Power5 at Microprocessor Forum 2003

[MacRumors]

The 2003 Microprocessor Forum takes place this year between October 13th - October 16th in San Jose, California.

Whether your interest is in server processors, PC processors, networking processors, high-performance embedded, or innovative "extreme" processors, MPF 2003 will have information you need to have.

Last year's Microprocessor Forum brought the first details (http://www.macrumors.com/pages/2002/08/20020808110527.shtml) of IBM's New 64-bit PowerPC -- which turned out to be the PowerPC 970 (G5).

This year (http://www.mdronline.com/mpf/conf.html#1) the Power5 processor will be presented by Dr. Balaram Sinharoy, POWER5 Chief Scientist of IBM. As a replacement for the Power4, it is unlikely that the Power5 will find its way into Apple-built computers, however, rumors have hinted that Apple may benefit from Power5-derived technology in the future.

Early reports indicated that the Power5 is already in testing and performs very favorably (http://www.macrumors.com/pages/2003/02/20030218021959.shtml) when compared to the Power4. The Power5 is reportedly due in 2004.

[Rustus Maximus]

originally posted by Macrumors
...it is unlikely that the Power5 will find it's way into Apple-built computers...

Now do you mean it's unlikely that the Power5 itself will find it's way into next generation Macs or that it's unlikely that next generation processors will even be based on the Power5 at all? I would think it very likely that the 980 (or whatever it'll be called) will be based on the Power5.

Rustus

[shadowself]

The POWER (Performance Optimized with Enhanced RISC) series is definitely moving forward. They definitely have POWER5 systems in test. The POWER5 is as much for scientific computing (it will be used in the IBM ASCI Purple machine for the DOE) as for banks and large enterprise servers.

However, I am a bit surprised they are presenting on the POWER5 as they usually present on things which are in design and will be shipping within the next 6 months to a year. If I remember correctly, IBM likes to put out teaser presentations at this conference. Since the POWER5 CPU will have been in test for well over 4 months by the time of the presentation maybe they are expecting significant performace gainse over the POWER4 CPU and plan on announcing the results from these test machines then.

Also there have been lots of rumors about concurrent development of a PowerPC being derived from the POWER5. Maybe there will be some semi-official statement made then too. It would be nice to hear.

Additionally, when they do create the PowerPC derivative of the POWER5, I hope they do an updated version of the vector processor and its associated on chip registers, etc. The vector processor in the PowerPC 970 -- while a screamer compared to competing chips -- was just sort of "tacked on" and it is not as advanced in architecture as some of the G4s have. Hopefully, with more time and foresight they will do a better one when they do the POWER5 derivative.

[arn]

Originally posted by Rustus Maximus
Now do you mean it's unlikely that the Power5 itself will find it's way into next generation Macs or that it's unlikely that next generation processors will even be based on the Power5 at all? I would think it very likely that the 980 (or whatever it'll be called) will be based on the Power5.

Rustus

"As a replacement for the Power4, it is unlikely that the Power5 will find it's way into Apple-built computers, however, rumors have hinted that Apple may benefit from Power5-derived technology in the future"

[wizard]

I have to wonder if AltVec is being incorporated into Power5. It seems like that would be the easest path to follow to also have Altvec in the desktop chip Apple would use. This would also offer Apple the oportunity to offer a wide range of servers of differrent class hardware.

As far as the 4X improvement I also wonder if Apple will be able to get that out of the single chip version. It would be a stunning accomplishment especuially if that 4X is based on clock to clock comparisons.

All in all things haven't looked this good in Mac land in sometime. Well as long as you don't look at the 7457 issue.

thanks
Dave

[idea_hamster]

Do we think that the Power5 could be the basis for a broader step up for Apple's XServe line?

If Apple's current technology is based on the Power4 and the Power5 may be suitable for enterprise server use, is it too simplistic to think that it would be a useful starting point if Apple wanted to give medium-large sized businesses that are too large for XServe an all-Apple alternative?

<swims back to shallow end>

[AidenShaw]

Originally posted by idea_hamster
Do we think that the Power5 could be the basis for a broader step up for Apple's XServe line?

No. Even the POWER4 would be a troublesome marketing message.

Apple has committed far too much intellectual capital to AltiVec to embrace a non-AltiVec high end processor. Even for a server, Apple is boasting about BLAST and other rendering apps that require AltiVec.

Even if the POWER5 does get AltiVec, its timing is too far out for the Xserve - something must be done sooner.

An update of a 1U Xserver to PPC970 would be good. It would also be a great idea to come out with a larger server, say a 3U or 4U quad-PPC970 system for those medium-sized businesses. Give it more slots, more memory, more CPUs - AND MORE REDUNDANCY (dual power supplies, ECC or RAID memory, hot-swap disks, hot-swap RAM, hot-swap everything).

[Freg3000]

Originally posted by wizard
I have to wonder if AltVec is being incorporated into Power5. It seems like that would be the easest path to follow to also have Altvec in the desktop chip Apple would use.

I don't think the Power4 has Altivec but its derivative, the 970, does. So just because Altivec is not on the Power5 does not mean it won't be on the derivative chip (980?).

[Rustus Maximus]

Originally posted by arn
"As a replacement for the Power4, it is unlikely that the Power5 will find it's way into Apple-built computers, however, rumors have hinted that Apple may benefit from Power5-derived technology in the future"

Nooowwwww I get it...

I get it...

I get things... ;)

(hey it was early when I was reading this :) )

Thanks arn,

Rustus

[nuckinfutz]

That Apple and IBM have concurrent development of the POWER5 and a derivative(PPC 980?) going.

This way the P5 hits and within 6 months the Derivative ships as well. The 970 will probably top out at 3Ghz so Apple will need this Derivative to take over from there on.

I'm hoping for ondie memory controllers and SMT. Beefed up Altivec would be nice.

[idea_hamster]

Originally posted by AidenShaw
Give it more slots, more memory, more CPUs - AND MORE REDUNDANCY (dual power supplies, ECC or RAID memory, hot-swap disks, hot-swap RAM, hot-swap everything).
Wow -- I didn't realize that the XServe is missing so many features that even I knew to be available on servers in general (e.g., ECC).

<concededly OT>
So who is the target market for XServe? -- that is, who wants a server but doesn't need this kind of redundancy or hot maintenance?
</concededly OT>

[visor]

Originally posted by idea_hamster
Wow -- I didn't realize that the XServe is missing so many features that even I knew to be available on servers in general (e.g., ECC).

