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.
Re: AMD told Anand there will be no new cpu architecture in 2006
Jerry Sanders and Dirk Meyers laid out AMD's current CPU architecture, AMD64, dual core, Hyptertransport, 14 stage pipeline, and on-die memory controller in 1998 and it hasn't changed since.
Seems to have been a real winner, too.
Maybe dumping everything, reversing course, and starting over from scratch every couple years (the way Intel does it) isn't... Ideal.
At least now they seem to have given up and are just copying whatever AMD does.
Mayber that will work out for them.
edit duplicate
Re: Shipments of the Itanium 2 processor grew 170 percent from the first quarter of 2004 to the first quarter of 2005.
And use of the smaller boxes let them qualify for lower postage rates, too.
Re: Man, they must really be dense to stick with a process that offers no benefits and has so many draw backs...
LOL !!
And they're left stuck with numbers like cinebench results. Judging by these test results, one of today's X2s isn't quite as fast as two of next year's Sossmans. (note how carefully they exclude any single socket results for the new Intel chips, or any single socket, single threaded results for the AMD chips, so you can't compare directly. Look at the single core sossman results to get a reality check).
http://www.computerbase.de/news/hardware/prozessoren/intel/2005/august/idf_benchmarks_sossaman_yonah...
If AMD would just lose the SOI, maybe they'd be 3 times as fast as Intel, instead of only twice as fast.
And the new Cinebench is apparently a favorite at Intel, too:
The benchmark application CINEBENCH, used internally by Intel to test the newest generation of CPUs, is also available in 64-Bit and lets proud new owners of 64-Bit systems put their computers to the test.
http://www.planetamd64.com/lofiversion/index.php/t7030.html
Re: despite the availability of SSE2, there are times when you need to just deal with a set of coordinates with operations in integer and floating point where using MMX and 3DNow instructions will be a bigger win than using SSE2 to operate on four objects when you only need two.
Do you think it would it make sense for AMD to extend 3DNow? Say to extend concurrency? AMD seems to be far enough ahead that they can afford luxories like that.
Re: Third, the "name of the game", so to speak, has been changed (flanking strategy). It's call platformitization, and
BZZZZT. Wrong answer.
That was last months final strategy for the next 5 years.
This month's final strategy for the next 5 years is encourage chipsets from other vendors to reduce capacity blips:
THERE IS a strong indication that Intel will leave the lower segment of the integrated graphic market. This will cause a great market share for companies such as Via, SIS and ATI to jump into with their offerings and sweep up some of the market share for them. Nvidia is still on hold, as it doesn’t have an IGP chipset for Pentium 4 CPUs.
Re: Let's just say I've gone from optimistic to agnostic. Montecito looks great, but Montvale barely fills the gap until Tukwila.
Face it, it isn't just dead, the body's starting to smell.
CPUs are going to AMD64, from now on, with some coming from AMD and some coming from Intel.
But AMD64 is it for the foreseeable future.
Sun's seen it.
HP's seen it.
Even Intel's figured it out (though they had to rename "Intel Architecture 64" to IPF so they wouldn't have to call their reverse engineered Athlon 64, AMD64).
EM64T brings to mind all the "EM" "T" promises Intel made to companies like HP and SGI about Itanium....
Re: the Athlon 64 micro-architecture will remain unchanged throughout 2006
That's right. AMD laid out a roadmap in 1998 that included AMD64, dual cores, on die memory controller, and hypertransport and they've been following it ever since.
Intel went from "the only 64-bit we'll ship will be Intel Architecture 64 (IA64)" to "Netburst is our core for the next 10 years and will take us to 10ghz" to "we're going to copy what AMD did and we'll copied all aspects of their 1998 design by 2007/2008!"
Re: From your own link....
Now read a little further down the page and you'll find the GX, like Dothan, comes in several versions. Two of the GX models use less than a watt, and that includes the on die 1280x1024x32 million color graphics controller.
