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.
IPF is dead. Alpha II. Give it up.
Make that 2 single-core multi-chip CPUs (Dempsey and Presler),
1 dual-core niche-ISA CPU (Montecito),
1 might-as-well-be-like-Presler "side-by-side" CPU (Smithfield),
and 1 possibly true dual-core x86 mobile CPU (Yonah).
I mean, really, after all the PR about being so advanced with dual-core progress, it turns out that 3 of 5 parts are really built from single-core parts! (Okay, quibble about side-by-side Smithfield, but it is clear from the FSB diagram that Intel gained nothing by making Smithfield a single-die.)
AMD should hold up a 4-socket blade populated with dual-core Opterons and claim it's an 8-core CPU sample.
--------------------------------------
The real message here is this:
There are NO true dual-core P4 parts. None. It's all PR. No engineering or design work to speak of. All 3 parts are effectively multi-chip modules, with each CPU having its own FSB connection. Yeah, yeah, on Smithfield the CPUs are "touching". Big deal. The ONLY reason that was done was... guess? PR purposes! Intel couldn't very well claim to be first to dual core if they rolled out something that looked like Presler.
AMD dual cores to be superfast
http://www.theinquirer.net/?article=21531
[...]
The biggest news is that the dual core parts will be fast, very very fast. If you are wondering, the first dual core parts out of the box will run at 2.4GHz, possibly more. The one show in the Alienware box below, not an Alienware product, just a cool case, was a production model. It came in a production package with a production bar code. I am pretty sure that the launch will happen early, I am just not sure how many they will put out in the first round of launches.
[...]
Really? Intel says Pressler will be a multi-chip product:
http://www.extremetech.com/slideshow_viewer/0,2393,l=&s=200&a=146843&po=3,00.asp
Pressler at 5 ghz ? LOL! Good one!
Intel has a ways to go on those dual 4GHz 65nm parts... they're at 2GHz right now. :)
That's because AMD did not make an "announcement". Reuters pulled the language from the 10-K filing that went public today.
It's a bit odd, given Rivet's comment that he "was not changing his guidance" last week. I've asked AMD IR via email to clarify their position on current CPG revenue guidance, and explain Rivet's comment, and the difference in 10-K language and 1/18/05 PR language.
A/H probably also has to do with the Novellus stuff, which is pulling all semis down.
One should also note that Reuters loves to amplify bad news. In this case the difference between "flat to slightly down" and "down slightly" probably amounts to a midpoint change of 1%, or about $7M, given that "slightly" has been used by AMD recently to describe a 2% revenue change, in both up and down cases.
That's not clear. The article says that Fujitsu sold 233 systems, and the others each sold less than that.
You do know what "on the order of" means, right? :)
We're talking on the order of 100 systems/year for those OEMs.
What a moron. There was no such Friday morning news.
LOL! Yup, here's the detail on the remaining vendors:
Dell managed to capture 5 per cent of the Itanium market with 1,371 server sales last year, Gartner said.
[...]
SGI moved just over 1,000 boxes to take 4 per cent of the market, and Fujitsu finished fifth with a whopping 233 systems sold in all of 2004, giving it 1 per cent of the market. Groupe Bull, NEC, Legend, LangChao and Unisys rounded out the top ten Itanium vendors - each managing to convince a few close, personal friends to take some Itanic kit off their hands.
If you feel that way, I'm not sure you have a good grasp of the term "leveling off".
I don't think you'll see "standard" cell-based workstations/servers for at least a few years, if ever.
I think the competition between Opteron and Xeon is generating solutions that attack the last remaining IPF niche: large, single-image systems. There's already no reason to use it in any other case. Even with superior FP performance, it doesn't have superior FP price-performance.
And look, IBM just announced a 64-way single-image chipset for Xeon, and we have similar Opteron stuff coming from Sun/Serverworks and Newisys.
I think that will continue.
The trouble is, the future is going to look like this:
2005 ~$2300m sales
2006 ~$1200m sales
2007 ~$300m sales
2008 ~$100m sales
Why does that surprise you? DDR1-533 far outperforms DDR2-533.
IIRC, the socket convergence idea has been scrapped.
Itanic may grow modestly for another year, after which it will whither away slowly.
That's just silly.
Clearly, this indicates even more strongly just how powerful a threat IPF has become.
- chipguy
-----------------------------
(I thought I'd help him out with his response. :)
IIRC you predicted it would reach higher than 10%
HIGHER than 10%? Would you have a link for that?
So what?
Just another nail in IPF's ever-shrinking coffin.
