The AMD Execution Thread [2007 - 2017]

[MTd2]

It's simply a question of getting over all the bad news in one go, and then moving forwards, rather than dragging the bad news out over many quarters.

I really don't buy this one hit and go, as I said before. They wont get rid out of the news, because it is not news, it´s like a desease.

 

[3dilettante]

Going by this, the first Fusion product is now pushed back to 2H 09. That means Nehalem's value product with an on-package IGP may potentially beat AMD's solution to market by six months to a year.

Impressive that AMD can have a ready-bought GPU company and running CPU business and manage to manufacture a lag time of that magnitude.

I can only suppose that if AMD hadn't bought ATI, it would have lagged by even more.

 

[Bouncing Zabaglione Bros.]

I really don't buy this one hit and go, as I said before. They wont get rid out of the news, because it is not news, it´s like a desease.

<shrug> lots of companies have done it. It's better than dragging out quarter after quarter of bad results because of something you bought a year ago. AMD have had a really crappy year, so a bit more crapness now doesn't make any difference.

Imagine they get their act together in Q1/Q2 08, and start getting good results, but those are dragged down by write-offs quarter after quarter. This time next year, instead of people looking back and seeing a crap 2007 followed by an upturn, they are just going to see crap 2007 followed by crap 2008. The trend is what matters to investors, not what happened in the last quarter, and a trend that shows an upturn will enable AMD to get more investement at better rates than a trend which is overall flat or poor.

And as I keep pointing out, if those write-offs are related to loans/interest, it saves them money in the medium to long term.

You seem determined not to believe it, but if you do your own research and spend a little time with google, you'll see that lots of companies do this rather than drag out the pain.

 

[MTd2]

Impressive that AMD can have a ready-bought GPU company and running CPU business and manage to manufacture a lag time of that magnitude

So, maybe 2009 will be the doom of NVIDIA, won't it?

I will search for it later then. I hope I can find it. Still, asking for it at a forum would be appropriate since this strategy don't have a name, so I guess it's easier if you have keywords.

 

[chavvdarrr]

Going by this, the first Fusion product is now pushed back to 2H 09. That means Nehalem's value product with an on-package IGP may potentially beat AMD's solution to market by six months to a year.

Impressive that AMD can have a ready-bought GPU company and running CPU business and manage to manufacture a lag time of that magnitude.

I really fail to get where's the problem just to slap together K8 cpu + dx9 IGP connected through HT or even embedded pci-e...
do it and beat the drum you're first and best...
maybe its time to start a thread "Is AMD d00med" and start making bets when? Really it looks like the question is no longer if, but when

 

[ShaidarHaran]

I really fail to get where's the problem just to slap together K8 cpu + dx9 IGP connected through HT or even embedded pci-e...
do it and beat the drum you're first and best...

If only it were that easy...

maybe its time to start a thread "Is AMD d00med" and start making bets when? Really it looks like the question is no longer if, but when

I believe that thread already exists, and you're reading it now

My guess: they won't make it to 2010.

 

[tangey]

In 2009, the IGP segment will see the demise of ATI, and the Rise of IMG.

 

[ShaidarHaran]

In 2009, the IGP segment will see the demise of ATI, and the Rise of IMG.

And then the rise of the robots in the following year.

 

[vertex_shader]

AMD's Hector Ruiz gets a raise

Company is going deeper and deeper day by day and he gets a raise, this world is crazy.

 

[MTd2]

http://www.theinquirer.net/gb/inquirer/news/2007/12/14/amd-ruiz-gets-raise LOL, at the same time, sorry.

 

[Geeforcer]

Someone did a comparison of "What AMD told us they expected during last year's analyst day VS what did happen in the real world". Can't find a link to it, but needless to say it will take a while for me to take their analyst day presentations with anything short of Everest of salt.

 

[Geo]

In 2009, the IGP segment will see the demise of ATI, and the Rise of IMG.

What do you think Fusion/Swift is about?

 

[Arun]

Well, most of the presentations were fluff and well-known facts anyway, except for Bob's (the CFO). I'm not saying that as being negative for AMD; clearly, their target audience likely wasn't looking for much more than that hearing their questions... (some of the answers, including those from Rivas, didn't exactly make me confident in these guys' understanding of their own business though!)

