Intel Inside For Apple?

iomud writes "Bear Stearns analyst Andrew Neff predicts that there's a better than 80 percent chance Apple will make the jump to Intel in two to four years. As the relationship with Motorola seems to be weaning the question may be what chip would you like to see in next-generation Macs and why?" It seems important to note that Bear Stearns owns shares of Intel and Dell, and has a banking relationship with Dell and HP. Oh, and even if it didn't, that I can't see any reason why anyone should care what Andrew Neff says. But that doesn't mean it can't be fun to talk about!

Comments

The real question is...

[BitGeek]

We all know that PowerPC chips get far more done in a given clock than x86 chips.

This was the great promise of the PowerPC, actually. By going to a superscalar Risc architecture, IBM and Motorola spent the effort to get a chip that really did more per clock.

The clock rate, however, is less of an engineering issue than a process issue. Intel has processes that increase their clock rate rather fast-- and so rather than re-engineering their processors (and paying the backwards compatibility penalty that apple paid when they switched from 68k to PPC) they have simply increased the clock rate and integrated more on chip cache, etc.

The thing is, this means that the PPC was at a very significant competitive advantage-- its really hard to beat architecture engineering, which the PPC has in spades, but pentiums lack. Design is hard. Process is easy. So, the Processes that Intel was using should have migrated to Motorola and IBM, and we should be seeing PowerPCs that run at 2GHz and leave no question as to the fact that the powerpc is much much faster.

So, the real question to my mind is-- why hasn't the process side of the house for PowerPCs kept up with intel? Certainly motorola and IBM have the know how, and they have the motivation-- competition with each other for the sizable sales to Apple, and the possibly even larger embedded and workstation markets.

I can think of two possibilities:
1) The increased complexity of a super scalar architecture on the order of the PPC makes timing more problematic and while process is there for higher speeds, the synconization of the clocks hitting all the subcomponents of hte processor at the same time is an issue. At these levels, the speed of light is a real factor when one signal goes a little further than the other, they arrive at the same place at different times due to the relative slowness it takes for the signal to go down the longer path.

2) Conflict. Motorola created Altivec and apple jumped all over it, and I don't believe IBM has a license to Altivec, giving motorola a bit of a monopoly. This combined with apple embracing altivec so much means that Motorola may not have sufficient incentive to grow the speeds. Plus, since the PowerPC has not had the widespred platform support that was expected-- NT for PPC has gone away, other Unix box makers aren't using it extensively, the market is smaller than was originally intended.

This creates quite a problem for apple. As long as they suffer from the perception- despite the reality-- that their processors are slower because people think MHz = speed-- they are going to have trouble not being seen as more expensive. Hell, even people who post here make this mistake.

So, I think Apple is planning something big. But it won't be a switch to x86, certainly as we know it.

I can imagine a couple possibilities:
1) Apple teams with AMD and brings the PPC instruction set to a future AMD processor that can handle it and the x86 instructions simultaneously. Gets AMD's process speeds, along with PPC compatibility running at native speeds (rather than emulated.) The downside is that IBM would have to agree to this, and its not clear what IBM's upside is-- unless IBM is part of the alliance and gets a competitive advantage to using this technology in its products (maybe low end power workstations)-- but still Motorola which controls altivec would have to be involved.

2) A new AIM partnership, this time its the AAIM partnership, all four companies collaborate on a new chip that will run OS X and Windows, IBM and Moto make PCs that dual boot, AMD gets Altivec and Power4 Multichip module technology, and IBM and Moto get AMD process technology, and IBM, Moto fab the chips for AMD. This gives IBM a weapon against windows, namely OSX, gives AMD the backing of two big competitors- IBM and Moto, along with a new customer, gives Moto a new jumpstart into the box making business that it gave up when Apple stopped subsidizing the clones industry.

3) The Death By Numbers Approach -- Apple goes to IBM and gets the four chip Power technology and migrates there from PowerPC, greatly increasing the volumes of these chips for IBM which is only currently using them in their servers and workstations. This drives down the costs, apple doesn't have to rewrite software (like quicktime) that was never part of the NeXT OS, and at the same time can emphatically claim the "fastest PCs in the world" title it now holds but nobody recognizes. Oh, and they sell them with 2 to 4 processor units per box.

4) Death By Numbers part 2-- apple starts shipping quad and 8 way PowerPCs running at moderate speeds, 1-2GHz using Motorola (or IBM) chips, and being competitive on price because the powerpc costs them so much less per cpu than Intel CPUs. Thus people will instinctively know that 8 1GHz CPUs are going to get a lot more done than one 3GHz intel cpu.

5) The Second Rebel Alliance-- Apple, AMD and Nvidia team up on an x86 processor that uses NVidea and AMD Hyper IO (or is it rapid io?) technology, and apple does go the x86 way..

