[NordicMan]

I read about this 'Cell' chip that IBM, Sony and Toshiba are developing, and on another site there is some question about Apple using this. It is to be used on a workstation by Sony, apparently, for game creation for its next Playstation. The chip will also power the next PlayStation. It is built using equipment capable of creating circuits as small as 90nm, not 65nm, as was earlier said.

Is not Apple on a course toward the Power5 cpu, following after the current Power4, used as the G5?

One also reads about rumours of Apple and OS X being on Intel. That seems unlikely, although there are these reports, said to be from credible sources, about Intel cpus in Apple's 'skunk works' basement running OS X.

The understanding that I have had from Apple's announcements, is that they are following the IBM Power server chip.

[Leonard]

Apple is on course to use a PowerPC variant of the Power 5. It's unlikely Apple would use the Power 5 as it is more of a high-end server chip and rather expensive. Compared to today's PowerPC 970FX which Apple uses the newer PowerPC 9xx based on the Power 5 chip may have dual cores, larger L2 cache, and possibly HT. Meanwhile IBM has started designing and prototyping the Power 6 chip which will be the basis of the next PowerPC 9xx chip.

I think it's unlikely Apple will go to the Cell chip. It is also unlikely that Apple would go to Intel, the version of MacOS X that Apple has/had in their R&D department is only for contingency's sake. It won't see the light of day, as long as the PowerPC has a bright future.

[NordicMan]

Yes, that is the way that I understand things, like you said, a variant of the Power 5.

Still, this subject of the Cell chip is very interesting.

[power142]

A lot of IBM's server business relies on Power and PowerPC technology existing and being competitive with the alternatives. Since IBM's just dumped their commodity items, their server business (ie business machines relying on Power) becomes as important as it ever was.

The Cell does look interesting - I'm sure we'll see in the coming years just how it will be implemented outside of PS3.

[Big Mac]

Many people like to fantasize about Apple moving to Intel, but if that were ever to happen the only thing that would be assured would be the death of the company. There are multiple reasons why such an outcome would be inevitable. 1) Apple needs a PPC ISA microprocessor to power its computers, since it lacks the resources to successfully transition to a foreign architecture like it did with 68k -> PPC. 2) A switch to Intel would mean running all your Mac software in emulation mode, and we can all imagine how enthusiastic developers would be when faced with the prospect of retooling their Mac applications. 3) Developers would increasingly expect Mac users to boot into Windows to use their programs, instead of creating quality Mac versions. 4) One of the key differentiations between the Mac and the PC would vanish, the Mac's unique processor. Ergo, the fate of Apple Computer is tied inextricably to the fate of the PPC.

[NordicMan]

The strongest part of your argument, I think, is that of the software developers having to change all of their programmes, if Apple made any change from PowerPC/Power chips.

The other strong argument is on behalf of the apparently good relationship with IBM, and 'Big Blue's' ability to come up with and produce viable chips.


One looks forward to the Mac OS taking advantage of 64 bit, and what that will bring.

[Catfish_Man]

An alternative that I haven't seen suggested much, if at all, is porting OSX to IA-64 (Intel's newer architecture used in their high end Itanium processors). As I understand it, most of the complaints about the performance of these chips have been silenced with the Itanium 2, leaving software compatibility and price as the remaining reasons that it will inevitably fail (some sarcasm here, "Itanic" failing seems a bit like Apple going out of business. Continually claimed, but with the reasons switching periodically). Since x86 and IA64 would be the same as far as compatibility (i.e. neither one is compatible), that leaves price. So, in out hypothetical situation, Intel releases a low cost IA64 chip, perhaps oriented towards blade servers or other high density applications.