Anyway, the most shocking part for me was the new roadmap. I sure as hell hope this is some kind of joke, because I've rarely seen anything so screwed up. Just compare the roadmap slides in these two presentations:
- July 2007: http://www.amd.com/us-en/assets/con...07_AMD_Analyst_Day_Phil_Hester-Bob_Drebin.pdf
- December 2007: http://download.amd.com/Corporate/MarioRivasDec2007AMDAnalystDay.pdf

This is officially the 'roadmap downgrade of the millenium'. I don't care what about in the remaining 993 years; there's no way in hell it'll beat this:

• 'Eagle' was replaced by 'Swift' as the Notebook Platform for 2009. The former's CPU was a Bulldozer derivative; the latter's is a Barcelona derivative. The GPU was also downgraded to certainly being a MCM (it was unclear previously) and being based on the 'current high-end discrete'. All this in... 2H09.
• Bulldozer is not anywhere else on the roadmap either. Sandbridge (the 8-core server chip) is gone, with Montreal apparently being the primary focus in 2009. Montreal is completely retarded, with even more cache (1MiB/core L2!) - thus getting rid of potential cost advantages against Intel.
• R7xx has definitely been delayed (or cancelled completely and something else took its place) from mid-2008 to 'later' (aka it's only on the 2009 roadmap, but since we're talking platforms it might also come out in late 2008).
• SB8xx and RD9xx are gone. Slide claims they'll use SB700 for their server 'platform'.

So, in the last week, 2009 turned from 'very interesting' to 'completely boring and retarded', and 2008 turned into a gigantic 'who the hell knows' thing as AMD doesn't even have Shangai samples back yet. Ugh. This new roadmap better be horribly wrong and Rivas better be completely nuts and clueless, because otherwise I don't know what to think here...

EDIT: Woops, I meant R7xx, and indeed SB8xx is on there (I just looked at one place where it was in July, and it was replaced by something else there, but not everywhere, apparently).

 

[silent_guy]

All right. Cite the example. I just can´t find it.

Let's see: JDSU, Nortel, AOL, McClatchy, Corning, ...
And recently: Citigroup, Morgan Stanley, Merril Lynch, UBS, ... Those financials are all writing down loans pretending they'll have a value of zero even if, after the credit crisis, there's still a possibility of getting something for it, but buy getting them off the books entirely, they can only provide upside. (Note that many of those companies saw their stock rising after the write-down announcement.)

If you've never seen this kind of action before, you're really not following financial news...

 

[Geo]

Uttar, do you mean RD8xx and not RD7xx?

 

[Dave Baumann]

I also see SB800 several times in the presentation.

 

[Geo]

I also see SB800 several times in the presentation.

I saw it referenced twice specifically for 2009, and possibly once by inference as "next generation southbridge".

 

[MTd2]

If you've never seen this kind of action before, you're really not following financial news...

I was not talking about just about finance, but primarily about the understimating talk during the presentaton and all their news these days. Not only the admited their faults but also put down the status their CPU line until 2009 by talking about delays. I just can´t see any past marketing strategy that was successful by showing all dirty work at once. The finance statement was one of them, but not all.

Sorry, I am to blame for the confusing thoughts. I admit that I don't write clearly. That´s an issue not related to english, it's the same situation in my mother tongue.

 

[3dilettante]

Eagle' was replaced by 'Swift' as the Notebook Platform for 2009. The former's CPU was a Bulldozer derivative; the latter's is a Barcelona derivative. The GPU was also downgraded to certainly being a MCM (it was unclear previously) and being based on the 'current high-end discrete'. All this in... 2H09.

Bulldozer is not anywhere else on the roadmap either. Sandbridge (the 8-core server chip) is gone, with Montreal apparently being the primary focus in 2009. Montreal is completely retarded, with even more cache (1MiB/core L2!) - thus getting rid of potential cost advantages against Intel.

The delay from the tentative Sandtiger introduction is at least 6 months, since it should have been 2H 09 and is either gone or in 2010.

This isn't the first "delay" of this magnitude AMD has experienced with a new design. It is at a minimum of the same order as the last ~6 month setback, when AMD failed to create a clean-sheet design to replace K8, and we all know what AMD got instead.

My guess is nothing more than numerology at this point, since the only real possible parallel is the length of the delay, but I do find this worrisome.
Barcelona and Shanghai represent a more involved attempt at revamping the uncore part of the chip.
Bulldozer should have been an extension of the changes to system infrastructure already present since Barcelona. It makes me wonder whether there is another layer of complexity in the uncore or if it is once again a core redesign failure.

To top it off, the slides on AMD's 45nm process indicate to me that it at least initially won't have metal gates or high-K gate dielectrics.
I don't expect AMD's 45nm chips to match Intel's thermals or clocks, and without serious tweaking, is potentially not much better in chip speed than AMD's 65nm.

