Cell Processor used for the next Powerbook???

[nietzsche2131]

Alright, I think that the cell processor could be a possible canidate for the next powerbook. I mean it's a powerpc variant. But the thermal output of the processor is really cool, and probably would not need liquid cooling. Plus, the power is way over 4ghz, the Playstation 3 is supposedly around 4.9ghz. The processor is said to be a "OS neutral chip", but still having the cell could make for a very good looking notebook. But then again the processor to even fit something such the current casing would require that it be smaller cell chip and around 3.2 ghz, but hey that's still really good. Anyways I just like to hear thoughts on this one because the G5 really is locked in a position of not being shrunk down enough. I mean a powerbook is a powerbook to me, as long it's running a powerpc variant and OS X.

 

[fryke]

What if it'd be an intel chip? ... Well: Cell isn't here. And it isn't _aimed_ at notebook uses. It could be as neutral as it wants, that does _not_ make it a good notebook processor, and not a good processor for Mac OS X either.

 

[ora]

What if it'd be an intel chip? ... Well: Cell isn't here. And it isn't _aimed_ at notebook uses. It could be as neutral as it wants, that does _not_ make it a good notebook processor, and not a good processor for Mac OS X either.

Fryke, I'm interested (and no chip expert), why do you say it wouldn't be a good chip for OSX? That said, I couldn't imagine a new chip debuting in a powerbook, too many issues of space, heat, experience with the architecture etc.

On th eother hand, when I saw the Cell specs it looked pretty dam cool, plus I did see a presentation at the IBM research labs in Zurich which mentioned the G5 and multicore (though i know cell isn't the same) on the same slide. I suppose that gave me hope of seeing some super-multi-G6 chip at some point. Am I being stupidly optimistic or just looking too far into the future?

 

[Mikuro]

Migrating to Cell would be a bit of a hassle. Not an x86-type hassle, because it wouldn't break support for current software (I think...), but it would force developers to learn how to optimize software all over again before any performance gain could be achieved. Remember when the G4 first came out, and it was pretty much exactly the same as the G3, because no apps were optimized for AltiVec? It'd be pretty much the same deal all over again. And AltiVec is hard enough to program for as it is; Cell is like AltiVec^2 as far as difficulty goes.

The reason the Cell is so powerful is because it uses a bunch of separate mini-processors, called SPEs (synergistic processing elements), controlled by one full-blown PPC. You can think of each of these SPEs like a standalone AltiVec unit, except — and this is important for any Cell-based Mac theories — they are not based on AltiVec. They're not as flexible, and they're not as powerful. They do what they need to do in the PS3, and they do it damned well, but that doesn't mean they'd make good desktop workhorses.

I wouldn't be really surprised to see some Cell-type architecture show up in Macs eventually, but I think it's a bit early, and I would expect to see a much more desktop-centric variant of the design that anything we've seen so far. The interesting thing about the Cell is that the design itself is very flexible — you could plug in as many SPEs as you want, and you could, in theory, use a completely different kind of SPE that anything that's currently been discussed. That's why I keep referring to it as an "architecture" rather than a "chip". The Cell as we (barely) know it will never be used in Macs; that much I feel confident in saying. But we might see a Mac-worthy processor that follows pretty much the same formula at some point. Although, whatever the specifics of the design, I think it would still require major re-optimizing of software.

Also, the Cell isn't really ready to go yet. By the time it is, I expect to see the current G5 advance far enough to make a move rather silly. By that time, we should have 2-core (or maybe even 3- or 4-core) G5s (and even G4s), which would present a big leap in power and a virtually seemless boost in performance (most MP-aware apps should be MC-aware right out of the box, so programmers won't need to re-optimize everything, IINM).

It's anybody's guess as to when we'll see multi-core G5s, though, really. From what I was hearing about last year, I honestly expected them by now. I still expect them by the end of the year — my feeling is that the whole 2.7GHz G5 was purely an interim upgrade to tide us over until the real stuff is ready. I could be entirely wrong, of course. But the PS3 isn't coming out until next year, and there's no real reason to think the Cell will be finalized very long before then.

 

[texanpenguin]

You won't ever see more powerful chips in the PowerBook range than in the Power Macs, that's for sure.

And personally I doubt we'll see a new PowerBook chip (G5 or otherwise) for some time. Jonathan Ive didn't make it easy for himself. The form factor of the PowerBook is just too slim to use with any of these super powerful chips (pretty much anything 64-bit). Liquid cooling (to whomever keeps suggesting it) is totally impractical in portable machines (too volatile). I think we'll get dual-core G4s before we get G5s in PowerBooks. G5s probably won't ever make it into PowerBooks in their current incarnation. They need to make a G5-M, or some such if it's ever to work. And it still has to be significantly more powerful than the G4 (and without much more battery drain).

That's really the only thing that adds credence to the Pentiums in Apples rumour - they might make a range of PowerBooks that are x86, and which you can run Windows on.

Heh, wouldn't THAT stir up the masses .

 

[ElDiabloConCaca]

The G4 is not difficult to program for at all... making something "vectorized," which is what AltiVec is all about, is pretty simple in most cases.

Instead of:

for (i = 0; i < 20; i++)
  myArray[i] += 1;

You would write:

for (i = 0; i < 20; i += 4) {
  myArray[i+0] += 1;
  myArray[i+1] += 1;
  myArray[i+2] += 1;
  myArray[i+3] += 1;
}

Done. It's vectorized. Non-AltiVec systems will process both of those pieces of code in approximately the same amount of time, while an AltiVec-enabled processor will process the second bit MUCH faster than the first. Of course, not all examples are that simple, but it's not a headache to make something take advantage of AltiVec.