Intel?s announcement last week that the P4 4 GHz would be delayed into 2005 may have been no surprise to those of us who follow the industry, but it can?t have improved the mood in Santa Clara.  It?s the latest in a recent set of hiccups and roadmap adjustments for Intel, whose had to recall early boards based on Grantsdale because of a problem with ICH6, delay next-generation Dothan chipsets, and now has been forced to scale back its P4 ramp.  With the Prescott 3.6 GHz CPU still in very tight supply, we?ve seen a fresh wave of doomsday predictions that we thought we?d address.  As tempting as it is to compare Intel?s recent trouble with the train wreck of 2000, there are major differences between the two. 

• Platform Issues Incomparable:  Its true, Intel?s recall of Grantsdale and the just-publicized Lindenhurst problems are black spots on the company?s record, but the i820 disaster was far, far worse.  Intel?s Grantsdale recall happened immediately after the chipset launch with few, if any, systems actually shipped to market.  The MTh-equipped i820, in contrast, shipped for months before Intel discovered (admitted) the problem.  Even the costs are vastly different?39 million (estimated) for the Grantsdale recall, over 250 million for MTH-i820.  Intel seems to have learned its lesson; a small chipset recall at the front-end may earn you a black eye, a recall after six months tends to give you a bad name.  One of them is much harder to shake than the other. 
  
• Prescott Isn?t at the Limits of its Process:  This may seem like an odd one to bring up since I?ve been vocal about the lack of high-end Prescott?s in the market place, but there?s a method to my madness.  The last time Intel had major availability problems was with the P3 1 GHz, but there?s a difference between the two.

The P3 1 GHz on .l8 micron was at the absolute limits of its design, and nothing anyone did could change that.  Not only was Intel unable to provide a 1.13 GHz chip, but even websites using high-end water cooling weren?t able to push a 1 GHz sample past 1050 stably.  Athlon?s from the same time period and at the same speed were pushing up to 1200 MHz. 

Prescott?s problem is different.  Unlike the CuMine P3, we?ve seen substantial evidence that Prescott can break the 5 GHz barrier today, if strong enough cooling solutions are used.  This implies that Intel?s problem is more an issue of keeping Prescott cool within their own desired thermal envelope using an OEM cooler.  A re-designed or heavier cooler is probably an option, but such coolers tend to be louder and inherently more expensive?and hence, not attractive. 

• Clockspeeds Less Important:  Not only are clockspeeds less important in today?s market than they were four years ago, but Intel has plenty of marketing fodder to keep advertisers happy.  Back in 2000 the Rambus fiasco(s) had killed any chance of using it as a successful advertising campaign, which left the entire program dependent on clockspeeds?which Intel couldn?t provide.  Today there?s a wealth of material to promote, from DDR2 and PCI-Express to new wireless capabilities and high-definition audio.  Intel can, in fact, whip out an entire suite of marketing material while barely mentioning Prescott?s clock. 

Continuation, Conclusion.

• Market Expectations are Different, AMD Not Pushing:  The ?AMD Race to a GHz" Classic run from August 1999 to January 2000 has become the ?Industry Plodding Onward of Indefinite Length.?  The truth is, neither AMD nor Intel is currently racing the other, nor does the market expect them to do so.  Considered strictly on the basis of clock, AMD?s 2.183 GHz Thoroughbred (Athlon 2600+) released almost exactly two years ago was only 10% slower than today?s current FX-53.  Granted, overall performance has jumped tremendously since then, but the clockspeed wars of yesteryear are over.

If you don?t believe the AMD example (the AthlonXP not being built for scaling, and all that), just look at Intel.  Right after AMD rolled the 2600+, Intel rolled the 2.8 GHz P4.  Two years later we?re at 3.6 GHz, which means that Intel?s clock has risen 28.5% in twenty-four months.  Using the 3.4 GHz as a comparator (since you can hardly buy a 3.6), we?re up only 21.4%. To put that in perspective, from August 2001 to August 2002, Intel raised the clock on the P4 no less than 40%.  That?s 3.3% a month, vs. the current 1.19%.

Comparatively, the August ? September period of 1999 ? 2000 was a different world.  Then we saw clock speeds jump no less than 100% in five months, or roughly 20% per month.  Just prior to Intel?s release of the P4 the Athlon was sitting at 1.2 GHz, up from 500 MHz just 15 months earlier, for a gain of 16% per month.  At the same time, the P3 was still sitting (and barely shipping) at a GHz.  Intel, to say the least, looked bad.

Today we?ve shifted marketing away from clockspeed, the industry is no longer quite so heady with the exuberance of the late 90s, and perhaps crucially, AMD isn?t pushing the way they once did.  All the roadmaps we?ve seen, in fact, imply that AMD plans to keep a rather leisurely pace when it comes to introducing new clockspeeds.

In Q3 we should see 2.6 GHz processors but nothing faster (despite the appearance of .09 micron) until Q1 of ?05, when AMD will roll 2.8 GHz.  In Q2 of ?05 AMD finally reaches 3 GHz with both the FX and Athlon 64 series.  By then dual cores should be on the horizon.  The point is, there?s no sign that Sunnyvale will use its own .09 micron push to put more pressure on Intel, and perhaps for good reason?AMD?s own state of financial recovery is still quite fragile; the company?s long-term financial commitments to Fab 36 are high, and Intel?s deep pockets could allow it to make very short-term unprofitable moves to stuff AMD?s attempts at market gain.

