Core2 Beginning to Crumble

The Inquirer gives us an interesting article that discredits that the upcoming products are as good as claimed. If they were that good, and easy to manufacture, Intel would be allowing reviewers to freely play with them, but it is happening the opposite.

Intel gathered reviewers, gave them Conroes and Benchmark Software to do the reviews which Intel expects to be published by June 4th. Also, the reviewers were promised to be sent units "at a later date", but haven't received them.

Intel is a marketing-governed company. If Intel had truly good products, in mass production as would be expected for something that will be launched in less than two months, it would leave the reviewers to play with those supposedly excellent products any way they wanted, to create a wave of word-of-mouth enthusiasm.

Nevertheless, the Conroe sightings (and Woodcrests, Meroms too for that matter) have been suspiciously scarse.

Let us review Intel's attitude towards Core2 processors:

• Many months ahead of launching, Intel disclosed that they have managed to develop killer new products,
• That those products were coming imminently to the market
• And Intel didn't care about totally eroding the credibility of all Pentium 4s embarrassing them by the comparison to the next products.

Thus, they followed an attitude of "tabula rasa" and "all or nothing" with Core2. How are we going to believe that they want to be secretive now? We must conclude that Intel is having serious problems with the new processors.

We already have pointed out predictions:

• That they are very complex, thus it is entirely reasonable to expect all sorts of development delays, production delays, and overall low quality like in Blue Screens of Death.
• Being so big (larger than the 90nm they come to replace), they must be very hard to manufacture

Naturally, the recent developments are in agreement with these predictions. That's why I will go beyond to speculate about exactly what's the matter:

I don't think the most important problem to be yields (the statistical distribution of operative processors versus clock frequency), because although they are gigantic in area, since Intel managed to ramp Yonah (Core Duo) quickly enough, it demonstrated it has enough control of the 65nm process, therefore Core2 would be at most harder to manufacture, but definitively mass produceable.

Then it should be processor malfunction. I think that Intel has not been able to iron out all the bugs in Core2. I guess that the frequent operational situations are well covered, but not the rare cases. To avoid Blue Screens of Death (BSODs), they must make sure that the software will not trip the bugs, and the only way to accomplish that in tests and benchmarks is to make sure they have been patched by Intel. It is becoming more probable almost by the day that Core2 will require Microsoft and other Software vendors to come up with patches to "support" Core2... Don't expect anybody to say that the Core2 need the patches because they are buggy, but expect any Marketing bullshit such as "since Core2 are so advanced they require updates to the Operating Systems and Applications to make use of all of its features".

In Bob Colwell's Stanford lecture, he complained that in the very old Pentium Pro the problem of race conditions already showed up with all of its nastyness. That is, with the re-ordering in Pentium Pro, sometimes the processor followed a re-ordering strategy that far from incrementing the efficiency made the processor to stall. I have worked as Software Architect in low level of abstraction software, therefore I have firsthand knowledge about how difficult the race conditions are to predict, detect, and correct. Being Core2 so much more aggressive in Reordering and other optimization strategies, it is truly a certainty that they are susceptible to lots of bugs and race conditions, that must be one among the many problems Intel is desperately trying to solve right now.

The insistence of management about this unrealistic schedule may be an indication of Intel's internal disconnection; although these design and development problems are crucial, they are also not at all easy to understand for those who don't have a solid engineering background. The Marketroids at Intel that compose top management must be just demanding from engineering something as impossible as demanding from nine women to deliver a baby in one month; the reality of the bees nest they got into will explode in their faces very soon.

Imagine then some problems in manufacturing (yields of less than 40% operating processors and most of those can't run at the promised frequencies) with a buggy operation, and all you can do to sustain the illusion of superiority is to cherry pick wafers, processors, tests, benchmarks, reviewers and do special manufacturing runs for demos.

Isn't it in Sun Tzu's "The Art Of War" that if you are weak you must pretend strength, and if you are strong you can pretend weakness?

If you care to read more about this subject, in "Steam..." I said, doing an analogy between Steam Engines/Diesel-Electric and Intel/AMD processors:

The design ideas and techniques of Steam Engines were exhausted, played out, but Diesel's had the whole world ahead.

Intel's new lineup of products are steam engines. There aren't any new ideas to further the basic Pentium Pro architecture; I, for one, am skeptical from the professional point of view (summary here), about some of the design compromises Intel chose; and in particular, there are some assertions that my considerable understanding of this subject are not capable of explaining, such as the claims of higher power efficiency, or how is Intel going to successfully deal with the problem of trading off combinatorial complexities for marginal benefits so pervasive in all engineering disciplines; especially with Intel's track record.