Advantages of Apple switching to such a chip:
1) backed by a company with much larger chip design resources than IBM, that has shown an extremely strong commitment to this architecture (putting a large design team on it for something like 10 years, with it only showing results now)
2) the chip would be being used outside of Apple in significant ways, unlike the G5. This helps assure that it has a future
3) very strong performance, especially in floating point (POWER5 based things could be a big improvement over the G5, but much of the POWER5's amazing performance comes from its expensive and elaborate memory system, which wouldn't be in a hypothetical G6)
4) relatively easy chipset design. Itanium is designed to work with 3rd party chipsets, where the POWERx arch is IBM internal.
5) no worries about being too compatible with PCs anytime in the near future (odd worry to have, but I can see it)
6) shares an architecture being pushed by Intel, HP, IBM, SGI, and others, which could bring software compatibility and compiler improvements

Disadvantages:
1) Biggest one: Incompatible with existing Mac software. (Intel has an x86 emulator with fairly reasonable performance. Possibly a sizable application of $$$ could persuade them to do a PPC one? It'd help pull customers away from PPC as well, which Intel would like)
2) Intel in the past has indicated that they would eventually like to switch PCs over to using this arch at which point compatibility concerns come back
3) IBM may pull out of their current manufacturing mess and start delivering really superior chips, in which case the whole switch would have been in vain
4) Apple wouldn't be able to pressure Intel into designing things that suit them better, since they're a relatively tiny customer

Comments, criticism, corrections?

[Madrag]

I agrre with having a bigger supplier than IBM, because I'm terrified to think if IBM goes down the drain, so may the processors...

Is this possible?

[Leonard]

IBM, big blue, go down the drain? Hahahaha

Not likely. IBM has too many corporate customers worldwide. IBM is just as big, if not bigger than Intel and AMD, the only thing is that Intel and AMD are more concentrated on the semi-conductor business wheras IBM is more diversified. I think there's a better chance of AMD or Intel going down the drain if one fell too far behind the other, which is unlikely.

[power142]

Yeah, I wouldn't be too eager to count out IBM because they are small fry, as they really are not. They may have had troubles with the chip fab recently, but that's a totally different operation (as far as IBM is concerned) from Power and derivative design.

Also, Power5 gives competitive performance with Itanium2, so there's no reason to believe that the next Power->PowerPC birth won't keep the G5/6 competitive with other desktop offerings.

There used to be several attractive 64-bit processors on the market, but in recent years these have really narrowed down to Power, IA64 and AMD64. Alpha was killed off, and MIPS and SPARC seem to linger for legacy customers, but don't have a bright future.

[P]

First up: porting OS X to anything is rather easy. This is not the old Mac OS that had been developed for one singe machine with lots of assembler - this is a modern UNIX-based OS that is compiled without assembler. It even runs on x86 today - witness Darwin x86 - though Apple hasn't compiled the Aqua parts. That would be the easy part. Rumors have it that OS X has been run on Ultrasparc and SGI machines to test the compability. What is missing is drivers for everything you might want to put in your system.

The main reason Itanium isn't going anywhere is that it never caught any steam and started growing. It's still at more or less the same volume - missing the estimated sales targets by 99% or so (estimated by Gartner, not by Intel). The Reg had a pretty cool plot of how the sales estimate has been revised downwards over the years. Note that the Reg is very negative towards the Itanium, but the graph was back up with data. I'll see if I can find it again.

The Itanium is based on the idea that the new instruction set style - EPIC - will bring massive parallellization and great speeds for free. It hasn't really, though the idea of integrating the compiler and CPU (which is essentially what they've done) is a good one.

I think that the Itanium is dying simply because Intel is getting fed up with it. Intel thought that they could force people into using Itanium by never making a 64-bit version of the x86 instruction set, but then AMD gave them Opteron with just that and any chance of forcing people over to Itanium went out the window. Intel had to respond, with the 64-bit Xeons, and Itanium is getting squeezed fro the low-end in a way it can't survive. Intel's latest comments about moving the chip into mainframes is a sign of this - it used to be that it should move into lower-end servers. More or less the same thing happened with RAMBUS - Intel tried to force people there by sabotaging existing SDRAM chipsets (the fastest PIII chipset is an overclocked 440BX. There were several chipsets after that, but all were slower that that) and delaying DDR to make RAMBUS look good. It failed when VIA and others made better chipsets and Intel had to respond by improving their own, effectively kiling RAMBUS on the PC.

