Leaked Toliman Die Image

Lately I have been receiving leaks from inside AMD and Intel from anonymous readers. Some of them I cannot reveal since it could put the person in a lot of trouble due to the sensitivity of the subject matter and their obvious involvement. But some do advise to reveal the secret. Like for instance this leaked image of the soon to be released Tri-Core K10 (code named Toliman). Since this leak was passed on to me by GutterRat from someone else, I cannot guarantee the authenticity of this image.

Reading Graphs 101

It should be obvious by now that a lot of folks in the semiconductor industry visit this blog and any wild speculative comments about chip production never gets a free pass. I get quite a few myself but I am always grateful for the learning experience. But for some people, it's like banging your head against the wall.

Take for instance making a claim about things beyond ones expertise. There's a big difference between making claims and stating an assumption. The 1st mistake is making claims about a company's future process. Nobody on this planet except an Intel or AMD approved spokesperson will disclose or make a statement about future yields. You get fired for breaking that rule and if someone does, normally you won't get proof anyway. So if you hear an outsider making claims about future yields, save the grain of salt and dismiss the claim outright. The 2nd mistake is presenting it as facts and using it for one's argument. The 3rd mistake is to keep on insisting even when provided with counter evidence.

Take for example when Scientia said:
"Intel's yields, in contrast, on its brand new 45nm process will take a couple of quarters to reach maturity... Intel will improve its 45nm process and this should pay off by Q2. The process will be mature when Nehalem launches in Q4 08. ".

InTheKnow countered with this evidence (from Mike Bohr, Intel senior fellow). But if you thought the discussion should have ended there, you're wrong.

Scientia's comments about this graph:
"No. You are reading the graph wrong. What it actually shows is that 90nm had worse initial defect density than 130nm but about the same improvement rate. The chart further shows that there was no improvement in initial defect density with 65nm but the rate of improvement got worse. It further shows that 45nm is close to 65nm and worse than 90nm. Again, this matches with what I said."

I'm not sure if Scientia is deliberately misreading this graph but this is the worst graph interpretation I've come across. The way I see it, and if I were to stick to the main point, 45nm matures before 2008 as each node reaches maturity quicker. This is right smack into the 45nm early ramp.

Here's InTheKnow's more detailed and accurate interpretation.
1) Intel required ~24 months to reach the same level on 130 nm that they eventually reached on 90 nm.
2) Intel reached the flat portion of the plot in ~22 months at 90 nm. This despite several flat spots on the graph that showed significant yield hurdles had been encountered. I would expect this since the 90nm transition also overlapped with the 12 inch transition somewhat.
3) On 65 nm Intel matched 90nm yields in ~19 months. Yields continued to improve from beyond the 90 nm levels.
4) 90nm launched around the end of December '04. Intel had reached the flat part of the graph ~2 months prior to this.
5) 65nm launched in Jan '06. Intel matched the 90nm yield levels ~3-4 months prior to this.
6) 45nm is now at about the 18 month point on the plot. If they are matching 65nm yields then they should be very close to the 90nm yield level now. The launch is believed to to be 2 months away and they should be well into the mature portion of the yield graph by then.

Intel Ticks On November 12

It defines Intel's cadence of process and design execution, but for AMD, "Tick - Tock" is beginning to sound like the countdown to its own dissolution. Every time Intel makes an announcement or a demo much like this week's IDF, AMD by comparison just looks more and more pitiful. It can't be said that AMD didn't bring this upon themselves. For the life of me I cannot comprehend why would AMD think that announcing plans to ship 3-core defects can be considered as stealing thunder while Intel talks about 4, 8 and 16 core CPUs.

Of course, the misery doesn't end there, while still trying to recover from its abysmal product launch, AMD is left with no choice but to look at their fluke 2.5Ghz Barcelona beaten to death by Harpertown, Intel’s soon to be launched, 45nm Xeons. The big difference is that the new Xeons with the 1600MHz bus has an official launch date and will be shipping in volume real soon.

