Stardate 20010510.1450 (Crew, this is the Captain): The sales company reports that my server will be shipped on the 14th and I'll get it on the 16th. Bummer.

Stardate 20010510.1405 (On Screen): This places Apple into a position of "exquisite agony". If indeed IBM begins to make its PPCs using better technology and design than Motorola (and I think this likely) then Apple has to deal with this:

First, it's behind in the "Megahertz war" and it's going to stay there. There's no possible way that even IBM is going to catch up with either Intel or AMD for the forseeable future, since both of them will have 2 GHz processors by the end of calendar 2001. (There's every reason to believe that for the next 12 months at least, AMD will have the most powerful processors in this market irrespective of clock speed. IBM has an astoundingly fast quad-PPC available but it's far too expensive for use in desktop computers.) And no amount of muttering "Megahertz myth" is going to alter the perception in the public's mind that Apple computers are more expensive and slower than PCs, especially as AMD's reputation continues to rise.

But worse is this: Apple will now have a choice of using Motorola's slower PPCs, or switching to IBM -- and discarding Altivec. Altivec is the name of a set of extra instructions implemented by Motorola but not IBM which are more generically known as "Single instruction, multiple data" or SIMD. This concept goes back a long way and it's central to how digital single processors work, since they are optimized for that kind of thing and do it better than any of the standard desktop processors. Intel's SSE and SSE2 and AMD's 3DNow are other examples of SIMD. SIMD is capable of making some kinds of operations vastly faster, though the areas where it can be applied are actually quite limited. It is very useful in many kinds of graphics and sound transformation operations, which are an important group of computer applications these days. On the other hand, it's useless for such compute-intensive tasks as database manipulation, or any operation where data is moved around but not heavily transformed, like sorting. It speeds up some kinds of arithmetic but doesn't have any affect on logic and execution branching.

Apple's strategy until now has been predicated on the assumption that the G3 and G4 would approximately track in clock rate, and that the G4 would be somewhat more expensive but that its Altivec capability would make up for that, and thus that at any given time the G4 would be more powerful. So Apple has been buying from both vendors, but using the cheaper IBM chips (without Altivec) in their cheaper slower units (e.g. iMac) so that the advantage of Altivec would be in the more expensive higher end units being sold at a price premium. So what happens if a G3 becomes available at perhaps 50% higher clock rate than the fastest G4 from Moto and at a lower price -- but due to its lack of Altivec still runs "selected Photoshop filters" far less rapidly though running everything else (including most of Photoshop) far more quickly? What processor do you use in the top-end computers? And if they decide to go exclusively with blazingly fast IBM G3's and kiss off Moto's turtles entirely, then what message does this send to Apple's most important ISVs, who have spent a fortune writing their code to take advantage of Altivec, which Apple would just have repudiated?

The best possible solution for Apple would be for Motorola to get its act in gear and to start executing the way that IBM, Intel and AMD have been. This is exceedingly unlikely. For the last three years Motorola has been doing an extremely poor job, and has been losing money. Now it's seriously cut back funding and staffing at its semiconductor group, and it's difficult to believe that it will suddenly start executing better with 20% less staff. Sadly, this isn't likely to occur.

Part of the reason is that though Apple is Moto's single biggest customer, it's still a relatively small part of Moto's total business, and Apple's needs from its processors (blazing speed and be damned to the power needed) are diametrically opposite to those of most of Moto's customers doing embedded work (use as little power as possible and who cares how fast it is). That's why the first new processor version Motorola had released in a year, last summer (7450), was only marginally faster than before but used much less power. (The only way that Apple could utilize it was to create the notorious Cube, a commercial flop. It was expensive, no faster than before, but didn't have cooling fans. Apparently the market yawned.) That's what Motorola's customers (except Apple) really wanted: low power consumption. What Apple needs will not be useful to most of Motorola's customers.

IBM is in a different position there: while they do sell a lot of PPCs outside of IBM, their main use of them is inside the company for powerful mainframes and powerful workstations. So for IBM, blazing speed really is the first priority. Unfortunately, Altivec isn't; none of their systems have ever had it and IBM has shown no interest whatever in adding it. I also don't expect this to change. (Apple has floated the idea a couple of times and both IBM and Moto rejected it.)