Intel doesn't need Itanium to survive. The company that really bet the farm on Itanium is HP, but Intel can survive fine with its Pentium line alone. IBM is in only because it wants to be everywhere, and Sun does pretty much the same. SGI has bigger problems, but they could live fine on x86.

There are many reasons why PPC is a better platform for Apple. IBM and Moto^H^H^H^HFreescale both make the chips, and more importantly can make the chips - as can Apple itself, they have access to the patents. With Cell using the same ISA, the market for PPC chips is significantly larger than the Itanium market. Itanium is only high-end servers and possibly mainframes, while POWER and PPC go everywhere: Servers, mainframes & clusters via workstations & personal computers to game consoles - not to mention the huge embedded market. IBM is working hard to get PPC and POWER architechtures everywhere - why walk away from this?

The future is also looking good for 970 and its successors. The large instruction window of the POWER CPUs and the highly threaded OS X mean that SMT (what Intel calls Hyperthreading) will mean a large performance boost. Also on the roadmap is an integrated memory controller - it is already in the POWER5 and the Opteron, and has provided a huge boost to those chips - and dual cores. Beyond that, we don't know much - other than the fact that not only is POWER6 in development, but POWER7 too. That is a long term strategy, and that is exactly what Apple needs.

[power142]

Originally posted by P:
I think that the Itanium is dying simply because Intel is getting fed up with it. Intel thought that they could force people into using Itanium by never making a 64-bit version
of the x86 instruction set, but then AMD gave them Opteron with just that and any chance of forcing people over to Itanium went out the window. Intel had to respond, with the 64-

I think you hit the nail on the head, there P. Intel believed they could make Itanium ubiquitous by it's infamous volume producing trick or "commoditization", only it backfired on them because after the introduction of Opteron, people looking at moving to Itanium and the associated headaches started thinking twice.

Intel doesn't need Itanium to survive. The company that really bet the farm on Itanium is HP, but Intel can survive fine with its Pentium line alone. IBM is in only because it wants to be everywhere, and Sun does pretty much the same. SGI has bigger problems, but they could live fine on x86.

Indeed... HP and Sun have made some strange moves over the last couple of years. Not so long ago, HP sold Itanium2 workstations, but dropped them earlier this year and now only sell P4/Xeon, PA-RISC and Alpha-based workstations. They also make servers based around not only Xeon and Itanium2 chips (in addition to their RISC), but Opteron too. IBM and Sun both produce workstations (as well as servers) based on their own RISC technologies as well as P4/Xeon/Opteron.
SGI is unique... although their niche continues to decrease, they have probably sold more Itanium2 chips for Intel than anyone else. The Altix is a fine system if you have the money, but I don't doubt they could switch to x86 (or, less likely, Power) if Intel decided (also unlikely) to dump the Itanium.

[Catfish_Man]

Overall, I agree with your post. The only things I wonder about/disagree with are below:

Originally posted by P:
First up: porting OS X to anything is rather easy. This is not the old Mac OS that had been developed for one singe machine with lots of assembler.

This is true to a certain extent, but if you run Shark on just about any program and look for Altivec instructions (which are PPC only), you'll see that much of OSX's performance is ISA dependent. Supposedly much of ObjC4 (the objective-c runtime Apple uses) is also PPC asm, but I'm not sure on that.

Originally posted by P:
The future is also looking good for 970 and its successors.

I'm not so sure on this one. I think that *if* IBM finds more customers for it, then yes, it should do fine. If they don't... I'm not sure Apple alone has the volume needed to make it profitable to develop. If IBM either pushes the 970 harder itself, or works to create a 3rd party market using it (i.e. more docs, easier mobo design, standard northbridge(s)), then I can see it getting the volume to survive. I agree PPC is one of 3 archs that has a chance though.

[power142]

All told, it's not as if they're designed the PPC970 from the ground up - they started off with the Power4+ and then chopped off the cache and bolted on Altivec. I know this is a much simplified assessment, but IBM has stated before that they rely on automated tools for their transistor layout during the chip design.

