The AMD Execution Thread [2007 - 2017]

[Aerows]

R600 makes an eloquent statement as to whether or not waiting for AMD to release a product is a good move. I've had AMD processors for years, but after the fiasco of a product launch, the lies that were told that haven't been answered for and the terrible execution that AMD has shown lately has given me cause for doubt.

Frankly, when I get ready to upgrade to a quad core processor (if I had a reason to do so), I'm going to go ahead and get whatever is available. If that doesn't include a product made by AMD, so be it. "The family launch", "we don't do soft launches" and "best drivers in the business" need to be addressed before I will go out of my way to choose AMD. Heck if an Intel Kentsfield gets released at a decent price, I might just go there anyway. I would not have felt this way last year, but much has changed lately at AMD and it hasn't been for the good, IMO.

They need to start restoring their image, and keep their traps shut if they haven't got something good to say. They aren't as bad as Sony yet, but they are getting there.

 

[Tim Murray]

Last, I want to comment very briefly on R600: while the pessimists and naysayers are already counting R600 down and out, I believe the fact is that the R600 saga is only just beginning. My own opinion is that AMD has a lot more knowledge of its own products than it is being given credit for by some sites, and that very possibly we should be paying more attention to AMD here than to sites like AT, which are, after all, pontificating on the future of products they have no input in with respect to either manufacturing or marketing.

Uh... what? The vast majority of what comes out of AMD is going to be filtered by AMD's public relations department, which exists solely to, you know... sell their products, not act as an unbiased source of information? This is like saying, "All those sites who wrote bad things about NV30? They're dumb! They don't know anything! NVIDIA clearly knows more than they do, so let's see what their PR guys have to say about this!" It simply does not follow at all.

edited: I've edited this a bit because I definitely replied a bit too emotionally at first. I took WaltC to mean, either fairly or unfairly, that the work of journalists is secondary to what information is actually released by IHVs, which I think is completely wrong. So, I got annoyed, went too far and accidentally ignored what Wavey, sireric, and co. do by posting on the boards and talking to us, which is greatly appreciated. Sorry guys

 

[Geeforcer]

No offense (no more then deserved, anyway), but I suggest going back a few years and reading what various people had to say about future of particular companies or industry as a whole, comparing that with what has taken place since and drawing conclusions about their insight based on that.

 

[_xxx_]

While GPU's, using external fabs, are often trailing CPU's on process node the difference isn't that far away (AMD is shipping 65nm GPU's already, Intel are still some months away from 45nm CPU's) and Intel don't have the advantage of the half nodes. Given the indications so far I'd suspect they would want the cutting edge to ensure the price/power/perf ratios - unlike chipsets, GPU's aren't small, so you want to maximise the wafer usage.

Those theoretical GPU's don't necessarily need to be high-end FWIW. Thus the "old" node would still be good for them I suppose.

 

[nicolasb]

A poster above goes so far as to claim that AnandTech "tested" said hardware, when AT's own comments conclusively debunk that idea completely.

Excuse me?! That was AT's choice of words, not mine. If you think it's an inappropriate choice of words, fine, but criticise AT, don't criticise me for quoting them accurately.

We were determined not to leave the island without running at least one test on Barcelona. We worked long and hard, and we were finally able to spend some time alone with Barcelona in Taiwan. But the story doesn't end there; it's unfortunately not that simple.

The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667; most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=3006

My own opinion is that AMD has a lot more knowledge of its own products than it is being given credit for by some sites

No one is suggesting AMD doesn't know the state of its own products, they're simply suggesting that AMD is lying about the state of its own products. The R600 launch establishes a fairly significant precedent for supposing that what AMD says and what AMD actually knows to be the case are two very different things.

 

[dizietsma]

Whether or not AMD or Anandtech are correct when the chip is finally produced it will need to be at about 2.4GHz or better to compete with Intel Penryn given AMD's own performance claims; so either it will or it won't and AMD know what the bar is.

After the R600 they cannot afford to fall below this bar, not when saying they are going to be X% better all year, so the release speed is fixed, it is only the timescale that is going to be flexible. Lets hope the latest revision of the chip is the one that can get them up to this level, if not then it's a case of hoping for the next one. I'm guessing the overclockers will not be too impressed though.

 

[3dilettante]

I wouldn't worry a whole lot about the tiny pieces of pre-production and engineering-sample hardware that sites like AnandTech have been able to *briefly* touch so far this year. A poster above goes so far as to claim that AnandTech "tested" said hardware, when AT's own comments conclusively debunk that idea completely. "Touched it" is a lot more apt than "tested it," imo. A couple of things to think about:

They were allowed to briefly touch the chip 1 month before the supposed launch date.
It was stated that they were told that the demonstrated chips were the latest stepping.
AMD prevented them from testing it further.

