Apple's choice of chips

kevinvanhaaren.net wrote:

We might see Itanium XServe's, but I really doubt it. I think Apple will go with the standard 64-bit extended Pentium with dual-core (and already rumors of Intel producing Quad Core chips.) Microsoft will only support Itanium for Server 2003. XP will be available in 64-bit for the Intel chips using AMD's 64-bit extensions. That's pretty much sealed Itanium's fate to be server only.

Apple docs only seem to talk about x86, so I seriously doubt they will touch Itanium with a 10-foot pole. I can just see all the excited developers out there hearing about needing to support another instructions set... NOT!

--Chris

Conrad Hirano wrote:

John Siracusa wrote about Apple's switch to Intel processors in Ars Technica.

<http://arstechnica.com/columns/mac/mac-20050607.ars>

What he wrote pretty much sums up how I feel. Perhaps I am a victim of marketing, but it seemed that the PowerPC architecture had a lot of promise. With Apple abandoning the architecture, that promise will never be realized.

That is a great article. I think there is actually a fairly obvious answer to the "why not AMD?" question that this article asks. Keep in mind Jobs emphasized the "bang per watt" issue. I suspect he wants to have the fastest notebooks out there (makes sense, because notebooks are now the bigger market and more importantly they are the higher margin market) more than he wants to have the fastest desktops out there. As great as AMD's chips and chipsets are, they just can't compete with what Intel has to offer in this area. Sure, AMD's 90nm CPU's are far more energy efficient than Pentium IV's (I didn't check, but I suspect a G5 can even compete fairly well with a Pentium IV in this regard), but Intel's Pentium M's are really in a class by themselves in terms of delivering top flight performance while still keeping power consumption low. It's kind of amazing because internally Pentium M's are said to more closely resemble the old Pentium Pros (famous for the amount of heat they produced) than Pentium IV's. Intel has managed to extract some amazing things from that processor core.

BTW, the Pentium M's are fast enough that a trend right now in the PC space is to put them in desktops. The only major disadvantage they have compared to Pentium IV's is that their memory bus is slower, and for a lot of applications this is trumped by the Pentium M's other advantages. It's not entirely inconceivable that Apple might just go with Pentium M's on their entire product line (Apple is very fond of having their machines cool and quiet). This would actually explain why the high end desktops would lag behind others in the conversion, as Intel's dual core Pentium M is still a bit of a ways out on their road map.

The other big question the article raised was about x86-64. I have to say I am also quite surprised that Apple isn't jumping to x86-64, but their docs are very specific about citing the IA32 ABI. I am speculating that they are going to stick with the model on OS X: kernel runs 64-bit, apps run 32-bit. They maybe have done this to make the porting job easier for developers, but realistically I'm sure every developer out there would tell them they'd rather make the transition to 64-bit now rather than later. This is particularly surprising as x86-64 actually brings the x86 ISA to something closer to the PowerPC, as it has more general purpose registers (enough so the standard x86-64 calling convention stuffs the first 4 arguments into registers like the PowerPC before falling back to using the x86's approach of pushing them on the stack).

The one thought that makes avoiding x86-64 in the short term make sense to me: Pentium M's don't support x86-64. :-)

--Chris

On 6/8/05 7:07 AM, Adam interjected into "John C. Welch"
<jwelchbynkii.com>'s message:

> [Indeed - this was notably absent. I wonder a bit if there will be a special
> Apple-only version of another Intel CPU for Macs. -Adam]

I hope not. That just puts Apple back where they are now: not big enough
to support the chip development by their sales.


[But it could be Apple-specific in name only - just a marketing thing. -Adam]


Also, I don't think it's "chip". I think it's "chips," with machines in the
Mini space perhaps sharing with iBook, and PowerBooks using a low power chip
used in contemporary Windows laptops, and high end using whatever is
impressive toward the end of 2007.

I suspect there's a reason for "low end first": Freespace has even less
reason to keep making chips for Apple than IBM has, and I would expect no
new chip designs out of them of interest to Apple. It will be interesting
to see whether Apple runs out of G4 chips before the Intel ducks are all in
line.