Montreal's die size isn't necessarily a problem due to costs, or at least wouldn't be if AMD hasn't confirmed repeatedly that it can't produce big dies very successfully.
The target market for a big chip like that would be one that isn't as CPU price sensitive, but unfortunately for AMD will look for better performance.

It is possible an 8-core MCM of Shanghai cores might compete in some respects with a Nehalem quad-core with SMT. It would be completely obliterated by the native 8-core monster Intel is promising in '09.

At least with the MCM, AMD can attempt to nip at Intel's heels. Since the big costs of software licensing are often tied to the socket count, AMD needs to have some product that keeps Intel from having free reign in socket consolidation.

Regardless, after Nehalem it is likely that there will be no measure by which AMD's chips are not demonstrably inferior to Intel's.
I'm not even sure if AM3 supplies enough bandwidth for 8 cores, and if does, it is only possible if each core performs about as well as Barcelona does(n't).
Intel's 2H 08 sockets have room to grow, and apparently the top SKUs are tripl-channel or have multiple FB-DIMM links.
Going by socket bandwidth alone, we'd have to conclude that Intels solutions will best AMD's by ~50% when bandwidth is a concern.

(edit: I'm going to back off on this last theory. I've read incomplete data on the pin count for the new socket that Montreal should use. edit edit (and some shenanigans with a new extender?) AMD didn't state the number of channels Montreal might get, hopefully more than 2. Chips on AM3 will still have this shortfall.)

So, in the last week, 2009 turned from 'very interesting' to 'completely boring and retarded', and 2008 turned into a gigantic 'who the hell knows' thing as AMD doesn't even have Shangai samples back yet. Ugh. This new roadmap better be horribly wrong and Rivas better be completely nuts and clueless, because otherwise I don't know what to think here...

Perhaps, if I were being snarky, I'd insinuate that AMD took the big write-down so that the prospect of delaying or cancelling its third attempt at replacing K8 (fifth or sixth at replacing K7)--the product that was to save the company's bacon--wouldn't look so bad.

I am really pissed with this news. It means the tech scene is going to be half as interesting for several years.

I know high-level chip design at deep submicron geometries is incredibly difficult, but damn, this is getting ridiculous.
The length of time AMD has tried to wring performance out of the K7 core is almost Via-esque.

I am pretty disgusted by this continued retreat from even a mediocre showing.
I'm so incensed, I'm hoping for an AMD apologist to post here so I can debunk any remaining excuses point by point.

 

[MTd2]

AMD Hints at 32nm Test Shuttles, Claims 45nm Samples Ready in January

AMD has had more than its share of problems over 2007 with possibly the worst issue being with its new Phenom and Opteron processors. With the current problems with AMD's K10 processors, now the question becomes, "Will AMD make its deadline for the next processor?"

ChannelWeb Network sat down for a phone interview with AMD Executive Vice President of the Computing Products Group Mario Rivas for some more information on the bug and how AMD plans to recover from the torrent of bad press that has resulted.

Rivas says that the bug started as an observation and it wasn’t until mid-November that it actually turned into a more serious bug. Rivas also said that the company tried to do BIOS workarounds and patches with a 90% success rate.

In a closing comment, Rivas details some heavy information about the company's next-process chips. "We have 45nm on the way. We will have initial samples also in January. I'm fairly confident that those puppies are going to boot, and then we can have a follow-up conference call and I'll tell you, 'The sucker is booting.'"

Typically, when a processor is first taped-out, a operating system boot is one of the markers of a successful design. Intel's 45nm Penryn, for example, loaded the Windows XP operating system on the first spin. 11 months later, the processor began shipping for volume.
According to AMD’s John Pellerin, director of logic technology and development, AMD plans to ramp production of 45nm high-k dielectric chips in the first half of 2008. Pellerin says that AMD is more concerned with finding customer applications for its processors rather that a process race with rival chip makers.

Pellerin said in an interview during the International Electron Device Meeting that AMD’s 32nm high-k parts were solidly in the development phase showing that AMD is already looking down the road and hoping to learn from the problems encountered with its early 45nm process.

Rivas closes, "We also have 32nm advance work in SRAMs, which as you know is the initial step. So we will be a fast follower again, and as long as we have architectural advantage, our 45nm will be as good as the other guy's 32nm."

Intel announced its 32nm test shuttle just three months ago in September 2007. If AMD does have a 32nm test shuttle already, the year differential for 45nm may shrink considerably with AMD's 2010 architecture launches.

http://www.dailytech.com/AMD+Hints+...45nm+Samples+Ready+in+January/article9995.htm

IBM Alliances Announce Advancement in High-K/Metal Gate Technology
http://www.physorg.com/news116860211.html

*******************

Question: why weren't these 32nm on the presentation?