It is a good question asking what volume IBM would need to continue development, but we shouldn't forget that IBM developed and is still speed bumping the PPC750 aka G3. Who is buying them now? I'm sure Apple must be buying a trickle of them for G3 repairs/replacements, but not >1GHz chips. It wouldn't be unreasonable to ponder who is buying those chips for them to make them worthwhile.

[macaddict0001]

I suspect that g3's may be going in embedded chipsets but it is nearly impossible to know.

[Catfish_Man]

Originally posted by power142:
All told, it's not as if they're designed the PPC970 from the ground up - they started off with the Power4+ and then chopped off the cache and bolted on Altivec. I know this is a much simplified assessment, but IBM has stated before that they rely on automated tools for their transistor layout during the chip design.

It is a good question asking what volume IBM would need to continue development, but we shouldn't forget that IBM developed and is still speed bumping the PPC750 aka G3. Who is buying them now? I'm sure Apple must be buying a trickle of them for G3 repairs/replacements, but not >1GHz chips. It wouldn't be unreasonable to ponder who is buying those chips for them to make them worthwhile.

Yeah, it'll be interesting to see how closely the 9xx series tracks the POWERx series. My take on it is that the POWER5 focuses on system-level rather than chip-level improvements, and that the 9xx series will diverge from it over time, but I could be underestimating how much of the POWER5's improvement is from architecture improvements, rather than its huge cache and memory improvements (4x bandwidth, 33% latency reduction).

[tooki]

Originally posted by Catfish_Man:
This is true to a certain extent, but if you run Shark on just about any program and look for Altivec instructions (which are PPC only), you'll see that much of OSX's performance is ISA dependent. Supposedly much of ObjC4 (the objective-c runtime Apple uses) is also PPC asm, but I'm not sure on that.

Considering that the Objective-C that Mac OS X uses wasn't born on PPC at all, but rather on 68K, and later on Intel, what you say is exceedingly unlikely.

Originally posted by Big Mac:
Many people like to fantasize about Apple moving to Intel, but if that were ever to happen the only thing that would be assured would be the death of the company. There are multiple reasons why such an outcome would be inevitable.

1) Apple needs a PPC ISA microprocessor to power its computers, since it lacks the resources to successfully transition to a foreign architecture like it did with 68k -> PPC.

2) A switch to Intel would mean running all your Mac software in emulation mode, and we can all imagine how enthusiastic developers would be when faced with the prospect of retooling their Mac applications.

3) Developers would increasingly expect Mac users to boot into Windows to use their programs, instead of creating quality Mac versions.

4) One of the key differentiations between the Mac and the PC would vanish, the Mac's unique processor. Ergo, the fate of Apple Computer is tied inextricably to the fate of the PPC.

1) What the f***? Apple's in much better financial shape than they were in the early 90's, when Apple not only successfully transitioned from 68K to PPC, but also ported System 7 from 68K to i486 (the "Star Trek" project). On top of this, the Mac OS X code is vastly more portable than the old System 7 code was. Moving Mac OS X to Intel is a fairly trivial thing, considering that OpenStep (Mac OS X's direct ancestor) ran on Intel.

2) Emulators suck, but recompiling applications to support PPC and Intel is trivial (again, that already existed in OpenStep with "fat" 68K/Intel application bundles). In fact, early developer builds of Mac OS X could make fat Intel/PPC applications. All it took was checking a single checkbox in Project Builder.

3) If Apple ran on industry-standard Intel hardware, perhaps. But Apple could also choose to switch to Intel processors without making the hardware even remotely Windows-compatible.