The benchmarks confirms Intel's claim to floating point superiority, further increasing the performance gap with AMD. The only thing worse than stumbling is doing so while the competition executes efficiently.

From the TechReport:
Those new Opterons will certainly have their hands full with Intel's 45nm Xeons, though. The Xeon E5472 extends Intel's performance lead over the fastest quad-core Opteron we've seen yet, the 2.5GHz model 2360 SE. ...it looks like Intel will continue to lead in the server and workstation markets. The same may be true in other markets served by these same basic CPU designs, but only time will tell for sure.

Raising the Dead

There is a rumour floating around that AMD is planning to ship triple core K10’s. Quality and Reliability engineers at AMD are scratching their heads while Intel’s Yield engineers are probably rolling on the floor laughing at this obvious desperate attempt to salvage revenue from traditionally defective products. If indeed true, I can make a guess that this is a marketing idea because no sane product engineer would want to be associated with such travesty.

Some may already be calling this a clever move since according to some, a gap exist in the market between dual and quad core CPUs. And on top of that, Intel’s MCM approach prevents it from creating their own backward stepping 3-core CPU. Let me just say that is a silly assumption. First of all, what’s stopping Intel (besides good yields and common sense) from disabling one core from the 2nd chip to create a triple-core? Secondly, while there may be a gap between the numbers of cores, you won’t find any gap in pricing. There is no niche segment to take advantage.

There is a known relationship between yield and reliability which is why most semiconductor companies do not ship products from a bad production lot even if some dies do past the test. To illustrate the real concern, medical (i,e., heart pacers) and automotive companies are very strict when it comes to yield and how maverick lots are handled. While microprocessors built for PCs and Servers do not require similar tight controls, consumer level quality requirements aren't that far off. What is alarming is AMD's willingness to sacrifice quality over revenue.

When a CPU is designed to run at 2.6GHz and the output maxes out only at 2.0GHz, there is clearly something wrong with the process and the product's physical make up. The farther the products are from the design target the greater the uncertainty is with the reliability. Salvaging the CPU by disabling one of the cores does not remove the risk that the product would perform normally over time or is free from time dependent defects. No matter how anyone would want to spin this, AMD's compromise with quality says a lot about how poor their yields are and how desperate they have become.

Update: 18th Sept 2007
Intel's official response to AMD's tri-core:
Asked in a press conference following his Intel Developer Forum keynote how Intel might respond to AMD's recent announcement of a planned three-core processor for early 2008, Otellini offered a brief but savage response: "We see a distinct advantage in having all the cores on our die work."

AMD’s Orwellian Product Launch

The controlling of the media, the use of strange words to confuse people, the manipulation of results to cast themselves in a better light and the propaganda about the cause they are fighting and how it is supposed to be for the good of all. No, I’m not talking about an authoritarian regime. I’m talking about a public corporation that is in a terrible crisis and that trying to manipulate reality seems to be its last resort. AMD’s spin and propaganda machine is running at a level that even a fully secured dictator would learn a thing or two. But journalists are taking note and are starting speak out. Well, at least those that were brave enough not to sign a Non-disclosure Agreement (NDA) in the first place.

The latest story, one which is related to AMD's NDA scandal comes from the INQ as the chip company once again tries to force journalist to sign away their integrity. Mike has a point in asking the real purpose of the NDA when in fact no secret is ever disclosed. The terms of the NDA including the one the Thai journalist “walked out” on seem to suggest that the underlying purpose is really to sanitise and control the stories. AMD’s desperate need for such tactics only magnifies the obvious fact that AMD has plenty to hide.

…Hiding not so pleasant details like Barcelona's performance and how it was indeed a paper launch with delayed and very limited samples sent out for benchmarking. The latest charge from AMD’s arch-nemesis George Ou, is the omission of Intel’s best scores while never disclosing the reason why the scores were lower (auto-parallelization=off). To be lying and cheating while comparing to the slower of Intel's offering is just reaching deeper levels of sleaze for AMD.

