Re: How many times must Apple make such DRASTIC architecture changes?

[Tim Cutts <timc@chiark.greenend.org.uk>]

In article <1118326364.4d45ce43aa1454279edfdf55a8ae4b33@teranews>,
Paul Schlyter  <pausch@saaf.se> wrote:
>Moore's law doesn't seem to hold as much anymore...
>
>
>In 1994 I bought a personal PC: a 64 MHz 486 with 8 MB RAM and 1 GB HD
>
>In 1998 I bought another personal PC: a 300 MHz P-II with 64MB RAM and 8 
>GB HD
>
>In 2002 I bought another personal PC: a 1.8 GHz P-IV with 512 MB RAM and 
>120 GB HD
>
>Next year it's time for my next PC, which should be a 10 GHz P-VI with 
>4096 MB RAM and an 1800 GB HD ..... I don't think that will happen....

Those figures are nothing to do with Moore's Law.  Moore's Law is often
misquoted to be about speed doubling every 18 months, or whatever it is,
but that isn't what he said.  Moore's Law is purely about transistor
count in processors, and it has held, more or less.  For a while,
performance increased at the same rate, so Moore's Law got abused for
that as well, but now we're seeing that second and incorrect use is
starting to show its faults.  As far as I know, though, Moore's Law is
still holding for chip transistor density.

Clock speed isn't rising very fast now, but performance and clockspeed
are only loosely related anyway.  At the bioinformatics applications I
deal with in my day job, a 1.6 GHz Intel Itanium2 is roughly 50% faster
than a 2.8 GHz Xeon, running the same suite of programs, and in both cases
compiled with Intel's compiler.

Different CPUs are good at different things; one reason for Apple's
switch, which Jobs alluded to, is that x86's integer performance is
blistering (in both Intel and AMD's processors).  Way faster than
anything else, and in the context of high performance computing, I've
tried almost everything that costs less than $250,000 in today's market.

IBM's POWER processors are very good at running very tightly coded,
floating point intensive code.  In my tests, they are much less good at
running run-of-the-mill, integer intensive codes such as the desktop
productivity applications that are the bread-and-butter of personal
computing.  And that's why Apple have switched.  They've designed their
OS to be CPU agnostic from the start (after all, it's just a new version
of NextStep, which was itself CPU agnostic anyway) and all most software
developers will need to do is recompile -- at least if they've been
developing in Cocoa.

And for those vendors shipping more finely tuned applications, the
switch to x86 is also a blessing; now they only have to fine tune the
CPU intensive parts of their code once, and then link it into both their
Windows and OS X ports.  That's a big saving of time and money.

My main worry about the switch is that enterprising hackers will find
ways of running OS X on non-Apple hardware.  And if they do that, and
dry up Apple's revenue stream from hardware sales, we could be in
trouble.

Tim