4) Eh, Apple made one CPU switch already, and did it masterfully. If they needed to, they could do it again. As it is, the PPC architecture is strong, so there's not much need to switch. But you'd be a fool to think that Apple is "inextricably" tied to PPC. As I said in point 1, Apple's in much stronger shape to make a transition now than they were the last time they did it!

tooki

[P]

Originally posted by power142:

Indeed... HP and Sun have made some strange moves over the last couple of years. Not so long ago, HP sold Itanium2 workstations, but dropped them earlier this year and now only sell P4/Xeon, PA-RISC and Alpha-based workstations. They also make servers based around not only Xeon and Itanium2 chips (in addition to their RISC), but Opteron too. IBM and Sun both produce workstations (as well as servers) based on their own RISC technologies as well as P4/Xeon/Opteron.
SGI is unique... although their niche continues to decrease, they have probably sold more Itanium2 chips for Intel than anyone else. The Altix is a fine system if you have the money, but I don't doubt they could switch to x86 (or, less likely, Power) if Intel decided (also unlikely) to dump the Itanium.

HP probably sold more Itaniums than anyone, but they're pitching it as the successor their main architechture, PA-RISC. Everyone else is pitching it as an also-ran. I agree about HPs strange moves, btw - I understand absolutely nothing they're doing these days. I'm glad I don't own stock in that company.

Sun is in trouble, CPU-wise. They can't really keep developing the Ultrasparcs themselves, as witnessed by their recent cancelling of an already taped-out CPU. They need something, and used Itanium as a "just in case it's a hit"-move. What they're going to do now? Solaris is great, but the CPU... Opteron? Best bet right now, I guess, but it's a dangerous road. Really think they'd be better off with the 970, but there are reasons against that too.

SGI has an even bigger problem - they also can't keep developing their CPU, but they don't have a good OS. IRIX is not a good OS - it's a barely working one. They have a niche, but they don't have a future. I really don't know what they should do. On the plus side, they know that they have a problem.

IBM is coming back. Power4 was the make-or-break CPU, and it didn't break. They can keep riding that CPU for some time yet, Power5 is looking extremely good and there are lots of nice stuff on the roadmap.

What is happening is that a lot of the server companies who used to develop ther own chips can't keep doing that. They have to bet on something, and there are risks with everything. With Power, or rather the 970, they're buying a CPU from someone they're also competing with in that segment. With Opteron or Xeon, they risk degenrating into just another PC supplier. Itanium was looking like the safe choice, but it's looking grimmer everyday since the chip isn't taking off. Intel saw this years ago, which is why developing Itanium was looking like such a great idea (no, I didn't see it then, but today I can understand their thinking. 20-20 hindsight here) but for some reason it doesn't take off.

Perhaps the x86 compability was the problem. Slow x86 emulation became news, and the first thing everyone heard about the CPU. If they hadn't bothered with that, they could have made a great new CPU and started off with migrating the PA-RISC users to it. Then, SGI and Sun could use it while Intel made the first 64-bit compatible x86 CPU, taking care to design the extended ISA to make it easy to work it into Itanium. Finally, when there was a stable Itanium base Intel could add in x86 support and start killing off the Pentium line - slowly, but kill it.

Wild speculation, but what do you think? Would that have been better?

[P]

Originally posted by macaddict0001:
I suspect that g3's may be going in embedded chipsets but it is nearly impossible to know.

The G3 began its life as the PPC613, the successor to PPC603. I know the 603 was a widely used embeddded CPU, and the G3 was the natural successor. It was used allright, at least the 740 model, although I don't know exactly how widely.

[P]

Originally posted by Catfish_Man:
Overall, I agree with your post. The only things I wonder about/disagree with are below:

<porting OS X is easy>

This is true to a certain extent, but if you run Shark on just about any program and look for Altivec instructions (which are PPC only), you'll see that much of OSX's performance is ISA dependent. Supposedly much of ObjC4 (the objective-c runtime Apple uses) is also PPC asm, but I'm not sure on that.

Oh, it may be quite some work to optimize it, but then it always is. Apple kept optimizing the old Mac OS after the transition from m68k, and I'm not speaking only of native code versus emulated. I'm just saying that it can be done without all that much work.

