Heat And Apple
I got two emails this morning. One asked me to comment on a blog talking about the heat-related reasons why Apple would decide to go with Intel; the other asked me why AMD wasn’t chosen rather than Intel, and the prospects of AMD ever being asked.
I think the responses would be generally informative, so here they are:
Heat: This is long, but a lot of background is needed
to understand why Apple did what it did in this
situation, and why it makes sense to go Intel rather
than AMD.
This is a tale of two extremes at Intel.
The mainstream Intel CPU is the PIV in its 90nm
incarnation called Prescott. When it comes to heat,
it is a TERRIBLE chip. The reason for that is the PIV
chip design leaks power. Lots of power; it was around
40% in the 130nm version of the chip, and that figure
is certainly worse now. Intel basically ran with a
design until it broke.
On the other hand, Intel also has its line of
notebook chips (aka Pentium M, codenames Banias,
Dothan and the upcoming Yonah. The
story behind these is that Intel Israeli engineers
were given the old PIII design and told, “See what you
can do with this for a notebook design.” They did a
lot.
The upcoming 65nm generation of PIV processors will be
a blend of the two technologies. What’s important to
note is that all areas of Intel has been obsessed
about power consumption the last few years. On the
desktop side, they delayed Prescott for a year, and
have continued to tweak it to cut power. They’ve also
been doing a lot of basic research to cut down
leakage, too.
On the other hand, current PowerPCs/AMD chips are
rather cooler than Prescotts because they use
silicon-on-insulator (SOI) technology.
It should be noted that AMD has had a lot of problems
with SOI and has been largely dependent on IBM to
solve its problems; AMD has paid IBM at least a
hundred million so far for CPU/SOI help. For
technical purposes, AMD’a SOI technology is
essentially an x86 version of IBM; it has the same
plusses and minuses.
AMD is just reaching crossover (i.e., new generation
of CPUs outselling previous generation) now with
Athlon 64s twenty-one months after product
introduction, a process which normally takes
six-months. They are now having problems with dual
core. This is a big part of the reason why Apple
didn’t go with AMD, more below.
However, everything is relative. SOI basically buys
you about a CPU generation of time before you have to
get serious about power consumption. The power
consumption of AMD’s Athlon 64s and Opterons is higher
than that AMD’s Athlon XPs, but not too much more so.
However, the G5 chews up a good deal more power than
previous PPC generations (IBM pretty much has been
evolving the PPC’s design in Intel’s direction to get
more speed; the G5 is a lot more like the PIV (more
power, longer pipelines) than the G3 or G4).
SOI also has some other limitations. SOI does quite
well in the power consumption area at low-to-moderate
speeds. Try to push it past that boundary, and power
consumption skyrockets. Past a certain (relatively
low point), it just stops working, and it’s not just a
matter of heat. Overclockers have found that they can
push Prescott a lot provided they use extreme methods
to cool it; they’ve had considerably less success
with Hammers (even after adjusting for different
IPCs).
Nor does SOI do terribly well in low-power situations.
Intel’s mainstream Dothan chews up a max 27W at top
speed; the AMD equivalent chews up 35W. AMD is only
starting to pay attention to power reduction on their
mobile chips, until recently, they just took the best
of the litter from the desktop and made them mobile
chips.
In short, Intel has been taking an intense crash
course in power reduction the last few years, and the
results of their homework should be apparent starting
next year. IBM/AMD thought SOI WAS the crash course,
and only recently realized that it was not.
Why Not AMD?
AMD may well have a shot at powering some AMD models
in the future. I can see two reasons why Apple would
be reluctant to use AMD any time soon:
1) Doubts about AMD being able to reliably supply chips. If I
were Apple, given what AMD has done with Hammer the
last couple year, I wouldn’t feel too confident about
them delivering, at least not until Fab 36 is up and
running.
2) Per process technologies (i.e., SOI), AMD is little
more than an x86 subsidiary of IBM these days. Apple
obviously doesn’t have confidence in IBM’s being able
to keep up with Intel on power consumption when we get
to dual- and quad-cores, so why should they have any
more confidence in AMD?
Of course, AMD will
inherently have a greater interest in this subject
competing against Intel than IBM will, preoccupied
with making Cells, so I suspect Apple’s attitude to
AMD is “Show me” hoping that they can so they can play Intel
and AMD off against each other. I also suspect
that the two year lag is at least partly meant to give
AMD a chance to “show them.”
P.S. If you want to look at personalities, there’s no reason for Steve Jobs to love Hector Ruiz, and some pretty good reason not to.
Ruiz headed the chip division at Motorola at the time the PPC began to trail x86 technology badly.
Granted, sometimes a new boss gets handed a poor hand of cards to play, but most of the Apple account got turned over to Motorola not that long after he left, so he didn’t exactly turn the situation completely around.
That’s even more reason for Apple to have a “show me” attitude towards AMD.
Be the first to comment