<concededly OT>
So who is the target market for XServe? -- that is, who wants a server but doesn't need this kind of redundancy or hot maintenance?
</concededly OT>

Well, 'blade' servers might be a good target, as well as low end small business apple based businesses.
Or imagine you have any apple style rendering to do big time - that belongs in a cooled environment with many sibling nodes.

Its ok for this kind of setup.

Personally I like clustered solutions better than big bad maschines anyway. Mostly because i don't much trust 'indoor' redundency of a single maschine. However - it seems to work even despite me not much liking it ;)

I know mid sized companies that work with small setups, but completely redundant, so that any one computer can take over the others work, while the other is in maintainance. that's mostly because noone would be willing to 'hot' work in a very narrow room anyways...

If you want to see a great cluster at work - just google ;)

[crenz]

Originally posted by idea_hamster
So who is the target market for XServe? -- that is, who wants a server but doesn't need this kind of redundancy or hot maintenance?

And who in that target market is willing to pay that kind of price (as compared to PC-based Linux/BSD servers, for example)?

[firewood]

So who is the target market for XServe? -- that is, who wants a server but doesn't need this kind of redundancy or hot maintenance?
The Google people gave a keynote talk at HotChips last year. Google apparently makes a very reliable system (responsive with a very high uptime) by using a huge linux cluster of the cheapest unreliable generic PC components. They designed server software that doesn't depend on any one machine or switch to stay running. The data is duplicated over multiple machines in multiple different sites. Much much cheaper than buying a smaller amount of more reliable systems. Apparently the only area where costly redundancy seems to be an issue is with building cooling and the facilities AC backup.

[The Grimace]

It's good to know that the maker of Apples high-end chip is already moving beyond that which spawned said chip. This can only portend good things for Apples' future.

(tig)

[Sherman]

Well I just hope the 970 doesn't have the shortest lifetime of any apple processor. With the Power5 in 2004 a 980 in late 2005/2006.

They'll probably live on in iMac's and iBooks for awhile.

[Docrjm]

Originally posted by Sherman
Well I just hope the 970 doesn't have the shortest lifetime of any apple processor. With the Power5 in 2004 a 980 in late 2005/2006.

They'll probably live on in iMac's and iBooks for awhile.
Don't mind if its short and sweet, so long as there is something sweeter coming afterwards!

[jaedreth]

Keep in mind, IBM has not announced any intent to actually use Power5 on desktop machines like Apples. Also, the G5 will likely have a lifespan similar to G4, but with a lot more frequent upgrades...

Jaedreth

[ddtlm]

jaedreth:

The Power5 itself will never be in desktop machines; its targeted at servers.

wizard:

I don't know where the 4x thing started exactly, but every time I hear it repeated it has become more exaggerated. As far as I know, originally it was an "up to" comment without specifying what was being done or what clockspeeds were involved.

[hasapi]

Originally posted by ddtlm
jaedreth:

The Power5 itself will never be in desktop machines; its targeted at servers.

wizard:

I don't know where the 4x thing started exactly, but every time I hear it repeated it has become more exaggerated. As far as I know, originally it was an "up to" comment without specifying what was being done or what clockspeeds were involved.

The reference for the 4x the performance of the Power 4, came from IBM Engineers themselves of its testing the ASIC Purple machine (sorry dont have the link, eweek or eenews?). The Power 5 is Dual Core with SMT. The 980 however is unlikely to be Dual Core, but I would be delighted with 2x (the performance) of a 970 wouldnt you?

Well just have to wait and see if the Power5 lives up to the claims from IBM - but its looking good for a G6 within 2years from now, perfect upgrade timing me thinks? :)

[jaedreth]

If IBM does decide to use Power5 or a derivative on Macs, I wouldn't expect it for 4 years or more. IBM makes chips for its own servers first, Apple second. But that's better than Motorola making its chips for cell phones first, and if they feel like it they might make some for Apple. The G4 premiered in 1999, and it's reign lasted until this year, 2003, about the same point in the year. So that's 4 years. Given that, I don't expect G6 until Fall of 2007.

Jaedreth

[wizard]

Originally posted by Freg3000
I don't think the Power4 has Altivec but its derivative, the 970, does. So just because Altivec is not on the Power5 does not mean it won't be on the derivative chip (980?).

Yes everyone knows that. What I'm trying to get at is if AltVec is better integrated inot the Power5 processor or will remain forever a tacked on unit. At some point in the development cycle it would be easier to keep Altvec technology in the processor HDL code that to keep tacking it on for the single chip variant.

Further; if Altvex is in the Power5 it does provide Apple and IBM with considerable marketing advantages. Even on the types of software executed on a Power5 there are a class of programs that can benefit form AltVec. It is a symmetry that I hope IBM is not missing.

Dave

[wizard]

Hi Jaedreth;

You making a big assumption here that Apple and its customers are happy with the state of the G4. I'm not to sure to many customers will jump up claiming undying devotion to the G4, the majority of those customers along with Apple would have rathered that it matured a bit in those 4 years. Instead it became the joke of the microprocessor world.

A microprocessor is not going to make it in this world if it is only improved every 4 years. To be specific I don't consider clockrate only changes to be improvements. Frankly Apple and IBM really have to deliever an improved 970 or a 980 by the end of 2004, just to demonstrate that they have a handle on the old Motorola stagnation. That improvement can be a simple as an enlarged L2 cache or built in DDRAM interfaces, or something more comples like an inproved Altvec implementation. The proof is in the pudding so to speak, the Apple - IBM arraingement needs to deliver on a regular basis if Apple expects to reach any of its former customer base.

Thanks
Dave


Originally posted by jaedreth
If IBM does decide to use Power5 or a derivative on Macs, I wouldn't expect it for 4 years or more. IBM makes chips for its own servers first, Apple second. But that's better than Motorola making its chips for cell phones first, and if they feel like it they might make some for Apple. The G4 premiered in 1999, and it's reign lasted until this year, 2003, about the same point in the year. So that's 4 years. Given that, I don't expect G6 until Fall of 2007.

Jaedreth

[Phil Of Mac]

Originally posted by shadowself
However, I am a bit surprised they are presenting on the POWER5 as they usually present on things which are in design and will be shipping within the next 6 months to a year. If I remember correctly, IBM likes to put out teaser presentations at this conference.

Just because it's by the POWER5 guy doesn't mean it'll be a POWER5 presentation. Maybe they'll announce the POWER5-based PowerPC!!!