No matter what comes out of this, Apple is going to make someone mad: their customers (because Macs keep being more expensive and slower than PCs) or their customers (because a cheapMac is faster on some things than a bigMac costing thousands of bucks more, which is already happening) or their ISVs (because Altivec isn't being used on the top-end Macs anymore), or everyone (if they toss in the towel, bag the PPC entirely, and switch to Sledgehammer or Alpha).

Stardate 20010510.1400 (On Screen): Ultimately any fair deal has to benefit both parties. In a business relationship, usually one side provides goods or services and the other side yields up money. So if the price for something rises a lot, then there should be a perceived increase in value.

I don't see one here. I can understand the naked greed of Microsoft in wanting to shift to a pricing model which both increases their revenue and also makes it more stable, such that people continue to pay for their product. What I don't see is why Microsoft thinks anyone would agree to this. Unfortunately, what's going to happen is that the smaller companies who have no bargaining leverage are going to get the shaft, while big customers (e.g. GM) will say "Sorry, we liked the old way better; write it the old way or don't bother talking to us."

It is not the case that Microsoft has no competition; their precise dilemma is now that their chief competition is their own three-year-old products. The customers presented with this new pricing option do have a choice: keep using what they have now. And in fact that is exactly why Microsoft wants to move to a per-year pay model instead of a once-only pay model. Their products have gotten mature and it's not really obvious just how Word can keep changing to justify regular upgrades. For what I want to do, I simply can't think of anything Office 2000 doesn't do that I need.

So Microsoft is looking the spectre of market saturation squarely in the face, and their solution is to charge infinite price for finite product, on the installment plan. I don't think this is going to fly -- the value proposition is not there for their customers.

Stardate 20010509.1930 (On Screen): There's been an extremely important development in the Rambus-v-Infineon trial. To review, Rambus sued Infineon on grounds of patent infringement, and Infineon counter-sued Rambus. There were 57 counts of patent infringement claimed by Rambus and last week Judge Payne dismissed them all. (Rambus announced that it will appeal to try to get its patent infringement claims reinstated.) He also dismissed most of the Infineon claims against Rambus but left three in place, which went to the jury today. It deliberated for four and a half hours, then found that Rambus was not guilty of violating RICO, but was guilty of two counts of fraud. It also awarded punitive damages to Infineon. That, now, is the only outcome of the trial. Rambus will appeal, naturally, but the consensus among observers is that it is very unlikely that an appellate court will override a jury decision.

Why this is particularly amazing is that Rambus had "home court advantage". Rambus was originally the plaintiff in this trial. Rambus chose the venue where it was tried. It was Rambus who insisted on a jury trial (in hopes of being able to use an emotional "We're just a bitty American company being picked on by the big bad foreign company trying to steal our property" victimization appeal). And they have now gone down in flames; this was generally considered to be their best chance of victory among all the trials they have pending.

The next stage in the trial will be a motion by Infineon for a judgement of "equitable estoppel" against Rambus. I had to hit Google on this one, and finally found an article describing it. What that term means is that a cheater is not entitled to justice. Someone going to court has to have "clean hands" themselves if they're claiming that someone else is cheating against them. In particular, with regard to intellectual property law, it means that if there is a proved history of cheating by the owner of a patent, then the court may permanently enjoin that owner from attempting to enforce its rights under the patent. In particular, this applies to a case where someone owning intellectual property is involved in an industry standards body and doesn't reveal it. Dell already lost such a case in the computer industry. Now a jury has found Rambus guilty of two counts of fraud precisely because of its silence about its intellectual property during its participation in JEDEC, which is exactly the same thing that Dell did. This establishes precisely the legal background for such a ruling. Infineon's motion is expected next week. Based on my knowledge of the case and the behavior of the judge to this point, I fully expect him to grant it.

The finding of fraud will drastically harm the Rambus company in a number of other regards. As far as I can tell, if Infineon is granted its motion, it will be a generic ruling applying to everyone, not a specific ruling applying just to Infineon. Second, if it is granted then the appeal on the dismissal of the claims of infringement will become moot, because even if infringement is proved Rambus will have lost its rights to enforce the patents on makers of SDRAM and DDR-SDRAM.

This also means that a substantial part of the revenue stream for Rambus will go away. They are reportedly receiving more than half their revenue from royalties on SDRAM.

Worse, they are now going to be embroiled in even more lawsuits. A lot of big companies have been paying them a l

Captured by MemoWeb from http://denbeste.nu/pregrey/20010501.shtml on 9/16/2004