  --John

On 6/8/05 at 7:07 AM, jwelchbynkii.com (John C. Welch) wrote:

>At what point did Steve say what precise chip they'll be using in a
>year to a year and a half?

>[Indeed - this was notably absent. I wonder a bit if there will be
>a special Apple-only version of another Intel CPU for Macs. -Adam]

I would be very surprised to learn Intel was making an Apple only version of one of their CPUs. The non-recurring development cost for any modern CPU is a very large number. It is only recovered by selling a whole lot of CPUs. Somehow, I just don't see Apple's market share as supporting this cost.

There is also another cost with having an Apple only CPU. That is you need more engineering staff to improve it (speed bumps etc). Given Intel has a much larger market for non Apple only CPUs, they would clearly devote more of their engineering talent for these chips than an Apple only chip. That would leave Apple fighting for resources to improve thier specific chip. In essence, that would simply change the name of Apple's CPU problem from IBM to Intel.

So, it seems to me an Apple only CPU wouldn't be in either Apple's or Intel's best interest and is highly unlikely.

So, it seems to me an Apple only CPU wouldn't be in either Apple's or Intel's best interest and is highly unlikely.

It is much more likely that Intel would produce a custom chip set for a custom designed Apple mainboard. The chip set would contain a standard CPU and non-standard support chips. Creating the glue logic for PPC CPUs is something that Apple has had to do for themselves in the past but probably would be quite happy to pay Intel to do for them.

The one thought that makes avoiding x86-64 in the short term make sense to me: Pentium M's don't support x86-64.

Here is a more likely reason. Apple has not even started on a x86 64-bit version of OS X. Unlike the PPC which was designed with 64-bitness as part of its design philosophy, the x86 started out as a hybrid 16-bit/32-bit chip. There was even an 8 bit version that was used in the original IBM PC (8088). The consequence of this is that unlike the 64-bit stuff in OS X where it can be done piecemeal, the 64-bit version of x86 OS X is probably going to be done all at once.

I'm no expert on x86-64 (intel calls it EMT64) but from what I've read you can only set 64-bit mode on at once for the whole CPU with a 32-bit compatibility mode being available from within the 64-bit mode of the CPU. That 32-bit compatibility mode is not likely to be useful for kernel level operations in the OS.

The roadmap for 64-bit x86 OS X might be difficult and take a while which is another unfortunate aspect of the need to switch away from the PPC. I hope that Apple gives a clearer picture of this soon.

Ars Technica has a great detailed article which looks at Intel's
roadmap and what it means for the future MacIntels.

<http://arstechnica.com/columns/mac/mac-20050608.ars>

An excellent read for anyone who wants a little education on what's
coming down the pipe and why.

--Nik

jdb wrote:
> Here is a more likely reason. Apple has not even started on a x86 64-bit
> version of OS X. Unlike the PPC which was designed with 64-bitness as
> part of its design philosophy

? not really. It's design was drawn from a 32-bit design, and it used a
52-bit flat address space. That hardly seems like it was planning for
64-bitness.

>, the x86 started out as a hybrid
> 16-bit/32-bit chip.

I'm not sure which chip you are referring to. The only one that would
match that description would be the 386sx, and it'd be hard to argue
that was the start of x86.

 > There was even an 8 bit version that was used in the
> original IBM PC (8088).

Actually, the 8088 was a hybrid 8-bit/16-bit chip. It was basically the
8086 (a 16-bit chip) with an 8-bit bus.

> The consequence of this is that unlike the
> 64-bit stuff in OS X where it can be done piecemeal, the 64-bit version
> of x86 OS X is probably going to be done all at once.
>
> I'm no expert on x86-64 (intel calls it EMT64) but from what I've read
> you can only set 64-bit mode on at once for the whole CPU with a 32-bit
> compatibility mode being available from within the 64-bit mode of the
> CPU. That 32-bit compatibility mode is not likely to be useful for
> kernel level operations in the OS.