(Or maybe not.)

[Phil Of Mac]

Originally posted by firewood
The Google people gave a keynote talk at HotChips last year. Google apparently makes a very reliable system (responsive with a very high uptime) by using a huge linux cluster of the cheapest unreliable generic PC components. They designed server software that doesn't depend on any one machine or switch to stay running. The data is duplicated over multiple machines in multiple different sites. Much much cheaper than buying a smaller amount of more reliable systems. Apparently the only area where costly redundancy seems to be an issue is with building cooling and the facilities AC backup.

That is fascinating...It makes sense, too. Buying cheap, unreliable parts, but lots of them redundantly. I guess by mere law of averages that will give you good reliability.

[daveL]

Originally posted by jaedreth
If IBM does decide to use Power5 or a derivative on Macs, I wouldn't expect it for 4 years or more. IBM makes chips for its own servers first, Apple second. But that's better than Motorola making its chips for cell phones first, and if they feel like it they might make some for Apple. The G4 premiered in 1999, and it's reign lasted until this year, 2003, about the same point in the year. So that's 4 years. Given that, I don't expect G6 until Fall of 2007.

Jaedreth
Well, I can't give you a link, but it was stated that, from this point on, PowerX development and PPC9X0 development would proceed in parallel. The 980 should arrive very shortly after the Power5.

The more I think about it, the more certain I am that the info above came out of a news item referenced here at MR.

[jaedreth]

I think you misunderstood me.

1) IBM has not reported that the next processor design, tentatively coded the 980 and/or Power5, would be used in Macs.

2) The reports I've seen seem to indicate they will *not* be used on macs, but technological advances based on this project may eventually be incorporated into Macs.

3) That the G5 would continue to be used, not as is, but as a growing processor architecture, for the next four years.

4) That *if* IBM were to eventually make the Power5 into a chip for Macs, aka a G6, that it would not happen until at least Fall of 2007.

My point is not that the G5 will stagnate. I don't forsee that at all.

I see G5 getting faster and faster. IBM is already working on a second Revision of the 970, that is still G5. And IBM will continue to do so. The chip will get faster and mature more, providing better performance at far more frequent intervals than with Motorola's G4's.

So just because I compare the *life span* of G5 to G4, and expect it to roughly match (my own personal opinion), doesn't mean that I forsee the same bs Motorola put us through happening with IBM. In fact, I'm expecting great things.

But also keep in mind, IBM was using G4 technology in it's current machines four years before Apple had its first G4. IBM makes High End Servers, so the processors they make for themselves are not going to the the processors Apple uses, perhaps except after a good period of time.

I hope that clarifies my position.

Besides, G5 will scale very well over four years, and perhaps beyond. I don't forsee needing a G6 before then, though we've needed G5 for two or three years now.

Jaedreth

[jaedreth]

Well, I can't give you a link, but it was stated that, from this point on, PowerX development and PPC9X0 development would proceed in parallel. The 980 should arrive very shortly after the Power5.

____

I must have missed that one. But what I'm describing is the relationship between Power4 and G5, Power 5 and G6.

If the PPC 970 is a *variant* of the Power4, designed for a workstation instead of a high end server, then I could see this. I think this may be the case, but I'd have to clarify this with my sources.

A PPC 980 *may be* a variant of Power5, but that does not necessarily mean it will be G6.

There is also no guarantee at this early stage that IBM will call this supposed variant of Power5 a PPC 980 anyways.

There *was* a report here about a supposed PPC 980, that more recent reports showed it would likely still be classified as 970, but a new revision.

I do expect technology to increase, and at a far faster rate after Moto is out of the picture, however I still don't forsee Apple releasing a "G6" computer any sooner than late 2007.

Jaedreth

[daveL]

Originally posted by jaedreth
Well, I can't give you a link, but it was stated that, from this point on, PowerX development and PPC9X0 development would proceed in parallel. The 980 should arrive very shortly after the Power5.

____

I must have missed that one. But what I'm describing is the relationship between Power4 and G5, Power 5 and G6.

If the PPC 970 is a *variant* of the Power4, designed for a workstation instead of a high end server, then I could see this. I think this may be the case, but I'd have to clarify this with my sources.

A PPC 980 *may be* a variant of Power5, but that does not necessarily mean it will be G6.

There is also no guarantee at this early stage that IBM will call this supposed variant of Power5 a PPC 980 anyways.

There *was* a report here about a supposed PPC 980, that more recent reports showed it would likely still be classified as 970, but a new revision.

I do expect technology to increase, and at a far faster rate after Moto is out of the picture, however I still don't forsee Apple releasing a "G6" computer any sooner than late 2007.

Jaedreth
Sorry, Jaedreth, I think you need to read some of the info out there. The relationship between Power4 and the 970 is well known and authoritative. Same goes for the next generation of Power5 and the 980. Search the forums and I'm sure you can dig up the links.

[jaedreth]

I'm not questioning that.

What I'm questioning is the relationship between PPC 980/Power5 and G6.

That's all.

Jaedreth

[Phil Of Mac]

Originally posted by daveL
Well, I can't give you a link, but it was stated that, from this point on, PowerX development and PPC9X0 development would proceed in parallel. The 980 should arrive very shortly after the Power5.

The more I think about it, the more certain I am that the info above came out of a news item referenced here at MR.

Which would indicate that the G6 will come very, very soon.

Originally posted by jaedreth
So just because I compare the *life span* of G5 to G4, and expect it to roughly match (my own personal opinion), doesn't mean that I forsee the same bs Motorola put us through happening with IBM. In fact, I'm expecting great things.

Perhaps the long lifespan of G4 was part of Moto's BS.

Consider:

The 601 (IBM exclusively) was the first generation, or "G1". It lasted from 1994 to 1995 before the next generation came out. 1 year.
The 603/604 "G2" (IBM and Motorola at Somerset) lasted from 1995 to 1997. Two years.
The G3 (IBM and Motorola at Somerset) lasted from 1997 to 1999. Two years.
The G4 (Motorola exclusively) lasted from 1999 to 2003. Four years.

Conclusion: The G4 lasted twice as long as it should have. It is the outlier. A one to two year lifespan for the G5 (until it is replaced by the G6, it will actually be used longer than that) is perfectly reasonable.