The thing is, 5 seems least likely to me. apple has just migrated accross platforms for the second time-- the first was 68k to ppc, and the second is classic Mac to OS X. Applications have to be re-written.

Are they really going to ask their developers to re-write their apps yet again, in only a few years? I really doubt it.

So, I think there is a new processor architecture or solution coming-- I'm sure apple recognizes that the PPC has not given it the marketability it needs.

But I think that solution will be PPC compatible natively.

Re:The real question is...

[danielwright]

You're oversimplifying things a little too much when you say that if a PowerPC chip were made with the same process that Intel uses for it's new P4s, it would have the same clock rate.

Modern CPUs are all pipelined, so they divide each instruction into several pieces - say Instruction Fectch, Instruction Decode, Execute, Load/Store, Write Back for example. Then, they interleave the execution of the different stages, so while one instruction is being decode, the next is already being fetched.

At a very rough approximation (it's much more complicated than this), the clock rate has to be low enough that the largest of the pipeline stages can execute in one clock tick, so if tou divide up the execution into more, smaller stages, you can raise the clock rate higher. However, there's a lot of complex machinery to avoid "hazards" where instructions depend on each other, so they have to stall some of the instructions, and this gets more complicated and slower with a longer pipeline. (This would be a gross simplification 10 years ago, and today's CPUs are much more complicated, but it gets the main point across).

The designers of the current PowerPC implementations chose fairly short pipelines (I'm not sure of the number of stages, but I think it's around 5), while Intel uses 20 stages for the P4. That means that the P4 can run at a higher clock rate, but get less done per cycle because more of the instructions are stalled.

So, my point is, at least IBM has CPU processes at the same level as Intel's, if not better - it's due to the fundamental design of the chip that the GHz number is lower, which makes the GHz a very uninteresting measure - hence the "MHz Myth".

Also, PowerPC is an instruction set, like IA32 or IA64, it's not a chip architecture. IBM and Motorola currently make chips that implement the PowerPC instruction set (and IBM's chip, the Power4, is currently the fastest chip available, BTW).

Just to add to the list of totally unfounded predictions, here's mine:

IBM released the Power4 a few months ago, as the fastest chip on the market. They want to use it for every server platform they make (AIX boxes, mainframes and AS/400 boxes). It's designed for servers, and that shows - you need something like 1 ton of force to attach it to the motherboard, and a pretty impressive cooling system as well. This makes it unsuitable for small desktop machines like the imac, and for laptops. Also, it doesn't support Altivec. I figure, they'll work out some licensing agreement so they can make a special, slightly slower version for Apple that does support Altivec.

The merits of this: they could use basically the same CPU design and processes (which are very, very good), and now software changes.

I don't think Apple can change to Intel chips because that would require new versions of all the software. They've just asked all their customers to replace old OS9 software with OS X software. If they came back in 2 years and said everyone should replace all their software again, their customers would start to get rather irritated by it...

Re:"Performance Boost" a result of the MHz myth?

[Toraz Chryx]

No, calculations != cycles

Different processors can handle a different number of instructions per cycle.

and hence, require a different number of cycles to perform the same calculation.

Re:"Performance Boost" a result of the MHz myth?

[DavidRavenMoon]

The ultimate smackdown: Mac versus PC

Dual 1GHz G4 versus 2.2-GHz Sony Vaio RX690G Digital Studio.

"Rather than argue the point, I decided to conduct my own comparison. Apple even provided the actual Photoshop picture used, a full-color photo over 44 megabytes in size, depicting seven bike riders with colorful bikes and helmets. This is the sort of file that graphic artists have to manipulate on a daily basis.

"Apple also sent me a copy of its actual test protocol, including a Photoshop Actions file, a set of scripts that automated the various rendering functions. They also provided a high-end twin-processor desktop Power Macintosh on which to run the tests.

"The Mac was upgraded to the latest version of Mac OS X, 10.1.5. The Sony had Windows XP. I installed the standard retail versions of Adobe Photoshop 7 on both computers."

"Running the tests proved exceedingly simple because Photoshop displays the actual timing of a rendering process rounded off to tenths of a second. Per Apple's directions, I conducted each test four times to deliver the most accurate results.

"Like all Adobe applications, Photoshop is a bit slow to launch. It took 15 seconds on both computers to get ready for the main event.

"In the nine test runs, the Mac came out on top five times, besting the Sony by up to 8.1 seconds. Where the PC emerged victorious, the margin was usually less than half a second.

"In all, the Mac took a total of 35.5 seconds to complete the nine rendering steps. The PC took 50.1 seconds, making it 41% slower according to my calculator's reckoning.

...

"The upshot of all this, however, is that, when someone tells you a Windows box is always faster than the Mac, point them to this article and tell them it isn't necessarily so."

Of course the PC beat the Mac in a game of Quake ;)