Assuming AMD isn't lying about those chips being the latest stepping, what time frame can we expect for a chip to make through the fab process?
I've heard 12 weeks quoted, but I can't be sure.

That means there better be another core stepping at least halfway through the fab process right now.
That is a lot of pressure on a new stepping.

It is also asking too much to characterize a new tentative release candidate as mass production.

(1) Prior to AMD shipping Opteron (which they did prior to shipping the desktop A64, remember), there was much despairing commentary also written about Opteron, which was based on early pre-production and engineering-sample looks that various people had. They jumped to several erroneous conclusions about Opteron and Opteron's future, as a result. That's pretty much exactly what we're seeing here, yet again, with respect to Barcelona.

I recalled slightly more positive buzz about a month before the supposed launch, or at least I looked forward to it.

(2) Prior to Intel actually shipping Core 2 to review sites and then the public, and lifting the NDA's for Core 2, general commentary about the upcoming Core 2 was that most of these sites doubted it would be giving Opteron/A64 much in the way of competition.

I would say the general commentary was being made by morons, then.
The Pentium M lineage alone should have been enough to know that it would be at or near the IPC of A64 at a bare minimum.
The rumors of significantly improved FP capability would have erased all but AMD's IMC as an advantage, and Intel was guaranteed to have better caches.

The architecture's outline has already been published, and its optimization manual is released.
We know what features Barcelona has. There's no pixie dust here.
We will not expect single-threaded IPC to exceed Core2, nor much that runs well on dual-core or below will run better on Barcelona.

For quad-core, the performance picture will be mixed, though most likely in favor of AMD if the clock speeds are remotely competitive.

It's the higher socket and core count systems AMD is hoping for, which should scale better (assuming clock speeds are competitive).

Scaling alone does not tell anything if there is no baseline, and AMD has been very obvious in avoiding that question.

In this case, too, you've got AnandTech repeating gossip it's heard from gosh knows where about how Barcelona "isn't scaling very well" (I assume they mean in MHz), yet in the same sentence they talk about AMD "being very optimistic" about Barcelona's scaling in MHz--but evidently they decide it's more prudent to listen to the rumors as opposed to the company that will be manufacturing and selling Barcelona.

Let's see what AMD's said (and not said) lately.

Clockspeed is not as important sometimes as ease of use and reliability.
Comparisons with lower-clocked Opteron HEs for the demonstration of drop-in replacement for equivalent performance at a lower power point (no indication of drop-in replacement for higher speed grade 65nm chips or the 90nm top-bin chips and AMD's top bins are still 90nm)

40% lead in selected undisclosed benchmarks, most predominantly scaling rate benchmarks
Single-threaded performance is not material as opposed to the platform (then neglecting to mention its advantages are highlights of projected performance on today's competition, because Intel's going to keep the same product line for over six months).
Barcelona will not make a material financial impact until 2008.

Earlier than that, AMD has made a number of statements about how its efforts are now focusing on platform and areas outside the desktop.

For the server/HPC, if Barcelona were in the 2.4 GHz range, this might be true in a wide variety of workloads, but not all.

At 2.0 or less, Barcelona must find a way to gain an IPC +50% that of already extant 3.0 GHz quad cores. As important as the "native" quad core design is as a feature, it is not enough to erase the overall gap.


For the desktop (not as important for AMD's fortunes, but still):
My interest was Kuma or Agena.
Everyone from Anandtech to Ars to Realworldtech to Xbit to some at AMD has stated that it is unlikely that the 10h line will beat Conroe at single-threaded performance.
That is still dominant for the desktop and likely will be by the time Penryn and Barcelona are replaced.

It's sad, but it is predictable. So, they give the rumors credence while implying with some gusto that AMD hasn't a clue about its own upcoming cpus... I think the record is clear that with Opteron/A64 AMD surely knew best, and so my money's on AMD being right about what they're going to ship this time, too.

AMD's silence and dubious guidance should be reason to wonder.
Did you miss the deluge of horseshit that came out about R600 delay-delay-"family launch"-UVD marketing that even tricked some of AMD's board partners?

Whose word should I trust?

My own opinion is that for some reason these sites are miffed about what they perceive as being left out of the loop when it comes to Barcelona, as, after all, they think, they are "the loop," so to speak... So then they reason that since they are so important in the scheme of things that the fact that they've been left out thus far means that AMD is just trying to hide some things--like Barcelona's scalability in Mhz--ahem, even though what AMD has officially told sites like AT is that AMD isn't worried at all about Barcelona's ability to scale.

