« Deep Thoughts | Main | Fortune of the Day »

June 07, 2005

Apple vs Intel (II)

OK. I was wrong. Without being too technical, let me try to explain why this third bad decision is probably the dumbest of all time...

In the world of processors, there are basically two types: CISC and RISC. Once upon a time there was a lot of debate about which was better, the bottom line is that RISC is simpler (and faster). This simplicity increases the complexity of coding (in assembler) which offsets the additional speed inherent in the chip. Nobody today really cares one way or the other except for one small thing: electron tunnelling. This is a problem that creeps into very dense electronic designs. Like CISC processors. The problem is what has stopped any real increase in the Pentium line. The problem has not yet affected RISC chips like the PowerPC series which still have another two or possible three iterations (each iteration is a doubling of performance).

Moore's Law
What does that mean in practical terms? It means that Moore's law (that performance doubles every eighteen months) does not hold any longer for Intel (current expectation is perhaps one more iteration without a significant technology change). On the other hand, the PowerPC has more potential growth - if pushed (by IBM), it will outstrip the Pentium easily.
Why doesn't Intel do RISC? Well they tried. It was Itanium. It was a disaster. That is why Intel are still tinkering with the Pentium.

Non technical comparison?
How about a non-technical analogy? Imagine, you are on a slow hydrofoil. You know that hydrofoils are going to get better and a hydrofoil is still faster than an ocean liner. Nevertheless, you commit all your passengers into climbing aboard the Titanic.

Short term results...
Why, at this point, would anyone buy a PowerPC-based Mac? Why would I replace my old G3's with a G5 - knowing beyond all doubt that the machines will be obsolete in less than a year - not just slower but actually incompatible. This will immediately harm all Mac sales.
It will also annoy IBM - Apple's partner. The one company that is actually innovating in the world of computer chips (unlike Sun and Intel). There is no incentive for IBM to make any more improvements to Apple's version of the PPC. There is no reason to push chip shipments and resolve any difficulties. In fact, Apple had better have a fairly significant stack of processors up their sleeve...
Finally (for this section) there will be a whole bunch of software that gets delayed. Why strive to port something to the Mac when you will have to re-port it in twelve months time.

Long term results...
The long term results are worse. Apple will have committed to a processor with no future. In 2 to 3 years they will be scrambling to move to AMD or SPARC or whatever. Possibly even back to PowerPC - in the meantime, the market will have moved on and instead of being able to dictate to IBM (as they do now), Apple will have to adapt to whatever they can get.
There will be intense pressure to allow customers to run MacOS X on non-Apple sanctioned hardware (i.e. Joe Blog's home-built PC's) which will destroy profitability as Apple looses the cross-subsidy between hardware and software. Some "pundits" were pointing out that MacOS X is cheaper than windows - that is only due to the cross-subsidy.

What about the shareholders?
If I was a shareholder (hang on: I am a shareholder), I would be praying that IBM would just buy the company and stop this stupidity before it starts. The whole thing is insane - insane because it will destroy profitability now, insane because it destroys future CPU growth, insane because it will destroy what makes the Mac great. The Mac will become no more than another PC vendor and MacOS X will join the legions of niche Unix implementations for Intel boxes.

Posted by Ozguru at June 7, 2005 12:00 PM


Note: This piece was written around 6:30 AM Australian time but is dated at midday so that it will remain at the top of the posts until tonight. It was written *before* I saw the video stream (which is not currently working anyway).

Posted by: Ozguru at June 7, 2005 09:56 AM

One thing to point out is that Intel could use the Itanium as the apple chip, after all no one else wants it. Intel does have the spare manufacturing cap to pump out as my chips as needed be apple.

But I agree that would be stupid, while apple is a good porter asking devolpers to do more then once every ten years is sucide.

Posted by: skipjack at June 7, 2005 02:45 PM

"The one company that is actually innovating in the world of computer chips (unlike Sun and Intel)."

What about Sun's Niagra core in development. Started back before dual core was the next big thing. Will include 8 cores, giganet controller, memory controller, potentially a crypo core. Lots of innovation.. That said, the Power5 is quite a nice little piece of silicon.

Posted by: 'gim at June 7, 2005 08:36 PM

You may be right.

My hesitation swings around the fact that Sun appear to be lagging in the design/implementation side of the chips. Despite an impressive roadmap they seem to stumble on the little bits (like the memory/cpu arrangements in the early 280R) and the failure to convince software vendors (like Oracle) about dual-core. Contrast this to IBM who are publishing white papers on the optimisation of cpu functions via localisation of memory (before the 280R shipped) and the implementation of multi-core within single processors (before Oracle announced that dual core = 2 for Sun). Sun may be doing it first but they end up looking like a follower.

As a big Sun fan, I find that disappointing. I work alongside AIX-bigots and at times it becomes hard to push the Sun-wagon :-)

Thanks for dropping in with a comment...

Posted by: Ozguru at June 9, 2005 09:58 PM

I think Sun/Apple would make a good combination as long as Sun didn't try to put Solaris on the Mac or MacOS on Sun.

Mind you, it would be a big leap for Sun to go back to BSD...

Posted by: Rofl at June 10, 2005 12:31 AM

Actually, RISC chips these days aren't significantly simpler than CISC chips; the decoder that turns x86 instructions into Pentium 4 or Athlon micro-ops is only a small part of the chip.

And the problems with leakage have hit everyone at the 90 micron node. Jobs pointed out that IBM couldn't ship the 3GHz G5, but the 4GHz P4 has been KIA by exactly the same problem.

Everyone is going dual- and multi-core now; it's the only way forward.

Posted by: Pixy Misa at June 12, 2005 10:57 PM