Originally posted by Catfish_Man:

<the future is looking good>


I'm not so sure on this one. I think that *if* IBM finds more customers for it, then yes, it should do fine. If they don't... I'm not sure Apple alone has the volume needed to make it profitable to develop. If IBM either pushes the 970 harder itself, or works to create a 3rd party market using it (i.e. more docs, easier mobo design, standard northbridge(s)), then I can see it getting the volume to survive. I agree PPC is one of 3 archs that has a chance though.

Right now, Apple pays a lot less for its G5s than it does for its G4s. I doubt that IBM is selling them at a loss, and the chips are bigger and more expensive to make, so the have to be recouping that development cost somehow. either they're selling a lot, or someone else is paying the bill. My guess? Microsoft. Xbox 2 has to be a hit - I think they're paying IBM a lot of development money.

Intel's in trouble now that they can't ride the MHz pony any longer. They're reacting as fast as they can, but CPU roadmaps are made years in advance. I have no doubt that they're throwing huge resources at it, but still... IBM is not small, and they have a head start right now. AMD is smaller, but they have an awesome CPU that's ready for the future. Intel's another company I'm glad I don't own stock in. Their main product line is in trouble - even Dell seems to be getting AMD chips in some machines now, the Itanium line is being put out to pasture, the competition in the chipset market is bigger than ever and has anyone heard about the networking stuff recently? They'll bounce back, sure, but they haven't seen the bottom yet.

With Intel sturgling, the traditional server ISAs dying and Motorola...let's not even go there, IBM has the chance of a lifetime. Time will tell if they grab it.

[Catfish_Man]

Originally posted by P:
My guess? Microsoft. Xbox 2 has to be a hit - I think they're paying IBM a lot of development money. [/B]

Yeah, that's definitely looking interesting. I recall IBM claiming that they were going to get something like Intel's volume advantage by taking over the gaming market, and if the XBox2 really is heavily 9xx based (in a way Apple can use, not some weird specialized gaming variant), then that explains quite a lot. Given that, then yes Apple would be piggybacking off a market with different requirements again (like the embedded market with the G4), but that would ensure some sort of future as well as a large supply of development dollars. There's also CELL, but I can't for the life of me figure out which company's plans that fits into and how, aside from the PS3. Another thing to watch This should be a fun few years.

tooki: about the ObjC runtime: yeah, I agree that sounds quite odd that it would be heavily PPC asm. I was vaguely recalling some posts on the GCC mailing list, I'll see if I can dig them up at some point.

[OreoCookie]

Originally posted by Big Mac:
...

1) Apple needs a PPC ISA microprocessor to power its computers, since it lacks the resources to successfully transition to a foreign architecture like it did with 68k -> PPC.

2) A switch to Intel would mean running all your Mac software in emulation mode, and we can all imagine how enthusiastic developers would be when faced with the prospect of retooling their Mac applications.

...

Apple does keep MacOS X for Intel running, so they could switch to another architecture (not necessarily x86). If it runs on two essentially different architectures, the code should be easily portable to another architecture.

[OreoCookie]

Some remarks are in order.

1. The Power5's advantage is not that it suddenly opens the door for multicore designs, it was introduced with the Power4 (the G5's big brother).

2. I don't think Apple will use the big Power CPUs any time soon (same goes for Itanium) as the chip alone costs as much as a PowerMac (or significantly more). Take a look at the price tags for a Power5 MCM (multi chip module) ... The same goes for Itanium: we will never see a switch to IA64, Intel itself has stated that the niche for the Itanium from now on will be big iron servers and main frames. Without cheap chips, it will never come to the desktop (HP, co-developer of the Itanium, has killed its Itanium workstations).

3. IBM will push for a more diverse PowerPC architecture, i. e. it will continue to develop cheaper Power derivatives for other uses (such as blades and smaller devices). IBM has apparently another low-power PowerPC chip under development, the PPC 300 series which is to replace the venerable PPC 400 series (that design has already been sold, I believe). It has been rather silent (i. e. no more rumors) about this line of chips, but the first results were supposed to be due in 2005.

I think these might be interesting for Apple.