I would fire any PR guy who admitted the guys back in corporate are wetting themselves.

And yes, AMD is hiding things.
We may disagree on how much is good and how much is bad, but their silence means they're hiding something.

I'm of the opinion AMD's best interest is not served by hiding good news about the platform it needs to use to survive fiscal year 2008.

Last, I want to comment very briefly on R600: while the pessimists and naysayers are already counting R600 down and out, I believe the fact is that the R600 saga is only just beginning.

Just in time for Nvidia's fall refresh.

My own opinion is that AMD has a lot more knowledge of its own products than it is being given credit for by some sites, and that very possibly we should be paying more attention to AMD here than to sites like AT, which are, after all, pontificating on the future of products they have no input in with respect to either manufacturing or marketing.

Those dumb sites just benchmark them and find the results inconsistent.
Perhaps I should put a list of all the things AMD did and did not say with regards to R600.
AMD, meanwhile, is trying to sell you something.

The rumor was that R600 would fall short, run hot, have buggy drivers, and be delayed.

AMD said things would be fine, then that there would be a family launch, and that the extra time would make their drivers (particularly Vista) top notch.

Nice to see how the naysayers were proven wrong.

 

[nicolasb]

They were allowed to briefly touch the chip 1 month before the supposed launch date.
It was stated that they were told that the demonstrated chips were the latest stepping.
AMD prevented them from testing it further.

Assuming AMD isn't lying about those chips being the latest stepping, what time frame can we expect for a chip to make through the fab process?
I've heard 12 weeks quoted, but I can't be sure.

Of course, if you believe FUDzilla, the chips being shown at Computex were B0 stepping, and we are actually just a few days away from getting the B2 stepping:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1424&Itemid=1

Our sources have confirmed that a few Barcelona chips that the company demonstrated at Computex behind the closed doors are B0 revision.

We also learned that B0 can go faster than 1.6 GHz but that it was a very buggy revision. Now our sources are talking about B2 revision that is suppose to save the day and this one is expected any day now.

 

[3dilettante]

What happened to B1?

 

[LeStoffer]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1424&Itemid=1

How totally clueless their are:

The same people also indicated that Barcelona will either be the best thing that even happened to AMD or it will be doomed to be the next Netburst.

Of course it will be neither.

The best thing that even happened to AMD? Better than K8? Doomed to be the next Netburst? Nothing wrong with the architecture here, just to little and to late. Just totally clueless.

 

[overclocked_enthusiasm]

More ASP declines and straight out of Henri "we don't do soft launches" Richard's mouth to boot.

http://economictimes.indiatimes.com...rop_chip_prices_by_50/articleshow/2117688.cms

"Advanced Micro Chief Marketing Officer Henri Richard said last week that he was ``concerned'' that Intel would cut prices of quad-core processors and drive down overall levels in the industry.

``That's the sign that I am getting right now,'' Richard said in an interview in Taipei on June 6. ``If they are pushing quad-core unnecessarily down because they think they can hurt us by doing that, then I don't think there's an optimistic view on'' average prices. "

 

[Arun]

More ASP declines and straight out of Henri "we don't do soft launches" Richard's mouth to boot.

Heh, for how long have I been saying the Intel Q3 price cuts would hurt Barcelona's pricing power badly unless it was noticeably faster than the Q6600, since that SKU will be repositioned at $266?

Obviously, since all current rumours indicate that at least some Barcelona models will NOT be as fast as the Q6600, they are effectively selling a 280mm2 chip without redundancy for less than $250 in the desktop market, and the price cut percentage of the model comparable to the Q6600 in the server market was most likely similar, I would assume. Compared to CPUs in the 2004-2005 era, the ASPs of the Barcelona Family will, I believe, be quite laughable.

So as it is, there is no reason to believe that Barcelona, Budapest, Agena, Kuma, Rana or Spica will ever achieve higher margins than AMD's current 65nm solutions did in Q1. It shouldn't be very hard to tell what this means, and what investors' reaction to that will be.

AMD has no right to complain either, as what Intel is basically doing is aligning the price of their quad-cores to twice that of their equivalent dual-cores. Given their package-based solution, that is not exactly a loss-making strategy last I heard...

 

[3dilettante]

Obviously, since all current rumours indicate that at least some Barcelona models will NOT be as fast as the Q6600, they are effectively selling a 280mm2 chip without redundancy for less than $250 in the desktop market, and the price cut percentage of the model comparable to the Q6600 in the server market was most likely similar, I would assume. Compared to CPUs in the 2004-2005 era, the ASPs of the Barcelona Family will, I believe, be quite laughable.