• Prescott Low and Midrange Yields are Excellent:  If there?s one drum Intel has pounded consistently for the past six months, its how excellent Prescott yields are.  Their last investor conference call listed this as a primary reason for the higher build-up of inventory, rather than channel weakness, and Intel executives defended it firmly in the face of skeptical questioning from members of the press.

Given the wave of crackdowns and standard-enforcing currently hitting Wall Street, it?s highly unlikely that Intel would dare state such a thing if it wasn?t true.  Saying that .09 micron is performing ?as expected? is one thing, but Santa Clara went out of its way to repeatedly emphasize high Prescott yields.  This seems, therefore, to be a problem only for the highest-grade Prescott?s; possibly 3.4 and 3.6 GHz (again, we noted that while 3.2 Prescott?s were immediately available at launch, 3.4?s took nearly six months to enter widespread retail availability.)

By comparison, price guides from the 2000 time period indicate that while Intel had announced the P3 1 GHz, processors at that speed were all-but impossible to find; price data and overclocking tests indicate that Intel?s yields were dropping off sharply after 733 MHz.  In May of 2000, a P3 933 was $778, an 866 was $572, an 800 was $377, and the 750 was $254.  Comparatively, AMD?s prices were as follows:  The Athlon 750 went for $179, the 800 was $231, the 850 at $313, and the 900 and 950 chips were $428 and $584, respectively. 

A simple price / availability check will tell us how bad the situation is.  At NewEgg, P4 3.2?s are shipping for around $275 (both Northwood and Prescott flavors).  The P4 3.4E / 3.4C are both around $420.  No 3.6?s are available.  If we treat the 3.4 as the highest available speed, we see price / performance drop sharply at the second-highest speed grade, then delta normally.  Intel?s mid-to-low range Prescott yields, therefore, most likely ARE excellent, and since these are the chips the company sells the most of, this is where the primary concern would lie.   Intel?s problem in 2000 wasn?t just an inability to ship the P3 1 GHz, but difficulty shipping the entire Pentium 3 line much above 733 MHz.  If Santa Clara were having trouble shipping 2.8 GHz Prescott?s we?d have a very different scenario on our hands, but there are numerous indicators this is not the case.

Northwood and Dothan Available:  In 2000, Intel had the Pentium 3 for a high-end product and the Celeron for a low-end product.  That was it.  When the P3?s yields went sour, there was no secondary to take up the slack.  In contrast, Intel?s decision to release Prescott 3.4 while pushing Northwood 3.4 into the channel instead was inspired.  Consumers who didn?t know the difference bought Prescott?s (after awhile) from companies like Dell and HP, while channel buyers who were more likely to know wanted Northwood anyway.  In the short run this was a double-edged sword that blunted Prescott uptake, but over the longer-run it was a very smart move.  Meanwhile Dothan is already known as the chip that?ll power the next-generation dual-core Pentium?s in one flavor or another.  Simultaneously being able to depend on the last generation for support while focusing attention on the next-generation for excitement has worked out tremendously in Intel?s favor even if it has turned Prescott into (even more) of a lame duck in the process. 

Conclusion: 

Is Intel having problems?  Yes, though their problems with .09 micron seem to be industry-wide.  It?s no accident, in my opinion, that AMD?s .09 micron ramp has been rather lengthy; it seems obvious that Sunnyvale has taken the time to make sure their ducks were in a row rather than attempting to push the conversion too fast, too soon.  Even removing the .09 micron transition, however, there?ve obviously been some glitches and snags in the chipset department. 

Glitches and snags, however, do not create the perfect storm of bad press, engineering problems, and suddenly-resurgent competitors that engulfed Intel in the 1999 ? 2000 period.  They do not constitute the same window of opportunity for AMD, and AMD is not in the same position it was to take advantage of them.  Even if Athlon 64?s .09 micron process was as finely-tuned and ready to roar as possible AMD might still be well-served by holding back simply so as to not overshoot what this considerably-more-fragile market is willing to absorb.  Rather than engaging Intel in the short-run, AMD seems to be setting itself up for a longer race. 

If Dr. Craig Barrett?s recently publicized rant on the topic seems a bit overkill its probably because he?s aware that while Intel?s situation may not be all that serious now, it can?t afford to turn that way.  Should Intel continue to suffer manufacturing delays and technical difficulties, it could stunt / delay the entire tech recovery.  Given that Intel?s business depends on convincing customers to continually upgrade hardware, the ripple-effect from such a situation could be severe. 

In the end, the delays and push-backs are probably nothing more than minor bumps we?ll hardly remember in a year.  On the other hand, if Christmas rolls around and Intel still can?t ship 3.6 GHz Prescott?s in volume, well, that?s a different story?but until that happens we wouldn?t be counting on it.  New technology and new manufacturing techniques take time to master, and we?re in the midst of the largest simultaneous technology transition in the past decade.  Bugs?and bumps?were inevitable.

 

 

(1 of 1)

---

• Continuation, Conclusion. (1 of 1)