The number of ARM based systems
You can't run anywhere near the same application classes on ARM as you can on iAMD64 or IPF systems. But you knew that.
By the end of this year, iAMD64 platforms will outnumber IPF platforms by a factor of HUNDREDS-to-1.
IPF is stuck in a shrinking niche.
What a laugh! You try to turn an Intel problem (overcapacity) into an advantage. Bzzz.
AMD, I would think, would be worse off.
Of course you would.
IBM server design drops Itanium support
http://news.com.com/IBM+server+design+drops+Itanium+support/2100-1006_3-5589603.html?tag=nefd.top
You believe that putting all your semiconductor eggs in one basket is the right strategy?
I believe you'll see some growth from the "other" segment, but largely, yes, at this point, AMD is strong enough, producing product in virtually all CPU segments (unlike earlier), with capacity growth coming online over the next two years, to forego the low-margin memory business.
In retrospect, they should have been lining up a Spansion sale during Q2 of last year, to be effective the day after the quarter closed.
Yes, but Spansion wouldn't, and current AMD shareholders would also get Spansion shares in a spinoff. It's a "the sum of the parts are greater than the whole" situation.
Business (as opposed to consumer) desktop and mobile. Dell's bread and butter, basically.
Intel puts Itanium saviour on ice
http://www.theregister.co.uk/2005/02/24/intel_nicks_tukwila/
"Intel has plumped, constrained and then killed a future version of Itanium once meant to save the entire franchise, The Register has learned."
[...]
"What was once Tanglewood - later renamed Tukwila - is dead."
Perhaps poorly phrased, but correct. Besides, I could never hope to surpass your recent set of amusing predictions, from Prescott IPC to the Lenovo-IBM deal.
Short term pain = 1-time charge. Long-term gain = AMD (non-flash) stock no longer has the flash-risk penalty impacting its valuation.
There's more than a memory controller on the K8, though. We're talking about everything that is not duplicated in the dual-core K8.
http://www.geek.com/news/geeknews/2004Oct/bch20041018027458.htm
Intel has released some roadmaps to OEMs detailing upcoming desktop processors, particularly the dual-core Smithfield line. Smithfield will debut in the third quarter of next year, supporting EM64T and the NX/XD bit (a.k.a., EDB--Execute Disable Bit) (see our previous coverage), and containing 1 MB of L2 cache per core. The model numbers for the initial Smithfield processors will be x20 (2.8GHz), x30 (3.0GHz), and x40 (3.2GHz). They will all sport an 800MHz FSB, not the newer 1066MHz FSB. This was expected, if you recall the initial announcement in August, which points out that the initial dual-core offering will lack the bus arbitration unit required to keep the load on the shared bus down, which in turn is required to maintain the higher bus speed.
BTW, I think that a spin-off would be short term gain for long term pain.
Precisely backwards.
some circuits on the smithfield will be shared.
Well, supposedly Smithfield is 2 complete Prescotts plus a little bit of additional wiring between them.
And one of the reasons Smithfield is supposed to suck is that there is no bus arbitration. Both parts attempt to talk to the bus. So I don't see much power savings there.
I see this as a huge win for the consumer.
I see this as a huge ripoff for the consumer. There's diddly that takes advantage of more than one core. The reviews are not going to be pretty from the consumer (read: gamer) sites.
and then listed an opteron at 30W... does that mean the opteron core really runs at just 8W
Of course not, in the LV Opteron case you cited (30W), I'm sure both core and MCT, etc use less power than they otherwise would.
Why not?
Neither would new Intel servers with new chipsets. Ghosting images will break in that case.
Tenchu, a separate memory controller should consume less than AMD's internal memory controller + crossbar + SRQ + HT communications links, don't you think?
I sense some backing off, but look, Paul, I think that with AMD claiming 20W, you need to prove that it isn't 20W, which you haven't done.
You've given an estimate for the external signal driver-related dissipation, claimed that was everything, then said the internal portion was trivial, then more or less admitted it wasn't trivial but claimed it must be less than 12W because... well, because that part of the chip (which does clock at full clock, BTW) isn't as complex as the rest of the core. (probably true, but how MUCH less power-hungry is the issue. 12W vs. 40W core is not unreasonable.)
That's not exactly a convincing proof, especially given your evolving explanations.
We have the AMD server VP saying 20W, and you waving your hands and throwing personal insults around.
Ah, Tom's may be repeating 110W from other articles, not claiming they heard that from AMD:
According to industry sources, Smithfield is rated at a maximum power consumption of 130 watts, compared to 110 watts of Toledo.
Basically, the article is a typical THG mess. :)