The POWER4 had been out for awhile when IBM developed the 970. If POWER and PowerPC development continue at IBM concurrently, than it's reasonable to assume that the G5 will be short-lived, the G6 will last about as long as the POWER5, until the POWER6/G7 comes out, and so forth.

[daveL]

Originally posted by jaedreth
I'm not questioning that.

What I'm questioning is the relationship between PPC 980/Power5 and G6.

That's all.

Jaedreth
Oh, so you are saying you're not sure that the 980 will makes its way into the Mac? I guess nothing is for certain, but Apple and IBM have worked very hard to communicate that there relationship is for the long haul and that Apple will benefit from IBM's continuing development of the 9x0 processor architecture. I guess we will see.

[jettredmont]

Originally posted by jaedreth
There *was* a report here about a supposed PPC 980, that more recent reports showed it would likely still be classified as 970, but a new revision.

I do expect technology to increase, and at a far faster rate after Moto is out of the picture, however I still don't forsee Apple releasing a "G6" computer any sooner than late 2007.

Jaedreth

So, if I understand you correctly, you are saying that the 980 (ie, a Power5 desktop derivative) will make it to Apple's lineup, but that Apple will still call it a G5.

If that's not what you're saying, it should be :)

There is nothing stating that Apple must brand a Power5-derivative "G6". In fact, the "G" number should, by design, escalate quite slowly, as a result of fundamental advances. Processor shrink is (obviously) not such an advance; increased FSB is not such an advance; bringing the memory controller on-chip is not such an advance.

G4: Altivec
G5: 64-bits
G6: (dual core? SMT aka HyperThreading?)

If the Power5 derivative does not have something fundamentally new, then it should not be designated a "G6". IMHO, SMT doesn't qualify either (heck, even Intel, they of the P2->P3 transition that meant next to nothing, didn't bump their proc number with HyperThreading!) Dual cores would do it. Any other ideas out there about what should be the next generational leap?

This is all IMHO, but because of the fairly small period of time between the 970 and the 980 (assuming previous rumors are correct there), I don't see the 980 being called "G6". I suspect the "G6" is a few years off, but that we'll see the "980" on our desktops a bit sooner than that.

[Phil Of Mac]

Originally posted by jettredmont
There is nothing stating that Apple must brand a Power5-derivative "G6". In fact, the "G" number should, by design, escalate quite slowly, as a result of fundamental advances.
G4: Altivec
G5: 64-bits
G6: (dual core? SMT aka HyperThreading?)

If the Power5 derivative does not have something fundamentally new, then it should not be designated a "G6".

Well, keep in mind that the G3 was always the exact same processor design. The G3 you use today is a variant of the same chip the G3 of 1998 was. The G4 you use today is a variant of the same chip that it was in 1999. The G6 is simply going to be the next chip that someone designs.

[jaedreth]

Jettredmont: Exactly what I mean.

And Phil, what we're saying is that it won't be the next chip someone designs, but the next *generation* of chip.

Jaedreth

[Phil Of Mac]

One and the same, jaedreth.

No one will design a new chip unless it has an inherent advantage over the current chip. Thus, a new generation. (Apple is more than happy to use one chip in all their products, they did it with the G3, couldn't do it with the G4, but with the G5, they'll probably go back to it.)

My point is, the G6 won't necessarily have any big new feature. It could, but there's no evidence that for Apple, the "G" designations mean anything other than "how much more new and improved the chip is".

[daveL]

Originally posted by jaedreth
Jettredmont: Exactly what I mean.

And Phil, what we're saying is that it won't be the next chip someone designs, but the next *generation* of chip.

Jaedreth
I agree that dual cores would demand the G6 moniker. I really don't care *what* they call them though, and I don't relate to threads that debate that issue endlessly. Then again, I'm an engineer, not a marketing guy or a sales guy or a poet. To each his own.

[jaedreth]

Apple didn't start using Gx designations until the G3.

Technically the 601s were G1, 603 and 604 were G2.

So more than one chip are within a generation. If IBM does release this new G3 chip, the chip has a new name, but it's still a G3 chip. It's still using third generation technology.

All of the G4 chips had their names, but they were each G4 technologies.

Neither a new revision of an existing chip nor a brand new chip will qualify as a new generation because this isn't a marketing thing.

Just because Intel artificially inflates numbers for the next-gen appeal doesn't mean others do. The Power5 is the next evolution of the Power4, however that does not mean that they aren't both "G5" chips. (In fact, a friend of mine does AIX L3 support at IBM, and allegedly IBM doesn't use the G5 moniker for either the Power4 nor the upcoming Power5, but a different processor that is their supercomputers that can only be sold to the US Gov't. So IBM's use of G5 does not even relate to Apple's. Yet this is still a 5th generation processor used in Macs, so G5 is appropriate, and Apple will not use G6 until the processors they use advance a generation.)

Jaedreth

[Phil Of Mac]

Originally posted by jaedreth
Apple didn't start using Gx designations until the G3.

Technically the 601s were G1, 603 and 604 were G2.

So more than one chip are within a generation. If IBM does release this new G3 chip, the chip has a new name, but it's still a G3 chip. It's still using third generation technology.

G2 was an exception only because the 603 and 604 were used at the same time. The G3 was intended to be used in all products at the same time, as was the G4. (Motorola's production problems scuttled that plan.) The G5 will probably be used in this manner. The "Gx" moniker relates to using the same chip across the entire product line because that one chip alone represents an entire generation of design.

Originally posted by jaedreth
All of the G4 chips had their names, but they were each G4 technologies.

They were variants of the same chip.

Originally posted by jaedreth
Neither a new revision of an existing chip nor a brand new chip will qualify as a new generation because this isn't a marketing thing.

Just because Intel artificially inflates numbers for the next-gen appeal doesn't mean others do. The Power5 is the next evolution of the Power4, however that does not mean that they aren't both "G5" chips.

Actually, neither is. G5 is Apple's marketing term for the 970 processor (and presumably its variants).

Originally posted by jaedreth
Apple will not use G6 until the processors they use advance a generation.

The next chip WILL be an advance of a generation. That's what the Gx thing is about.

[jaedreth]

Whatever. We're not going to agree on this one. My prediction stands. No G6 products til late 2007. Feel free to disagree.

Jaedreth

[mim]

Originally posted by Phil Of Mac
G2 was an exception only because the 603 and 604 were used at the same time. The G3 was intended to be used in all products at the same time, as was the G4. (Motorola's production problems scuttled that plan.) The G5 will probably be used in this manner. The "Gx" moniker relates to using the same chip across the entire product line because that one chip alone represents an entire generation of design.