And it's in production now, used in (among other devices) winterm workstations:
http://www.wyse.com/products/winterm/S90/index.htm
Meanwhile, you guys are giddy that Intel will have something similar, but without graphics, a year from now.
Re: Intel fixes EM64T
That's the problem with getting behind in this business. While the other guy's moving forward, you're doing maintenance to copy the other guy's changes, then you have to do regression testing against anything new you try to add and so on. Then you're stuck fixing whatever that turned up.
Meanwhile, the other guy is moving ahead, creating new problems for you.
AMD had to deal with this challenge for years until Intel screwed the pooch with "Intel Architecture 64" Itanium, Prescott, Rambus, and no SOI.
Unless AMD does something equally moronic, Intel may never catch up.
Re: By the end of the decade, Otellini said, Intel would have another new architecture running x86 code as low as 0.5 watt
End of the decade?
LOL !!!
AMD sells those now. They're in production. Under 1 watt TDP including on board video and I/O controllers. Depending on how much the video controller is using, the CPU is likely under 0.5 watt in the GX466.
http://www.amd.com/us-en/ConnectivitySolutions/ProductInformation/0,,50_2330_9863_9864^9865,00.html
Re: The "Enhanced Intel Deeper Sleep" technology,
So, how many years before Intel catches up to what AMD's shipping today. These are generally available production systems.
4-core, dual socket, 64-bit server including 1 terabyte of storage and 8GB of RAM that draws 145 watts at the wall.
You can put 144 of these systems in a standard rack without creating thermal density issues. That's 576 cores, 1.1 terabytes of RAM, and 144 Terabytes of storage in one rack.
Rackable Systems' new C2002 server, winner of the Best Server award, draws an astoundingly low peak of 145 watts in a configuration featuring two dual-core 1.8 GHz AMD Opteron HE processors (four total cores), 8 GB DDR400 RAM and one terabyte of SATA storage leveraging Hitachi 500 GB hard drives, with a mere 500 BTU/hour of peak thermal output. Mounted back-to-back in Rackable Systems' cabinets, the C2002 provides the greatest possible efficiency in power consumption and heat evacuation.
http://issj.sys-con.com/read/117897_p.htm
It really looks like Intel found a 2 year old AMD roadmap, crossed out "AMD", pushed the dates a couple of years out, the wrote "Intel" at the top.
I wonder why AMD's taking so long - are they just toying with Intel?
Chartered’s 300mm ramp goes mad!
Tuesday, 23 August 2005
Chartered Semiconductor is attempting something no other foundry has been able to achieve to date! With the big four foundries (TSMC, UMC, SMIC & Chartered) all having produced second quarter results and provided some forward production projections, one big thing sticks out a mile!Chartered is projecting that its 300mm facility, Fab 7, will have gone from 4,133wspm in the second quarter to 22,800 wspm by the end of the third quarter this year!
According to our in house "300mm Activity Report" data, Chartered, if successful, would have achieved the most aggressive ramp of any 300mm fab to date! That means easily beating its rival foundries but also obtaining a much higher ramp rate per month than that achieved by memory manufacturers such as Samsung, Infineon, Inotera and Powerchip.
The nearest competitor ramp rate has been that of TSMC's Fab 12. In 3Q03 it ramped from 4,000wspm to 13,400wspm by the end of 4Q03. Chartered's ramp is almost double that achieved by TSMC!
It has been well publicised that IBM will be using Chartered for 90nm and 65nm chip production this year. Production for IBM is dedicated to Chartered's Fab7, so the big push must be coming from IBM, and what a push this is!
However, as evidenced by TSMC with both Fab12 and Fab14—after an initial ramp spurt—the ramp rate started to slow quickly. The momentum of customer demand can therefore reach a peak quickly. Continued ramp momentum from other potential customers did not materialise.
In Chartered's case, this may not actually happen! The reason for this is that starting in 2006, AMD will also be using Fab 7 and we expect wafer starts to be significant. Should that ramp prove successful Fab 7 could reach capacity in 2006, barely a year after first silicon. Chartered's revenues from 130nm and below technology are forecasted (see blog story "Foundry sales by feature size set for surprise in 05") to reach $ 327 million (30 percent) of sales by year-end. On a percentage basis, the 30 percent of revenues could jump through the roof in only a few more quarters, beating all other competitors.