Time Of Death: 0:00 am September 10 2007

AMD’s new micro-architecture was officially launched today and at no surprise to anyone who contributes to this blog, Barcelona is dead on arrival. Doubters like ourselves were told to wait for this day. Sept 10, 2007 was supposed to be the day AMD takes back the performance crown to the rejoicing of its followers. But yet again, AMD disappoints. The promise of a 40% performance lead at the beginning of the year was cunningly and gradually reduced to a trivial performance-per-watt advantage. An advantage that current K8’s probably already had but nonetheless insignificant to prevent AMD’s market share erosion.

To AMD's credit, the improved IPC that Barcelona brings with it is quite noticeable. Whereas in the past the K8’s were severely lagging both in clock speed and in IPC. But with Barcelona the gap narrows down to raw clock speed. Like we mentioned several months ago, GHz is once again King and AMD’s clock frequency problems aren’t quite easy to overcome. Anyone speculating that AMD can ramp its 65nm SOI to 3GHz by Q1’08 is only preparing for another public humiliation. For a number of reasons; Barcelona’s already high thermal dissipation at 2GHz, its poor 65nm SOI process and the very large die, all create an insurmountable barrier that AMD can never overcome. You should notice how Barcelona is debuting with a very low clock frequency with an already high 2.6Ghz original power draw.

From the TechReport:

Note: Sample benchmark from TechReport and may not be representative of the entire comparison.

“Nonetheless, AMD now faces some harsh realities. For one, it is not going to capture the overall performance lead from Intel soon, not even in "Q4," which is when higher-clocked parts like the Opteron 2360 SE are expected to arrive… On top of that, Intel is preparing new 45nm "Harpertown" Xeons for launch some time soon, complete with a 6MB L2 cache, 1.6GHz front-side bus, clock speeds over 3GHz, and expected improvements in per-clock performance and power efficiency. These new Xeons could make life difficult for Barcelona… this CPU architecture may not translate well to the desktop, where it has to compete with a Core 2 processor freed from the power and memory latency penalties of FB-DIMMs…”

From Anandtech:
“When you are looking for the highest performance however, Intel has still a solid advantage with it's 3 GHz Xeon x5365”

Note: Sample benchmark from Anandtech and may not be representative of the entire comparison.

Anyone thinking this is a good start for K10 is fooling themselves. This is a terrible start for any new product, never mind one that is desperately needed to be better than the competition. When Core2 came out and trounced everything AMD had, even Hector Ruiz was forced to admit that a new generation processor is always expected to "leap-frog" the competition. Clearly things are different this time around. Barcelona's problems and delays when combined with Core2's massive performance improvements over previous generations created a gap too wide for AMD close. For the first time in several processor iterations between Intel and AMD, the latter's new processor falls short. While AMD will be stuck with an already beaten K10 design for the next 3-4 years, in a few months Intel will move to 45nm that will allow it to clock even higher, reduce the cost even further while drawing less power. Expect the gap on any meaningful metric to continue to increase as AMD's K10 is simply not good enough.

The Leaked Barcelona Numbers

Leaked Barcelona launch slides have reached George Ou. Obviously the numbers doesn't look very good for AMD. But if you've been in this blog before, then you shouldn't be surprised.

Instead, I am quite intrigued by the nature of the leak. It looks to me that if Barcelona's numbers are weak, AMD has much to gain by leaking the numbers and slowly deflating everyone expectations before the launch. These sort of tactics never really amount to anything in the long run as we've already seen how Core2 was benchmarked to death when it was launched. Thanks to the internet and the growing numbers of benchmarking sites, AMD's PR Department, specifically the Damage Control Team will be very busy for the next few years.

Barcelona's Entry Level Dilemma

Even George Ou agrees with the dangers of launching a next generation product that isn't quite ready and worst, in small quantities.