They were variants of the same chip.


Well, this discussion is just about all G'd out - but I think it's worth pointing out that the original G4 bears little resemblance to the current design. The original G4's are more like G3's + an altivec unit. It's the altivec that makes it a "G4".

Apple will name thing's Gx where ever they see the bigest marketing impact - and that's reflected in leaps in evolution, not tweaks.

[Gyroscope]

This is what I've heard.

Unlike Power4, Power5 is not exclusively designed to be used in big iron servers. Its design will allow for use in smaller servers even high end workstations. It will come with an Altivec(SIMD) unit and it will incorporate hardware multi-threading. Its gonna be very easy to make derivative of this chip just by scraping one of its two CPU cores or there will be almost no need to make any changes to its design in order to be used in Apple high end boxes.

By the way!

G3+Altivec!=G4

G3 was derived from PPC 603 while G4 borrowed heavily from PPC 604 including far superior FPU unit.

[Phil Of Mac]

Originally posted by Gyroscope
G3+Altivec!=G4

Altivec factorial?

[Gyroscope]

!= =not equal C/C++/Java :):)

[Phil Of Mac]

Oh. I thought that either you meant Altivec factorial or you were really excited about Altivec :)

[shadowself]

The generations did mean something even though the "Gx" series monikers were not used openly until the "G3" series.

The original chip -- the PowerPC 601 -- was a derivative of a combination of the original single chip version of IBM's POWER CPU and Motorola's 88000 CPU (The original POWER CPU actually contained either 5 or 7 chips, depending on the model, to house all the units in it and to dissipate heat -- I believe it took IBM 2 or 3 years to get the first single chip version of the POWER CPU out the door after the original CPU chip sets shipped and within a year after that IBM Manassass started shippind a radiation hardened version for the military and space applications. The Motorola 88000 CPU was a 1-5 chip configuration depending upon how much static RAM Level 1 cache you wanted in your system.) The original chip was a truly 32 bit design just like the chips it was based upon, but, just like the POWER CPUs it was designed from the start with the intention of eventually taking it to a 64 bit chip. The AIM consortium (Apple, IBM and Motorola) took the execution units, predominantly, from the POWER chip and the internal busses and the ability to be extensible in nature (ability to add unique execution units as the chip evolved) from the 88000. This first hybrid one out the door was the 601.

The 603 and 604 (and their variants) were the next generation. They were fundamentally different in that the 603 emphasized power coservation and simplicity while the 604 had the fullest implementation, in hardware, for fully symmetric multiprocessing. (Full SMP could be done in the 603, but IIRC it only had MEI of the full MERSI suit so a lot of the cohernecy task had to be done in software -- which Be Inc did when it used 603s. IIRC, the 604 had the full set of MERSI hardware hooks.)

The "G3" from Motorola was specifically tuned -- in hardware -- to the Mac OS. Motorola had been negotiating to be a Mac OS compatible vendor. Motorola wanted the G3 tuned to that OS. Motorola worked with Apple to carefully model how the PowerPC chips ran the Mac OS of the time and tuned the hardware in the G3 to run the Mac OS and Mac applications as fast as possible. In fact Motorola not only got a license to be a Mac cloner, Motorola got the right to license OTHER companies as Mac cloners -- thus getting Motorola a possible double, triple, or even more, revenue stream in the Mac clone market.

(This hardware tuning of the G3 for the Mac caused no end of shouts in the benchmarking world that Apple had "cheated" and that such things as the Byte benchmarks -- remember them? -- were tailored to the Mac. In fact the opposite was true, Byte actually went in and modified its benchmark to run faster on the Intel chips of the time -- actually going in and changing a malloc call to make the Intel chips perform better. No such modification of any kind was done for the G3.)

We all know what happened to the Mac clone market. Motorola became very, very annoyed (to say it mildly) with the abolishment of the Mac clone market.

Meanwhile IBM loved the concept of the G3. It was a "clean design". IBM went forward with it, enhancing it dramatically. I remember hearing of samples shipped to various users (even Apple), but I don't remember any significant implementations of this very fast G3.

Motorola had the G4 under development before the death of the clone market -- IIRC, internally referred to as the PowerPC 7XXX series rather than the "G4" label. The big change for the G4 from the G3 was that it was the first chip to really use the extensible nature of the 88000 derived internal bus. Motorola added a vector unit (actually a SIMD unit not a true vector unit like on the old Cray computers and IBM's 3090-600VF machinines). IBM decided to pull out of the AIM aliance and sold its interest in the joint development center to Motorola. To my knowledge IBM never did a G4 with, or without, a SIMD unit.

After the G4 Motorola did do a G5. It actually shipped some test chips. Some of these machines actually found their way out into the world, and many of us who had friends with these boxes heard great stories about how these chips were real "screamers". However, these were rarely referred to as G5 chips by anyone other than Apple, but rather as "Book E" chips after a set of specifications referred to by that name. Motorola never got these chips into production. The rumors were that there were serious design flaws in the chip that kept it from going into production. Motorola decided the cost to redesign the chip would be greater than any profit they would make off it -- because Apple would be the only significant customer for it. Thus Motorola killed the project. (As an aside, Apple and Motorola had an agreement that Motorola would give Apple a significant lead time before killing any PowerPC production or PowrPC development. Motorola appears to not have given Apple the appropriate lead time. Thus Apple is suing Motorola over killing the G5/Book E chip and not giving Apple enough notice.)

Enter IBM. Motorola is not going to do anything beyond variants of the G4. Apple needs something significantly faster. IBM offers to do a derivative of the POWER4 chip and add a SIMD unit to it. This would not be a true PowerPC derived from the original concept, but instead a POWER derivative with SIMD added to it, yet still use the PowerPC instruction set. Thus IBM created the PowerPC 970 based upon the POWER4 with SIMD added. Is the PowerPC 970 different from the POWER4? Definitely, but it does carry that heritage. In fact its hardware implementation carries more heritage from the POWER4 than it does from the PowerPC G4 (other than the SIMD unit, that is). It has the PowerPC instruction set and is software backwardly compatible so everyone is calling it a PowerPC.

In adding a SIMD unit to the POWER4 derivative, IBM did a sub optimal SIMD to get the chip out the door in a reasonably short period of time. It is definitely NOT a bad SIMD unit, but it could have been better if they had been given more time. IBM engineers have admitted this openly. No one but Apple refers to this chip as the G5. IBM calls it the PowerPC 970.