Chartered is also on track to catch UMC's 300mm capacity in early 2006. At the end of 2005 UMC will have a combined figure of 36,000wspm from its two 300mm fabs, while Chartered could reach 30,000wspm from Fab7 in the same period.
The turn around story at Chartered Semiconductor looks like happening sooner than many has predicted!
http://www.fabtech.org/index.php?option=content&task=view&id=574
Re: Yawn, same old rerun
How about something LIVE - actually available shipping systems!
http://www.hpcsystems.com/datasheets/ds_a5880HS.pdf
A 16-core single memory image system for $25K. If you pay for air freight, you can have it up and running in your server room in less than two weeks.
That's what Intel is facing NOW. When will they have an answer for it? What will they be facing by then?
Re: remember the ODMC added ~33%
Wanna run that one by us again?
More like 3% to 30%, depending on the application, with the gain being around 5% to 8% in the vast majority of applications.
Re: it's looking like both 45nm and on-die memory-controller might be coming to a Conroe somewhere near you in 2007/8
AMD will have been refining its long in production quad core and asymmetric core CPUs by that time frame.
AMD will be discontinuing as obsolete the parts that Intel will be struggling to introduce.
Re: That's impossible! We have been told for years that IPF will deliver twice the price/performance as x86.
Now they're telling us that Conroe/Merom will deliver twice the performance/watt of Dothan.
Don't tell me you're skeptical !!
Re: If you were to ask me a year ago, before Tukwila and Montvale got downgraded, before details on Merom and Woodcrest began to emerge
You never learn, do you? Next year, when Merom and Woodcrest get "upgraded" in power use, you'll be waxing poetic about whatever snake oil Intel's peddling for the following year.
Re: The latter seems to be where all the 'Dro^^^, I mean AMD speculators, have ended up.
Meanwhile you're delerious with joy over Intel's announcement that they're going to try to copy AMD's roadmap from 2 years ago.
Whoopee !!!
In 12 months or so Intel hopes to ship a 14 stage pipeline chip that can run the current 64-bit windows (Conroe/Merom/woodcrest). 2 years after that Intel will add on die memory controllers and do something about the single FSB bottleneck. Stuff that AMD has been shipping, literally, for years - in volume.
By the time Intel matches what AMD was shipping last year, AMD will be delivering quad core and asymmetric mulit-core chips.
And Intel still can't figure out how to do SOI....
Re: Quadcore running at
It was two dual core parts in a dual socket board.
AMD will have quad core next year, Intel won't have it until 2008:
Intel quad core not before 2008
In a post-keynote briefing to reporters, Intel vice president for the Digital Enterprise Group, Stephen Smith, provided some further details about the new architecture. He said Conroe's core may be given as few as 14 pipelines, as opposed to Pentium 4's current 31. He also said Conroe may, possibly, take on as many as four cores by early 2008, though no formal decision has been reached. When pressed about cache size, Smith indicated a possible 8 Mb cache for Conroe, but again, would not officially confirm.
http://www.tomshardware.com/hardnews/20050823_133123.html
http://www.siliconinvestor.com/readmsg.aspx?msgid=21630661
Re: Can you think of any reason why they shouldn't do this (assuming they can continue to get good margins on Xeon MP)?
Xeon faces direct competion, Itanium didn't. You get better margins if you have a monopoly on your segment.
Itanium was losing even against indirect competition, so it's something of a moot point.
Looks like Itanium is augering in.
The entire argument for Itanium pretty much resolved down to one thing:
It was to be the only commodity 64-bit processor and, as such, the only one that would benefit from economies of scale.
If you were going to be satisfied with an obscure high end architecture, it would have been better to go with Alpha - a lot of code already existed for it. But Itanium was to benefit from being the only 64-bit CPU in very high volume production.
