially if they're comfortable enough with the price. AMD won't be capacity limited in 2006 so they'll probably grab the opportunity to increase market share, also in T&L. I think Keith' ~11% is significantly conservative. I'm thinking ~15% is more likely.
To be complete: Desktop will be AMD's stronghold during 2006 in the consumer space. Opteron will continue to significantly outperform Xeon at least most of 2006, and for years after that when it comes to virtualization (Intel won't have MC on die or CSI until AFTER Tigerton).
Regards,
Rink
Joe, Mike, older Geodes that are taken over from NSM are fabbed by TSMC. Newer ones (based on K7) are fabbed by UMC.
Regards,
Rink
wbmw, re: Virtualization Yonah VP - A64 Pacifica.
Rink wrote: Yonah is going to include Vanderpool virtualization. Isn't the usefulness of virtualization in this case more limited by the 32b memory constraints (when compared to 64b virtualization hardware that AMD will have Q2 '06)? I take it that this is the case but am not entirely sure.
Wbmw wrote: You might be confused on what the virtualization acceleration really does. You don't need 64-bits to make it possible; you can manage a few virtual environments with just a little bit of memory (1-2GB). I think the flat 64-bit address space will accommodate larger virtual environments, but this isn't the usage model for first generation virtual environments.
Every virtual environment will need memory. If you have two virtual environments running two different OS's in 2GB total memory it's going to be amazingly tight for the applications that need to run in the two virtual spaces. Virtualization was mentioned in that article along with low power and 64b as criteria for choosing A64. Apart from the adress space argument AMD's version is said to be ahead of Intel's Vanderpool.
Still doesn't mean ofcourse that AMD didn't simply won on price. That's quite possible and even somewhat likely.
Wbmw wrote: AFAIK, 6-7W is the limit for passive cooling. Everything else, including the 9-10W range needs a fan.
The article mentinoed the 9W A64 is fanless...
Regards,
Rink
wbmw, re:
Reasons the article mentions for picking Athlon 64 are: 64 bit, low power, and upcoming virtualization. Especially the last will not be possible with Transmeta's Efficeon fafaik.
Secondly two questions I have:
- Is that 1.6GHz Dothan that only consumes 10W TDP that you are talking about really fanless?? I can only presume that the 1.2GHz Dothan at 5W TDP is.
- Yonah is going to include Vanderpool virtualization. Isn't the usefulness of virtualization in this case more limited by the 32b memory constraints (when compared to 64b virtualization hardware that AMD will have Q2 '06)? I take it that this is the case but am not entirely sure.
Regards,
Rink