Not to pile on, but to elaborate on the redundancy point:

AMD's native quad-core strategy may be about one (and a half, if there were a half for x86) process node too early.
It's not just defect rates that go up with large dies, but apparently process variation has become a worsening problem as well.

If a single core out of 4 in a Barcelona chip can't clock above 2.0 GHz, it doesn't matter if the other 3 can push 3.

It was also pointed out elsewhere that device leakage is even more variable than speed bin.
While this probably won't hurt Barcelona's lower speed grades too much, it may significantly impact top bins if one leaky core pushes a chip over the TDP.

Intel's packaging strategy allows mixing and matching of leaky parts, allowing better bins.
Once chips shrink further to change the balancing point between variation and die size, Intel will put out a native design.

 

[INKster]

Not to pile on, but to elaborate on the redundancy point:

AMD's native quad-core strategy may be about one (and a half, if there were a half for x86) process node too early.
It's not just defect rates that go up with large dies, but apparently process variation has become a worsening problem as well.

If a single core out of 4 in a Barcelona chip can't clock above 2.0 GHz, it doesn't matter if the other 3 can push 3.

It was also pointed out elsewhere that device leakage is even more variable than speed bin.
While this probably won't hurt Barcelona's lower speed grades too much, it may significantly impact top bins if one leaky core pushes a chip over the TDP.

Intel's packaging strategy allows mixing and matching of leaky parts, allowing better bins.
Once chips shrink further to change the balancing point between variation and die size, Intel will put out a native design.

That would be "Bloomfield", more than a year away.
According to the latest leaked specs, it's a 45nm native quad-core design, supports up to 8 threads (HT reborn, perhaps) and it will come with an integrated, triple-channel (!) DDR3 memory controller.

Two of those in a MCM package could pose a huge threat to AMD by then.

 

[nAo]

If a single core out of 4 in a Barcelona chip can't clock above 2.0 GHz, it doesn't matter if the other 3 can push 3.

I hope they can at least sell CPUs with defective cores as 1,2,3 cores CPUs.

 

[3dilettante]

The disclosed products so far don't say much about that.
It's also not a question of one core being defective, it just clocks slower or leaks more. It does mess up binning and margins, however.

I haven't seen much about tri-core SKUs, and it's also a question on whether a tri-core at a higher clock can be priced above a quad-core with a lower clock.

A quad core could possibly be have fuses blown to a dual core. The question is what that means for AMD, which already has separate dual core and quad core chips. It seems reasonable there could be a way to downgrade a chip, though I haven't heard it talked about.

 

[ZioniX]

According to JPR, Intel claimed a 92.0% share during Q1, up from 88.9% in Q4 2006, up from 89.1% in Q3 and up from 86.7% in Q2. AMD in turn, dropped from 13.3% to 8.0% in the same time frame and is almost back to a market share level that was recorded by JPR for Q3 2005 (6.6%).

“We’d expected AMD’s share to moderate or level off by the time Intel improved its dual-socket Xeon platform in mid ’06, but we hadn’t anticipated the decline we’ve seen,” said JPR analyst Alex Herrera. “The extent of Intel’s rebound will put that much more pressure on AMD to deliver quad-core Barcelona soon - and with better performance than Xeon.”

AMD loses market share in x86 workstation segment faster than expected

 

[3dilettante]

AMD loses market share in x86 workstation segment faster than expected

Sounds like the dearth of competitive chips and weak execution is leading new sales to avoid AMD and some of those who had AMD gear to give up waiting.

This calls for desperate measures. Obviously, many people are not assuaged by the 2.0 GHz chips shown recently.
AMD should take it to the next level and demo even buggier chips at 1 GHz to restore its credibility.

 

[overclocked_enthusiasm]

That server market share was from Q1 and has already been known for sometime. Henri Richard implied at the Q1 CC that AMD hadn't lost as much share as some of the analysts were asking about. In hindsight it was another very disingenuous response and another attempt to sugarcoat a turd. Here are some of my favorite exchanges with Henri:

http://seekingalpha.com/article/32901

David Wong - AG Edwards

Right. Could you give us any quantification of sequential microprocessor units growth or declines in any of the segments?

Henri P. Richard

We typically do not give you those indications.

Robert J. Rivet

The only comment I would make to give you the best color would be ASPs were the small piece of the equation in the sequential movement quarter to quarter. It was mostly units. ASP were a piece of the equation but the driver was the unit drop quarter on quarter. That was in all segments, whether it is desktop, mobile, server.