There will definitely be future versions of the PowerPC 970. IMB has alread hinted rather heavilyt at adding memory management, a better SIMD unit, etc. to the PowerPC 970. Will they be a 970+, 970e, 971 or 980? Who knows. I doubt the decision has even been made yet. Apple will almost definitely refer to these as G5 chips in G5 systems.

IBM is most definitely going forward with the POWER5 chips. Will Apple use the "G6" label with the PowerPC derivative of the POWER5 CPU (already rumored to be in development for several months now)? Who knows. I doubt Apple's marketing team has made that decision yet.

We should never confuse the POWER chips with the PowerPC chips. They are two different CPU families for very different markets. It is true, one has always had a heritage from the other, but they are two different chip families. Also we should never confuse IBM's G5 CPUs with either the POWER CPUs or the PowerPC CPUs. IBM's G5 CPUs (as were the IBM G4s before them) are mainframe CPUs and are a totally different beast.

Bottom line as I see it is IBM is agressively pushing the POWER series forward (funded through the purchases of truly massive machines by DOE and the EU). IBM has made so many hints over the past 6-12 months that they will agressively create an even better PowerPC derivative of the POWER5 that this seems like a virtual certainty. Will Apple call the POWER5 derivative a G6? Who knows? I doubt anyone does.

Is this all good for Apple and Mac users? Definitely.

[wizard]

Currently the Power5 and the imainged 980 are considered to be two different things just as the 970 is a differrent product from the Power4. Further I don't think IBM would ever comment on a customers projected usage of anything they make. This would be especially the case with respect to Apple, considering Steves obsession with keeping everything secret.

Reports or not I would see Apple being driven to use a 980 as soon as possible. For one thing it si still up in the air as to how well the 970 actually performs. Second; by the time the G5 MACs are in full produciton both AMD and Intel will have signifcantly improved hardware on the market. Apple will have to implement the 980 quickly just to keep up with Intel hardware.

Both IBM and Apple realize that they need a single processor (chip) Power 5 implementation. I would not be surprised to see the 980 (G6) in a Mac 6 month after it is taped out for the first time. As good as it is (or a hack depending on veiw point) the 970 will not improve at a fast enough rate.

I don't believe that the 970 will stagnate either, but simple clock ramping won't be enough to save its behind. Hitting 3 GHz may help but it will have to be judged against what Intel and AMD are selling at the time. A 50% speed boost may be to little by the end of the year.

DAve

Originally posted by jaedreth
I think you misunderstood me.

1) IBM has not reported that the next processor design, tentatively coded the 980 and/or Power5, would be used in Macs.

2) The reports I've seen seem to indicate they will *not* be used on macs, but technological advances based on this project may eventually be incorporated into Macs.

3) That the G5 would continue to be used, not as is, but as a growing processor architecture, for the next four years.

4) That *if* IBM were to eventually make the Power5 into a chip for Macs, aka a G6, that it would not happen until at least Fall of 2007.

My point is not that the G5 will stagnate. I don't forsee that at all.

I see G5 getting faster and faster. IBM is already working on a second Revision of the 970, that is still G5. And IBM will continue to do so. The chip will get faster and mature more, providing better performance at far more frequent intervals than with Motorola's G4's.

So just because I compare the *life span* of G5 to G4, and expect it to roughly match (my own personal opinion), doesn't mean that I forsee the same bs Motorola put us through happening with IBM. In fact, I'm expecting great things.

But also keep in mind, IBM was using G4 technology in it's current machines four years before Apple had its first G4. IBM makes High End Servers, so the processors they make for themselves are not going to the the processors Apple uses, perhaps except after a good period of time.

I hope that clarifies my position.

Besides, G5 will scale very well over four years, and perhaps beyond. I don't forsee needing a G6 before then, though we've needed G5 for two or three years now.

Jaedreth

[panphage]

All I know is the IBM guy at the WWDC said IBM was well along in development of the *next generation* of chips. I had just assumed that meant a POWER5 derivative, I didn't think a die shrink of the 970 warranted a next generation label. But maybe IBM's feelings differ.

Can someone with some chip design knowledge explain/speculate on why apple wouldn't want a dual-core chip for their powermacs? I mean since they keep pumping out dual processor systems, isn't a dual-core a shortcut?

[Quila]

And some of it builds on what others posted.

1) The POWER5 will never show up in workstations, and not likely in any Apple server. It is very expensive and extremely reliable at the expense of speed, and people in Apple's market want speed and are not willing to pay for that class of reliability.

2) PPC970 was a POWER4 derivitive, but the POWER5 and next gen PPC970 will be designed together, as will future generations.

3) In the pipline are supposedly better predictive logic (very important with the grouping it does), on-die memory controller, their version of Hyperthreading and a better SIMD unit. Better SIMD was left out due to time to market constraints.

4) I doubt the PPC970 and heirs will show up in Apple blade servers since IBM has already announced they'll be making those.

5) According to an Ars interview, gcc has a hard time dealing with the PPC970 architecture, and they're working on a better compiler. I'd think a dedicated compiler would be very important for the PPC970 with its group dispatching.

[jaedreth]

Thanks Shadowself, you explained this all far better than I could, or did.

IBM G5 mainframes (which IBM is still KING of mainframes) are only sold to .gov. Period. Federal law prohibits machines this powerful to be sold anywhere else.

So G6 as Apple uses it is purely an Apple Marketing thing.

So it will still be a few years to G6. But who cares, this supposed 980 will likely be counted as G5, but that doesn't mean it would be any less awesome. It's just a name.

Jaedreth

[jettredmont]

Originally posted by Phil Of Mac
Well, keep in mind that the G3 was always the exact same processor design. The G3 you use today is a variant of the same chip the G3 of 1998 was. The G4 you use today is a variant of the same chip that it was in 1999. The G6 is simply going to be the next chip that someone designs.

I disagree. For one thing, there are two designs of the G3 at least (Moto's and IBM's). Amongst the G3 line we have various "designs" as described by IBM and Moto, corresponding to the various manufacturer part numbers.

For another, the G4 above 500MHz is a quite different design from those G4's below 500MHz (many more pipeline stages).

Chip design (as all engineering design) is an evolutionary process. Each design is based on the last to some extent, and borrows concepts if not concrete implementations from it. The difference between a "tweak" and a "whole new design" is one of degree and semantics.