From ZDNET:

Barcelona at 2 GHz will only threaten high-margin AMD CPUs but it will not threaten high-margin Intel CPUs because the clock speed isn’t high enough yet. When a $250 Barcelona DP quad-core kills a $750 Opteron DP dual-core processor on performance, who will buy that $750 part? When a $700 Barcelona MP quad-core kills a $1600 Opteron MP dual-core processor, who will buy the $1600 part? Barcelona will effectively erase all of the high-margin CPUs in their line-up next week when it launches.

Like I mentioned in my previous blog, AMD's decision to go ahead with the launch renders its entire Opteron line obsolete. AMD even make it worse for themselves by making Barcelona socket compatible therefore giving its customers the ability to quickly switch to the the new platform. This is very risky from a business point of view. The much needed revenue that the original Opteron line provides AMD can never be replaced by the new but lower volume Barcelona as it slowly ramps. It would be interesting to watch AMD's server inventory levels in the next few quarters.

Value Gaming - game.amd.com

AMD recently launched a new gaming website. This is an excellent idea from a company that can deliver a complete gaming platform. There is just one tiny problem, it's 2 years too late. Where was this wise guy who came up with the gaming website idea two years ago when AthlonFX was king? The gaming crowd has moved on and AMD is just embarrassing itself by pretending to offer anything attractive which is at the moment quite impossible with a $300 quad core from Intel.

This 'gaming' site gives the same feel as the 'break free' website. It feels rushed, half baked and hollow. Browsing through only gave me a feeling of sadness for the smaller company that once had the right product for this very demanding and unfaithful market. The target audience for gaming sites are smart and knowledgeable. They read reviews, benchmarks and are constantly aware of what is the best and the latest. In a very obvious way this website hides information a gamer needs to create the best system. Understandably the site only offers AMD and ATI products. The problem is every gamer already knows they're not the best nor is it the cheapest. Which really leads to the question whether this was a useful exercise for AMD.

All in all, this is yet another poor marketing attempt with an advertising gimmick that is off timing and completely useless for its indented audience. If you think you have a different opinion then ask yourself the question, do you find the need to return to AMD's gaming website? I didn't think so.

AMD and NIVIO

AMD recently announced a partnership with NIVIO along with an offer of an undisclosed amount of investment. NIVIO is a web based virtual desktop service allowing individual users the right to own virtual PC's for a monthly fee. As for AMD's motives for the investment, Giuseppe Amato, AMD's European technical sales and marketing director, said that this was part of the company's 50/15 initiative (50% population with Internet access by 2015). That is of course if you think Pinocchio's step father isn't lying.

If NIVIO becomes successful we're not talking about third world revenues. From a business point of view, AMD knows it has much to gain as this company can become a driver for demand. The potential to use thin clients to access a more powerful virtual PC is the concept behind NIVIO. AMD's Geode has the ability to expand in this space as the basic CPU for web access. Opteron can power both web servers at the front-end and the farm of virtual machines at the back-end. If this company is dead on about the trend in virtual computing, the decline in desktop PC revenue will be compensated by the pickup in server CPUs and AMD wants to be there.

But it is all a big IF whether the business does pick up. Currently there are quite a few "remote access" PCs being offered as a free service for basic features. This is for people with their own PCs at home which they can access remotely from anywhere. This somehow encroaches into NIVIO's remote PC concept. But, the significant part of the NIVIO's service is providing its users their very own "virtual PC". For a fee of $12.99, you can get your own virtual PC which you can turn on and access anywhere in the world if you have Internet access. The PC is a virtual machine in a shared resource environment.

Again the price is $12.99/mo for a virtual PC. Unfortunately, here's a real, non-virtual Dell PC selling for $11.00 a month. I'm not quite sure how well AMD thought this investment through because it only gets worst. The $11/mo PC from Dell is Sempron based. Talk about shooting yourself in the foot while stabbing your customers.