Henri P. Richard

I would like to add one point. We need to wait a couple of weeks until all the analysts are publishing the sales out figure for the quarter, because we know that customers really dispose of inventory that they had built in the fourth quarter, and so you might be actually surprised by our sales out market share numbers versus what you are looking at today, which is sales in.

Krishna Shankar - JMP Securities

As you look at the quarter, did you see the pricing pressure growth in the consumer and the corporate market? Can you give us some sense for the success you had in the corporate notebook and desktop market, now that you have API under your fold?

Henri P. Richard

Sure. So the first part of your question, which is there was no real significant difference in terms of the overall pricing environment, but of course we have different participation. I would say that in the consumer market, there is a broad level of competition in both mobile and desktop across all geographies, where today in the enterprise market the bulk is in the server segment.

Now that leads to the second part of your question, which is what is going on with our commercial clients. We are, as planned, continuing to increase quarter after quarter the number of platforms offered by our key OEMs in increasing number of channels and increasing number of geographies and sales are growing quarter after quarter.

Furthermore, there is another indicator which is a leading indicator of success, which is the latter part of your question which is since now we have ATI and therefore are able to offer a more complete platform, is that giving us some traction? As I indicated earlier in the call, I am very pleased with the level of design-in wins in the first quarter, even though I am not pleased with the level of sales in the quarter.

Michael McConnell - Pacific Crest

Looking at the graphics business, any way we could get an apples-to-apples comparison between Q4 and Q1? I know there was a limited amount of revenue in Q4, but is there any way we could get an idea of what the growth was quarter to quarter there, full quarter to full quarter?

Henri P. Richard

I will give you -- basically the revenue was slightly down. Units was up and ASP was slightly down. Frankly, it is logical since we are waiting for the launch of our brand new R600 line of products, which will enable us to restore the price positioning in the marketplace.

Michael McConnell - Pacific Crest

Okay, and just on that note -- last question and I will go away -- the R600 stack coming out in late May, is that enough time for the window before back to school, or should we be thinking about this more for the holiday season?

Henri P. Richard

Well, you know, when you think about back to school, a large majority of those machines use integrated graphics, so there is very little impact of discrete graphics. As you know, the launch of the R600 is essentially a discrete graphics launch. Those cores will find their way into integrated chipsets later in the cycle.

I feel that when you have the right technology, particularly in the enthusiast space, it gets gobbled up pretty quickly and those people do not actually march their purchases based on cycles.

Jim Covello - Goldman Sachs

Thanks so much. A quick question -- the guidance for Q2, does that assume share gain, share loss, or no change in the microprocessor segment?

Henri P. Richard

From which quarter, from Q1 or from a year ago?

Jim Covello - Goldman Sachs

No, I’m sorry, the sequential guidance of flat to up slightly.

Henri P. Richard

By definition, given the guidance of our competition, it means share gain.

Jim Covello - Goldman Sachs

Okay, I wasn’t sure what the mix was relative to some of the other segments as opposed to pure microprocessor. So you think you will gain share in the second quarter?

Henri P. Richard

Yes.

Jim Covello - Goldman Sachs

What gives you confidence in that after the last couple of quarters?

Henri P. Richard

Well, again, if you look at the third and the fourth quarter, despite some of our challenges, we were gaining share. As Hector pointed out, we had a very difficult quarter. But as I also told you, our design-in pipeline has never been stronger. What is selling in the market is one thing. How many systems are your OEM proposing, how many countries are they offering your products in, how many customers are having access to your programs -- all of these things are not changing and we are seeing, particularly in the mobile space, we had a 30% year-on-year growth when the competition actually went down.

We are continuing to see good momentum on design win and that is why we are expecting that we could have a decent second quarter.

Gurinder Kalra - Bear Stearns

Just a question on your R600 launch. Can you provide us more details in terms of timing, whether it will be a hard launch or a soft launch, how much product do you expect to have and in what categories? Would there be a performance product or a performance mainstream product?

Henri P. Richard

We are going to announce those products in the second part of May. The location has been picked. It is a series of 10 products covering the entire stack with DX10 capabilities, leading edge performance and more importantly, a solution that works with [WECO] qualified drivers.

Gurinder Kalra - Bear Stearns

Would you expect to have substantive quantities of product available at the launch?

Henri P. Richard

Absolutely. We do not do soft launches.

 

[Geeforcer]

Let me ask this hypothetical question: If AMD/ATI deal has not gone through, what do you think stock price for each company would be right now? This is probably the end of the stretch when we can make such a guess, since pretty soon we will be seeing jointly developed products hit the market and separating former ATI from AMD side of business will become impossible. On a semi-related note, Intel has hit 52-week high today and Nvidia has been creeping back toward theirs as well.