Apple exists on a slightly different plane. They are not a chip manufacturer. They only need to redesignate their computers when there is a significant, user-visible difference. As I said, the G4 and G5 each have real, concrete, user-visible differences relative to their previous generations. I would expect no less from the G6.

Alongside this, Apple will want to embue some facade of stability in their computer lineup. If there is a "generational leap" every six months, you have much less buyer confidence ... why not just wait six months and get the next generation instead? Not to mention the brand recognition that you just throw out the window by shifting G5->G6 so quickly. G6 less than a year after the debut of G5 is silly.

So, it comes down to:

1) There is no reason for Apple to name a 2004 980 chip "G6"
2) There is a business reason to avoid "G6" for a few years.

Therefore, I don't expect G6 next year.

[jaedreth]

I think you misunderstand.

So many are getting hooked on the idea that the Power5 derivative *will* be called PPC 980. This hasn't been decided. Also many are getting hooked on the idea that the 980 (which DNE), is G6. Not true. IBM has nothing they call G6. Apple has no plans as far as G6. G6 DNE. (Does Not Exist, math lingo)

3 altivec cores, improved "vector" cores, and hyperthreading *are* planned for this "next generation" chip.

As great as the 970, and as far better it is than the current G4's, it is admittedly "slapped together" in a rush.

IBM is taking their time with it's replacement, and wants to do it *right*. So we win big time.

Also these new chips are likely to hit first in Xserve and in high end workstations if those rumors turn out to be true. Then Desktops.

Thus, when that happens, eventually, the *consumer* market can transition to 970 while the *professional* market will move to "980". This being what will bring the transition to G5 to completeness. However, this could take 3-5 years for this cycle to complete.

But rest assured, it only gets better from here.

Jaedreth

[jettredmont]

Originally posted by panphage
All I know is the IBM guy at the WWDC said IBM was well along in development of the *next generation* of chips. I had just assumed that meant a POWER5 derivative, I didn't think a die shrink of the 970 warranted a next generation label. But maybe IBM's feelings differ.

Can someone with some chip design knowledge explain/speculate on why apple wouldn't want a dual-core chip for their powermacs? I mean since they keep pumping out dual processor systems, isn't a dual-core a shortcut?

1) Dual core is inherently more expensive than dual chips on low-yield processes
2) Dual core is not as flexible as dual chips from a design perspective
3) A dual-core chip will put out more heat than a single-core chip, and so you have a concentrated heat source which requires more intelligent dissipation design.
4) Dual cores generally share a single FSB, which the G5's don't (G4's did, though, getting the "worst of both worlds" award).

On the other hand:

1) Dual-core can be cheaper if you have a very high-yield process
2) Dual core chips use less power than two single-core chips (in theory because the existing dual-core chips are on power-hungry monster computers ...)
3) Dual-core chips physically take up less space, making design relatively easier (aside from heat dissipation, see above).
4) Dual Cores Are Cool (tm).

[Hattig]

Originally posted by jaedreth
If the PPC 970 is a *variant* of the Power4, designed for a workstation instead of a high end server, then I could see this. I think this may be the case, but I'd have to clarify this with my sources.


The 970 is a tweaked single-core version of the POWER4 with an added altivec unit. It does incorporate some stuff that went into the design of the POWER5, but overall it is very closely related to the design of the POWER4.


There *was* a report here about a supposed PPC 980, that more recent reports showed it would likely still be classified as 970, but a new revision.


No, that was the 90nm shrink of the 970 that will get the processor to 3GHz and above by the middle of next year. The 980 will most likely arrive between the end of next year and the middle of 2005.


I do expect technology to increase, and at a far faster rate after Moto is out of the picture, however I still don't forsee Apple releasing a "G6" computer any sooner than late 2007.

Apple fell behind enough with the G4, I can easily see Apple releasing a G6 in 2005.

[jaedreth]

If the official 980 isn't expected until 2005, that could be the case.

Whether this 980 winds up being G6 or not is an Apple Marketing decision. But whether the 980 Power5-based derivative is the G6, or if a later variant of a Power6 winds up taking that name, the chips will be released as they will be released, and they will rock for sure. :) The only way out is up. :)

We'll just have to wait and see.

Jaedreth

[Phil Of Mac]

A few things:

1. This is all speculation based on rumors. It's stupid to point out that nothing is set in stone yet. Of course not. This is MacRumors.
2. No one's saying the G6 is coming out in six months. All I'm saying is that in terms of release, it might take anywhere between one year from the G5 (i.e. next summer) to two years. The G5 is probably a transitionary chip, considering how quickly it was slapped together out of spare parts, so to speak. A single Power4 core there, an Altivec unit tacked onto the side, with a few modifications to make it work. The design of IBM's PowerPC counterpart to the Power5 (rumored to be called the "980", so I'm calling it that) is reportedly being done more methodically, alongside the Power5. So it's possible that the 970/G5 is simply a 601-style immediate solution that's due for a quick replacement. That would lean toward one year. The fact that the 980 corresponds with the Power5 suggests that the 980 will probably be out around the time the Power5 is, and if later, then not too much later. That puts an upper limit of 2 or so years on the 970.

G1 lasted one year, G2 and G3 lasted two years. G4 lasted four years, but that was for many reasons (end of Book E, Motorola's troubles, etc.) a fluke.

Also, G3 and G4 did refer to variants of the same chip. The G3 was called the 740 or 750, while the G4 was called the 74xx. Motorola's Book E G5 was the 8500--indicating that in Motorola's eyes, it was a "new processor". The above-500 MHz was not a "new processor" because it was still a 74xx.

And the G5 is the 970.

IBM obviously considers the 980 to be a "new processor", because it corresponds with Power5, which is also a "new processor" compared to the Power4.

Therefore, the 980 will be universally considered a "new processor", justifying the name G6.

[hasapi]

Originally posted by Phil Of Mac
A few things:

1. This is all speculation based on rumors. It's stupid to point out that nothing is set in stone yet. Of course not. This is MacRumors.
2. No one's saying the G6 is coming out in six months. All I'm saying is that in terms of release, it might take anywhere between one year from the G5 (i.e. next summer) to two years. The G5 is probably a transitionary chip, considering how quickly it was slapped together out of spare parts, so to speak. A single Power4 core there, an Altivec unit tacked onto the side, with a few modifications to make it work. The design of IBM's PowerPC counterpart to the Power5 (rumored to be called the "980", so I'm calling it that) is reportedly being done more methodically, alongside the Power5. So it's possible that the 970/G5 is simply a 601-style immediate solution that's due for a quick replacement. That would lean toward one year. The fact that the 980 corresponds with the Power5 suggests that the 980 will probably be out around the time the Power5 is, and if later, then not too much later. That puts an upper limit of 2 or so years on the 970.