The same is true with the G5. It runs all kernel space in 64-bit mode
(if you know a thing or two about programming you would run screaming if
you tried to imagine a kernel that switched between 32-bit and 64-bit
mode), with programs in user space running in a 32-bit compatibility
mode. The same is done with x86-64. If anything, the transition to
x86-64 is actually easier.

> The roadmap for 64-bit x86 OS X might be difficult and take a while
> which is another unfortunate aspect of the need to switch away from the
> PPC. I hope that Apple gives a clearer picture of this soon.

I don't think there is a lot of evidence to support this notion. The
x86-64 ISA is so similar to the x86 ISA that porting is generally very
easy (which is exactly what AMD had in mind of course). Linux was able
to support x86-64 fairly easily, and more importantly FreeBSD (OS X is
built on FreeBSD) already has x86-64 as one of its Tier1 platforms. The
Mach kernel would have to be ported as well, but the NeXT/OS X version
of the Mach kernel is claimed to be easy to port these days because of
all the porting they've done with it in the past. Given that the OS team
has already dealt with endianess and 64-bit issues in the past, I'd be
suprised if they hadn't already completed an x86-64 port internally. The
only thing that suggests they haven't is the noteable absense of a
Darwin x86-64 release.

--Chris

James Bailey wrote:
> On Jun 10, 2005, at 4:44 PM, Christopher Smith wrote:
>> jdb wrote:
> Apple makes the claim here:
> http://www.apple.com/g5processor/
>
> "[...]because Apple and IBM designed the PowerPC architecture to scale
> from 32- to 64-bit from the beginning."
>
> I'm too lazy to find a less marketing driven quote but I've read it many
> times.

Yeah, the actual reality is far different from that. The one bit of
truth is that when the PowerPC alliance was first formed, the roadmap
included the 601, 603, 604 and 620 processors. The 620 was designed to
be a 64-bit chip. Of course, by then all the major workstation chips had
roadmaps going to 64-bit as well, and much like the PowerPC folks, few
of them had actually figured out the engineering behind how they were
going to get there, let alone incorporated it in to their current designs.

The POWER architecture, however, was not designed for 64-bit from the
get go, and the PowerPC ISA was the POWER ISA, give or take a few
instructions.

The actual chip cores themselves were not designed with 64-bit support
either. The 620 was a completely different design from the 60x chips.

The biggest thing though, is that the 620 was a complete failure. First
of all, they defined a new set of instructions for the POWER/PowerPC ISA
to support 64-bit (it wasn't there from the get go). Then there was the
actual track record of the chip. There were three attempts made to bring
the 620 market. In each case, by the time the 620 was "done", it turned
out to be more expensive to make, ran slower, ran hotter, and had more
bugs than the 604. In the end it was abandoned. So much for the
advantages of having designed the architecure to scale to 64-bit "from
the beginning".

That being said, IBM did actually get 64-bit POWER architecture chips to
market in the form of the POWER3 and POWER4 series (and the specialized
RS64 series). I'm sure their experience with those processors made it
much easier to design and develop the G5.

In generally, planning something 15 years ahead of time in the chip
engineering business is rarely going to help you much anyway. By the
time the 15 years have passed, the whole market will have changed and
sticking with your engineering plan from 15 years ago is going to
prevent you from being able to take advantage of many of the lessons you
learned along the way.

Keep in mind the Itanium was engineered for the get go to be 64-bit, and
yet even with Intel's might behind it, it has been all but vanquished by
the x86-64 designs from AMD. Sometimes incremental stuff is actually
better, even if it makes us crazy to think we're still carrying baggage
from two decades ago.

>> The same is true with the G5. It runs all kernel space in 64-bit mode
>> (if you know a thing or two about programming you would run screaming
>> if you tried to imagine a kernel that switched between 32-bit and
>> 64-bit mode), with programs in user space running in a 32-bit
>> compatibility mode. The same is done with x86-64. If anything, the
>> transition to x86-64 is actually easier.
>
> I'm not up to speed on either PPC 64 bit nor x86-64. I hope you are
> right. I'm going to do a little investigation but I haven't yet. I
> haven't been keeping up with computer architecture much recently. The
> CPU is just not as important as it once was and I don't write kernel
> code or compilers. I haven't even written a device driver in 10 years.
>
> Thanks for the information.