And then suddenly, it was over. Had Intel managed to keep everything else either proprietary and super expensive, or 32-bit, Itanium would have been a huge success. Intel managed to scare HP into killing off Alpha and PA-Risc, and almost got IBM to fall for the Itanium bogey man story, too. Intel had everyone believing they'd never ever ship a 64-bit X86 part except for Mike and Charlie at the Inquirer.
Itanium got THIS close to making it !!
Then along came AMD64 and it was just a matter of time. You can buy a 16-core Opteron box for $25K that will run pretty much ANY software out there. Unix, Windows, Linux, Solaris and myriad applications for each OS. You can develop for it on a $799 Turion Notebook - on the road.
What kind of Itanium system can you put together for $25K, and where will you get the software to run on it? What will you develop the software on?
Itanium is just pointless, at this point.
Re: it would be a good time to demo a 65nm quad core Opteron running at 4 GHz and burning 20 watts...
Hey !
That's supposed to be under NDA !
Re: <i.explain where 31W comes from
Amps * volts.
And the early roadmaps for Dothan had it as a 21w chip - as you well know.
Meanwhile, some commentary on what's been going on in the real, as opposed to the theoretical and/or future world:
http://moneycentral.msn.com/content/P125123.asp
Re: We need a 17" model from an OEM that really matters, like HP or Acer
7.7lbs for a 17" screen model is pretty good. I've talked to some MSI distributors about bringing in the S270 (another darned good looking Turion systems from MSI) and apparently MSI is still trying to work through an OEM.
Maybe it will show up as an HP, maybe it will show up as a Gateway, maybe it won't ever make it out of the Orient, and maybe it will show up as a Dell.
Re: Please elaborate as to which Intel fabs currently capable of making chipsets are making flash instead.
The following Intel FABs are mixed use for flash and logic (chipsets and com parts, since these Fabs don't make CPUs). They are all explicity listed as being dual purpose FABs producing both flash and logic.
FAB D2
FAB 10/14
FAB 8
FAB 18
FAB 11
http://www.intel.com/pressroom/kits/manufacturing/manufacturing_at_a_glance.pdf
Re: the center is putting much of its efforts on dual-core technology products, which are expected to take off next year
Will it really take that many new staff to handle the addition of AMD parts to their line up?
Oh well, at least now they have the folks they need to add more skus and to do more in house design that doesn't rely on Intel.
Re: I have direct hands on experience, with cutting edge semiconductor manufacturing process technology, and you don't.
Not with one that isn't seeing its competitor gain server market share at a 570% annual rate, you don't.
Re: the Widely Reported issues that IBM/AMD have been having with their similar SOI process in Fishkill
You mean wiping the floor with Intel in the server market because the IBM/AMD process delivers somewhat faster chips that use 1/3 less power?
It's let them increase server share 89% quarter on quarter. That's a 570% annual growth rate in the server market.
And a pretty nice "issue" to have to deal with.
Re: everybody was assuming that Prescott was a watt-monster because of the process
Intel said in their last conference call that power envelope limits precluded them from adding 64-bits to Yonah.
Meanwhile, 64-bit Turion chips from AMD are delivering 4-hour battery life in notebooks with Radeon X700 gaming level graphics.
This when Intel's addition of adequate graphics to its chipset has users complaining that recent Centrino models have dropped from 4 to 2.5 hours.
http://www.pocket-lint.co.uk/review.php?reviewId=1144
I think you're confusing Intel's chip, and its process, with special low platforms.
Re: Semi has you on ignore, just like everyone else. You're a troll
LOL !!
This from a professional poster who does nothing else but work the boards all day every day.
Who monitors your boards for you when you're on vacation?
Re: 65nm in less than 2 years, well, I don't expect it.
The jury's still out on that one (there certainly aren't any 65nm systems out there yet). But I'll certainly agree now that something called 65nm looks like it will ship a couple of months ahead of that 2 year deadline, beating it rather than following it.