And the G5 is the 970.

IBM obviously considers the 980 to be a "new processor", because it corresponds with Power5, which is also a "new processor" compared to the Power4.

Therefore, the 980 will be universally considered a "new processor", justifying the name G6.

Logic suggests you will be right, the 980 will achieve greater performance (speculation about twice as fast per clock), so its more than likely that Apple marketing will introduce this as a G6. Of course its not a '6th generation' processor but then neither is the 970 a '5th generation'. Its just naming to differentiate processors of disparate performance.

Everything ive read from IBM suggests this chip (980) will be available around early 2005 (at least in samples).

I just hope (not looking good?) Motorola can get some more legs out of the 7457 to 2GHz+ for the consumer range over the next 6-9months before we get 0.09um G5's into the PBooks and then into the iMacs/iBooks in time for the PMacs to go to G6.

[Phil Of Mac]

Originally posted by hasapi
Logic suggests you will be right, the 980 will achieve greater performance (speculation about twice as fast per clock), so its more than likely that Apple marketing will introduce this as a G6. Of course its not a '6th generation' processor but then neither is the 970 a '5th generation'. Its just naming to differentiate processors of disparate performance.

Exactly. It doesn't necessarily mean they add a new feature.

Originally posted by hasapi
Everything ive read from IBM suggests this chip (980) will be available around early 2005 (at least in samples).

That pegs release in the summer of 2005. Sounds like a two-year generational period, quite reasonable and in line with past history.

Originally posted by hasapi
I just hope (not looking good?) Motorola can get some more legs out of the 7457 to 2GHz+ for the consumer range over the next 6-9months before we get 0.09um G5's into the PBooks and then into the iMacs/iBooks in time for the PMacs to go to G6.

I think the current G4's (the late ones) are the end of the line. IBM's working on 90 nm, which is probably the only real obstacle to putting the G5 into the PowerBook. I'm guessing that summer 2004 or the following winter will see the G5 in all lines. Remember, the G3 went across the entire line in a matter of 1 day, back in 1997. (Discontinuing the entire product line and replacing it with the Power Mac G3 and a hacked PowerBook 3400 upgraded to G3 was a feasible option then, but not now.) An entire G3 product line took until 1999 to complete, but that's because they had to develop four all-new products, using completely different technologies, from the ground up. (The "New World" computers, based on CHRP technology.)

If IBM takes 6-9 months to go to 90 nm, then I think Apple can put G5's across the line by next summer. The iMac and PowerBook are the obvious choices, then the eMac and iBook.

And no, I don't think the iBook will ever see a G4. I don't think it should. The purpose of the iBook is to be a consumer portable that's not just yesterday's technology. Using yesterday's processor makes it yesterday's technology. The iBook was a G3 when the PowerBook and Power Mac were both G3. Apple liked that, and probably wants that type of thing going. I can believe that we'll see the G5 across the line. I can believe that since the 980/G6 will be a 90nm chip. it'll be across the line within a matter of months. That's what I would do if I was Apple's CEO, and I think Steve Jobs is just as smart as I am.

[hasapi]

Originally posted by Phil Of Mac
If IBM takes 6-9 months to go to 90 nm, then I think Apple can put G5's across the line by next summer. The iMac and PowerBook are the obvious choices, then the eMac and iBook.

Again, it appears that this is indeed the case, the problem for Apple right now is that in terms of price/performance both the PowerBooks and the iMacs SUCK!. The G4 is relatively stuck at 1G, discounting the super hot 7455 1.42's (unsuitable for PB's/iMacs).

Both the PB's & the iMacs NEED G4's at around 1.6G at the high end or G5's at say 1.2G. But what are the odds of Moto shipping 7457's @ 1.6G and in quantity!, probably why Apple engineers are trying to get G5's to work in PB's now - before SJ blows a fuse! :eek:

[AidenShaw]

Originally posted by jaedreth
IBM G5 mainframes (which IBM is still KING of mainframes) are only sold to .gov. Period.

IBM is not selling anything with a PPC970 chip today (the "PPC970" is the real name for what Apple marketeers have dubbed the "G5") - nothing, nada, zilch, zero. (Actually, IBM is already selling G5 and G6 systems in the S390 line (http://www-1.ibm.com/servers/s390/pes/), but that has nothing to do with PowerPC.)

IBM POWER4 systems (of course, the PPC970 (Apple G5, not IBM G5) is a "POWER4 Lite" chip) are in a number of systems, and are on the open market. (At least, I know that the big black system in the 2nd floor lab is a 32-processor POWER4 system, and I *don't* work for the government.)

You are so full of it that the whites of your eyes are brown....

[Phil Of Mac]

Originally posted by AidenShaw
IBM is not selling anything with a PPC970 chip today (the "PPC970" is the real name for what Apple marketeers have dubbed the "G5") - nothing, nada, zilch, zero. (Actually, IBM is already selling G5 and G6 systems in the S390 line (http://www-1.ibm.com/servers/s390/pes/), but that has nothing to do with PowerPC.)

I think that's what jaedreth was talking about. And yes, it was dumb to bring it up.

[Fitzcaraldo]

With any luck the next revision of G5 will include 64 way altivec like this:

CS301 (http://www.wired.com/news/technology/0,1282,60791,00.html)

If not then the next generation should? I quit like the idea of a personal 'Super Computer'...

[yamabushi]

Seperating the Altivec unit into a large math coprocesor similar to the CS301 might be worthwhile. It certainly sounds interesting. Alternatively, IBM could leave things the way they are and supply an extra Altivec PCI-X card for those who need it. However it kind of seems llike the opposite of their current stategy of putting everything onto the chip.

[Phil Of Mac]

Originally posted by yamabushi
Seperating the Altivec unit into a large math coprocesor similar to the CS301 might be worthwhile. It certainly sounds interesting. Alternatively, IBM could leave things the way they are and supply an extra Altivec PCI-X card for those who need it. However it kind of seems llike the opposite of their current stategy of putting everything onto the chip.

PCI-X? Altivec currently operates at the clock frequency. The PCI-X bus won't even be a tenth of that.

Bad idea.