I should clarify one thing. While the x86-64 supports the 64-bit
kernel/32-bit user space model (indeed, this is one of it's key selling
points), it isn't a requirement. I know lots of Linux folks using
Opterons who have almost everything 64-bit (up until recently MATLAB
wasn't available in for x86-64, so they used the 32-bit version instead).

--Chris

>>, the x86 started out as a hybrid
>>16-bit/32-bit chip.
>
>I'm not sure which chip you are referring to. The only one that would
>match that description would be the 386sx, and it'd be hard to argue
>that was the start of x86.

Old saw:

Windows is now a 64 bit tweak
  of a 32-bit extension
  to a 16-bit user interface
  for an 8-bit operating system
  based on a 4-bit architecture
  from a 2-bit company
  that can't stand 1 bit of competition.

--
_______________________________________________________________
Keith Dawson Layer of ash separates morning and evening milk.
kadawson "at" mac(dot)com http://technologyfront.com/

Google Kreme wrote:

On 10 Jun 2005, at 13:41 , David Weintraub wrote: Cringely: > Question 3: Where the heck is AMD?

Not making laptop chips.

Actually, they do make laptop chips (even 64-bit ones). It's just that the Pentium M and its related chipsets are so much nicer.

--Chris

Keith Dawson wrote:
> Old saw:
>
> Windows is now a 64 bit tweak
> of a 32-bit extension
> to a 16-bit user interface
> for an 8-bit operating system
> based on a 4-bit architecture
> from a 2-bit company
> that can't stand 1 bit of competition.

Yeah, the only catch there is that today's Windows is built on Windows
NT, which is not a 32-bit extension, but a ground up implementation of a
32-bit operating system with an emulator for the 16-bit environment
(much like 680x0 emulation with PPC's). It also was designed with the
notion of supporting 64-bit from inception, although the original 64-bit
platform was Alpha (and it took a very long time to get that done due to
declining interest in NT on Alpha), following by Itanium, and now
finally x86-64.

--Chris

Keith Dawson wrote:

> Old saw:
>
> Windows is now a 64 bit tweak
> of a 32-bit extension
> to a 16-bit user interface
> for an 8-bit operating system
> based on a 4-bit architecture
> from a 2-bit company
> that can't stand 1 bit of competition.

Rather funny when you consider who that 2-bit company is.

--On June 14, 2005 2:51:15 PM -0700 Christopher Smith <xxman.org> wrote:

> Yeah, the only catch there is that today's Windows is built on Windows
> NT, which is not a 32-bit extension, but a ground up implementation of a
> 32-bit operating system with an emulator for the 16-bit environment (much
> like 680x0 emulation with PPC's). It also was designed with the notion of
> supporting 64-bit from inception, although the original 64-bit platform
> was Alpha (and it took a very long time to get that done due to declining
> interest in NT on Alpha), following by Itanium, and now finally x86-64.

Don't forget that NT actually ran on PowerPC at one time too. I've never
looked at XBox but some say it's a version of XP Embedded and it runs on
PowerPC Cell for XBox 3 (which was demoed running on G5's).

Kevin

I think we're missing one of the biggest implications of this news. Everyone mentions that Powerbook g5 issue, but has anyone really thought about it?

Thanks to those hot G4 and G5 processors, Apple engineers have become wizards at heat and power management techniques. Look at what they've managed to do with Powerbook sizes and battery life, quiet G5 desktops, and the iMac. Now imagine what they can do with low power processor chips with power management built in, yes, I'm talking Centrino :-(. This opens up a whole new era for Apple in the portable market. Radical new designs for laptops with amazing battery life, ever smaller and quieter iMacs, and consumer electronics that don't sound like an airplane engine. Can an Apple tablet be far behind?

Joe Krajnak