4. Dual-core chips are en vogue right now, and Apple is lucky that IBM is basically the chip designer with the most experience in that field. My guesstimate is that IBM will do it like AMD: lower the clock speed for a dual core CPU so that the power envelope will stay the same as a higher clocked single core CPU.

5. I have no idea if Apple will use the cell processor, too little is known. But the basic design suggests that it probably won't be used (the PowerPC chip just acts as a `controller' for the SIMD units). It seems to be more of a vector cpu (and that speeds up just certain types of applications, such as encoding, rendering, scientific calculations, etc.), but it will rely on the (smaller) PowerPC core for normal OS stuff.

6. Maybe, some time in the future, we will see massively parallel systems with smaller, simpler CPUs, à la Cell or PPC440 (in the fastest supercomputer), but this means that applications and MacOS X that take advantage of this architecture need to be in place.

[OreoCookie]

Originally posted by macaddict0001:
I suspect that g3's may be going in embedded chipsets but it is nearly impossible to know.

No, it's not.
I have a G3-derivative in my Kyocera printer (PPC740). Oki uses them, too, and so do many other printer manufacturers. It is already used as an embedded CPU.

[OreoCookie]

Originally posted by P:
...

Sun is in trouble, CPU-wise. They can't really keep developing the Ultrasparcs themselves, as witnessed by their recent cancelling of an already taped-out CPU. They need something, and used Itanium as a "just in case it's a hit"-move. What they're going to do now? Solaris is great, but the CPU... Opteron? Best bet right now, I guess, but it's a dangerous road. Really think they'd be better off with the 970, but there are reasons against that too.

...

Sun basically asked Fujitsu-Siemens for help as they design their own SPARC CPUs that can keep up with the Itanium and Power CPUs performance-wise.

They will co-develop the next-generation SPARC CPUs.

[power142]

Originally posted by P:
Perhaps the x86 compability was the problem. Slow x86 emulation became news, and the first thing everyone heard about the CPU. If they hadn't bothered with that, they could have made a great new CPU and started off with migrating the PA-RISC users to it. Then, SGI and Sun could use it while Intel made the first 64-bit compatible x86 CPU, taking care to design the extended ISA to make it easy to work it into Itanium. Finally, when there was a stable Itanium base Intel could add in x86 support and start killing off the Pentium line - slowly, but kill it.

Wild speculation, but what do you think? Would that have been better?

I agree with your assessment

I'm sure that originally Intel hoped to woo the tech types with all the EPIC stuff, but then people in the field saw the hopeless performance that Itanium1 gave. Itanium2 was _much_ better, but the damage was already done.

The same goes for it's future... other things I've read indicate that Intel will make motherboards compatible with both Xeon and Itanium at some stage, and gear Itanium at the high performance, and Xeon at the low-end. I don't know for sure that this is true, though. Maybe the chickens have already fled the coup, after the wolf came around the first time?

[P]

Originally posted by Catfish_Man:
Yeah, that's definitely looking interesting. I recall IBM claiming that they were going to get something like Intel's volume advantage by taking over the gaming market, and if the XBox2 really is heavily 9xx based (in a way Apple can use, not some weird specialized gaming variant), then that explains quite a lot. Given that, then yes Apple would be piggybacking off a market with different requirements again (like the embedded market with the G4), but that would ensure some sort of future as well as a large supply of development dollars. There's also CELL, but I can't for the life of me figure out which company's plans that fits into and how, aside from the PS3. Another thing to watch This should be a fun few years.

tooki: about the ObjC runtime: yeah, I agree that sounds quite odd that it would be heavily PPC asm. I was vaguely recalling some posts on the GCC mailing list, I'll see if I can dig them up at some point.

The Xbox 2 requires something very similar to the 970: It should be a CPU that is mainly linear and similar to x86, for several reasons:

* The old games must work
* PC games should be easy to emulate
* The required parts of Windows should be easy to port or emulate
* It should be easy to develop for it (it's the only advantage the Xbox has over the PS2)

Also, the system needs a good internal bandwidth to compete with the PS3 - another upside for the 970.

Cell and the PS2 are both about parallellization, which may be the way of the future but won't work well for emulating the very serial Pentium III used in the old Xbox.