But given all the "just wait for 45nm" talk we're hearing from Intel, lately, it may or may not be a big gain.
Re: Centrinos may never reach their power goals due to process problems...
And they didn't. They were supposed to have a TDP of 21w and it's 26w (31w if they hadn't changed the T from "Total" to "Typical."
Re: Certainly, something could go wrong, but I wouldn't take a roadmap identical to Athlon's as a bad sign...
Yep, it's taking longer than hoped as Intel "re-invented itself" as a full blown criminal organization and extorted AMD out of a number of markets, so UMC wasn't needed since Intel was blocking market access anyway.
But things are moving along. AMD now has nearly 30% of the profitable 4-way and above server market, and is gaining ground steadily in the other segments, including mobile, despite the despicable, criminal actions of Intel.
Having Intel enter the 4mbit to 16mbit flash market (formerly Spansion only) by bombing prices from $6 down to $1, even though it meant big losses for Intel wasn't expected either. But Intel is paying in interesting price - Intel is giving up $30 chipset sales in order to sell $2 flash parts for $1. the resulting shortage of chipset has left an opening for AMD's Turion to bypass Intel's extortion.
Brilliant management by Intel....
They can't even cheat without screwing it up.
Intel was handed a goldmine of a monopoly by IBM 25 years ago. Leave it to greed and stupidity to muck it up.
Re: it was too expensive to design for both lines...
What a load of crap - they buy boards from Asus, Foxconn, and Microstar just like anyone else. Dozens of different models. To have one of them be one that supports AMD adds zero to their costs.
Re: No surprise that AMD is relying on the two biggest fish, Microsoft and the federal government, to succeed where they couldn't.
It's not just Intel - bookies, hitmen, and drug dealers just hate it when somebody calls the cops. Interfering busybodies who hate "competition." Maybe the four groups could form some sort of lobbying association.
Nice to know Intel is proud that it's become a member of Local 12 - the Villains, Thieves, and Scoundrels Union.
Re: Memory performance
It shouldn't have made any difference if the same number of loads were on the bus. Did you go from single sided to double sided DIMMs? (I don't think there is such a thing as double sided SODIMMs).
Many bios's adjust timing down when adding more loads to the bus, but swapping two SODIMMs for two other SODIMMs shouldn't add any additional loads to the bus.
Could the new DIMMs just be slower RAM?
Re: For the past 2 years, Intel has had to make due with what they had
Why is any of this any different from 2 years ago, and 2 years before that?
For years, the industry had been letting Intel set the standard for sockets (which give them a big advantage in timing and technology choice) and AMD still came out with better parts - so Intel blocked AMD from sharing the "industry standard" Intel socket designs.
So AMD came out with a better socket (with a lot of help from DEC) leaving Intel even further behind (now AMD's better chips had the benefit of a better socket).
Intel tried to foist a crappy, unreliable, and expensive (for everyone but Intel) RAMBUS memory standard on the industry. AMD replied with a better memory standard (DDR2- which Intel claimed couldn't be made to work) and Intel fell even further behind.
AMD pioneered mass production of highly power efficient SOI as a process, and Intel fell even further behind.
AMD pioneered AMD64 and Intel (and its IA64) fell even further behind.
AMD developed dual core servers, and Intel fell even further behind.
Almost everything Intel has touched for the past 5 years has turned into crap, forcing them to "re-invent" themselves as a criminal enterprise to maintain market share.
Now you say this corporate culture is suddenly going to stop making bad choices and start making good ones. You say Intel will suddenly stop coming up with poor designs and start coming up with good one - and your statement confirms that you agree that the choices Intel's corporate culture has been making for years have mostly been very poor ones.
Yet you claim it'll be nothing but blue skys and sunshine from now on.
Why?
What's changed?
Do you think that GM and its unions are suddenly going to start wiping the floor with Toyota?
Why?
Re: Clearly, at this point in time, AMD is falling further and further behind intel in process technology
You don't make any sense. They may or may not fall behind in the future, but they are currently well ahead.
The only thing AMD can get "further" is ahead.