I think that the actual CPU they'll be using will be a future version of the 970, with an onboard DDR memory controller and maybe something else to make it unique. And yes, I agree that this is going to be fun for several years...

[P]

Originally posted by power142:
I agree with your assessment

I'm sure that originally Intel hoped to woo the tech types with all the EPIC stuff, but then people in the field saw the hopeless performance that Itanium1 gave. Itanium2 was _much_ better, but the damage was already done.

The same goes for it's future... other things I've read indicate that Intel will make motherboards compatible with both Xeon and Itanium at some stage, and gear Itanium at the high performance, and Xeon at the low-end. I don't know for sure that this is true, though. Maybe the chickens have already fled the coup, after the wolf came around the first time?

Too little too late. Sure, it can be a big iron kind of chip, but it has competition there. The volume is in the low-end servers, and it's not gaining traction there. I just can't see how it can win at this point.

[Simon]

Originally posted by P:
The Xbox 2 requires something very similar to the 970: It should be a CPU that is mainly linear and similar to x86, for several reasons:

* The old games must work
* PC games should be easy to emulate
* The required parts of Windows should be easy to port or emulate

Do you think this could eventually lead to improvements to VPC? I mean, if MS is now porting components of Windows and other x86 code to a 970-like CPU, could this benenfit Windows emulation on the G5?

[OreoCookie]

Originally posted by P:
The Xbox 2 requires something very similar to the 970: It should be a CPU that is mainly linear and similar to x86, for several reasons:

* The old games must work
* PC games should be easy to emulate
* The required parts of Windows should be easy to port or emulate
* It should be easy to develop for it (it's the only advantage the Xbox has over the PS2)

Also, the system needs a good internal bandwidth to compete with the PS3 - another upside for the 970.

Cell and the PS2 are both about parallellization, which may be the way of the future but won't work well for emulating the very serial Pentium III used in the old Xbox.

I think that the actual CPU they'll be using will be a future version of the 970, with an onboard DDR memory controller and maybe something else to make it unique. And yes, I agree that this is going to be fun for several years...

AFAIK there is no confirmation that old XBox1 games will work on the XBox2. It has been a persistent rumor and it helped the PS2 a lot (you could use all of your old games!). But on the other hand, the thing is damn fast and so it should be doable ...

Windows is not MacOS X (aka Darwin). It is not so easy to port. There is a lot of assembler under the hood since the versions for alternative platforms got canned. Maybe it's getting better (Itanium), but just take a look at how long it took to port Linux/FreeBSD to AMD64 compared to Windows ...

The PS3 will probably use the same chip, so I think the internal architecture will be similar, ditto for the GameCube NG.

[P]

If the Xbox successor is truly called Xbox 2, or anything with Xbox in the name, then I think it will play older games. Since they're written mainly for DirectX, MS would only have to port DirectX to whatever architechture they're running it on and hack up some new drivers and it should work. Personally, I think it would be stupid not to make it compatible. If MS were the leader in the field, they wouldn't do that, but as it is they're third in world-wide sales and losing a lot of money. They can't risk alienate any customers.

On porting: Porting all of Windows, including the Win16 compability stuff, would be very hard. Porting the NT kernel (used in NT, Win 2k and XP) and the newer APIs, or a subset of them (DirectX, mainly) is doable. NT 4, which includes the kernel and the Win32 APIs, ran on CHRP PPCs, at least in beta. NT 3.5.1, which admittedly didn't include the Win32 APIs, did run on PPC and was extremely stable. I think that MS can use put the kernel on the G5 and port DirectX in a fairly short time. The would need VPC or something similar to run the older games plus whatever part of Windows they need and can't port, but doesn't have to be much.

This _could_ benefit VPC, if MS ports DirectX to PPC. If they can run that part of it natively on the G5 of a Mac, the part needing emulation would drop markedly. How much assistance they would need form Apple I don't know, but the main question is this: Why would MS do this? They're not exactly being pushed out of the Windows emulation market.