More Dual PowerPC 970 Info...

[MacRumors]

eWeek provides (http://www.eweek.com/article2/0,1759,1627893,00.asp) more information on the upcoming dual-core version of the PowerPC 970 due from IBM.

The original Power4 processor which was the basis of the PowerPC 970 design began as a dual core processor. According to eWeek, the additional of the second core should improve efficiency:

IBM documents suggested that hardware and software optimizations would make this processor more efficient in many computing situations than two separate processors at the same clock speed.

Similar details were previously provided (http://www.macrumors.com/pages/2004/07/20040723172623.shtml) by ThinkSecret.

[GroundLoop]

mmmm..pseudo-4-way PowerMac...A shame that I just bought a dual 2.5 GHz. Maybe, I will upgrade to one of these when they become avaialable.

[Diatribe]

Woooh, now this sounds interesting.
Even FASTER than 2 processors running at the same clock speed? The last info said that they would be running around 3-3,5 Ghz so if you put 2 and 2 together the new ones would be running faster than a dual 4 Ghz.
This is mind boggling :D Me wants one...
I guess that also means that this will be the end of dual proc. powermacs.

[Chaywa]

Yeah,

to me this seems to spell the end of dual processor powermacs. That way apple can save money by only putting one processor in without sacrificing power and it will reduce the heat problems they have been creatively challenged to deal with. The trick will be selling these machines with one physical processor to a customer base that has come to believe that dual procs are the way to go for best performance.

[Knox]

I guess that also means that this will be the end of dual proc. powermacs.

At the bottom end I suspect you might be correct, but I wouldn't be surprised to see the top end retaining dual processors, along with the Xserve. The existing dual-core Power 4/5 machines are all multiple CPU for example, although admittedly they are a different market and price range. I think dual core is being seen as a replacement for increasing clock speed, rather than replacing multi-cpu.

[aussiemac86]

Yeah my dual 2.5 is on its way, but no way am i prepared to put up with the waiting for an unreleased mac again. I almost got a dual 2 in decemebr of last year, but put off until the eventual release of the rev B's..........and i am still waiting....it will all be better once it arrives

[Diatribe]

Yeah,

to me this seems to spell the end of dual processor powermacs. That way apple can save money by only putting one processor in without sacrificing power and it will reduce the heat problems they have been creatively challenged to deal with. The trick will be selling these machines with one physical processor to a customer base that has come to believe that dual procs are the way to go for best performance.

If they are really this fast it is going to be a breeze. I mean seriously, if they had a processor out by the same time next year that was about as fast as a dual 4Ghz would be right now.... how many people would be cheering and wetting their pants? :D

[AmigoMac]

Powermac : Dual-Core-Dual-Processor @ 3 GHz :cool:
iMac / Powerbook : Dual-Core-Single-Processor :cool:
iBook / eMac : Actual G5 :cool:

????

Yes! Bring them up! Go IBM! Go apple! :p :p

[AidenShaw]

Even FASTER than 2 processors running at the same clock speed?

The IBM statement most likely means "more efficient at SMP than 2 separate processors", not faster at a single-threaded task.

The reason is that a lot of "chatter" occurs between CPUs in an SMP system, for example interlocked instructions need to be coordinated and processors have to check each other's caches. (If one CPU changes a chunk of memory, if that chunk is in the other CPU's cache the cached version is no long correct.)

In standard SMP, this "snooping" occurs on the FSB (or the coherent HT link in the case of Opteron).

In a dual core chip, the snooping is within the chip itself, at full core speeds.

[~Shard~]

Although many of your are getting all excited with the thought of dual dual-core chips going into PowerMacs, has anyone every contemplated that perhaps, once these chips are ready, they won't go into dual processor machines? Perhaps the fabled 3 GHz PowerMac will not be an actual DP system, but instead have only 1 of these dual-core 970s. Sure, the effect will be the same (and from the sounds of it, it will actually be faster), but I wouldn't automatically assume that whenever these chips are ready to go, Apple will be able to simply keep their dual processor PowerMac config and slap these dual-core chips in place of the single core chips for gobs of power. I don't see this as being too likely.

As Chaywa said, this, if anything, will spell the end for the DP PowerMacs.

[manu chao]

... this, if anything, will spell the end for the DP PowerMacs.

Will everybody please stop repeating this nonsense about this being the end of dual processor Macs.

A dual-core is nothing but two processors on one chip, there are still two processor cores in the computer. The OS and all programms will see two processors, you will even have a somewhat better performance than with a two processor computer.

If you think that the wide public and maybe yourself would perceive a dual-core Powermac as somethink less powerful than a dual-processor Powermac, you should for once accept that the opposite is the case and also rest assured that Apple marketing would try its best to correct that impression.

Naturally, everybody who said this would make it relatively easy for Apple to create a four-way computer is absolutely right.

BTW, the timestamp on the eWeek article is 26 July, three days after the Thinksecret story and three days before macrumors picked it up.

[AidenShaw]

The OS and all programms will see two processors, you will even have a somewhat better performance than with a two processor computer.


Note that IBM/Apple needs to double the amount of memory bandwidth per chip in order to maintain performance.

If the dual-core chips shares one memory bus (for example, if the dual-core chip (or 2) are dropped into the current G5 architecture) then the memory bandwidth *per CPU* will be halved.

That's gotta hurt.

They'll need to double the speed of the bus, double the number of busses, or double the width of the bus to maintain the current balance of memory bandwidth per CPU.

[bathysphere]

Woooh, now this sounds interesting.
Even FASTER than 2 processors running at the same clock speed? The last info said that they would be running around 3-3,5 Ghz so if you put 2 and 2 together the new ones would be running faster than a dual 4 Ghz.
This is mind boggling :D Me wants one...
I guess that also means that this will be the end of dual proc. powermacs.
that's not what they meant. they mean that a dual core 3ghz machine will run more efficiently (and a bit faster) than a conventional dual processor 3ghz machine.
a dual core 3ghz machine will not run as fast as a 6ghz, 5ghz, or 4ghz machine. it's the same logic as a regular dual processor computer, it's basically just more efficient.

[dizastor]

A Dual Dual Powermanc has such huge drool factor.

Must... start... saving...

[Frobozz]

"This move, according to sources at IBM, is intended to expand the PowerPC's top frequency capabilities, currently at 2.5 GHz, in a dual-core configuration."

Looks like the January - March revision of the PowerMac will contain dual core 970mp's to get performance past 3GHz. The article doesn't state what GHz the chips will run at, but it does say this about the real world performance:

According to Glaskowsky, customers running imaging and scientific applications developed for the Mac platform will appreciate the multicore design. "A dual-core chip is more effective than a single-core chip on problems that stress the computational resources of the chip, more than the front-side bus bandwidth. Because the 970FX has a very fast, efficient front-side bus, most Mac applications will favor the dual-core configuration."

This Glaskowsky guy is an "independant microprocessor analyst," so I wouldn't take his word as gospel... HOWEVER... this suggests an approach that will definately do two things for the Apple G5 architecture, and maybe a third:

1) Spread our heat production to a wider area, thus effectively making a much easier to maintain heat sink and dissipation profile. Though the article suggest that, since it will likely be clocked higher, it will porbably still need a water cooling heat sink.

2) Increase the relative performance of the PowerMac by making it more efficient than the dual processor design.

3) Speculation: If they don't need to use two physical chips to achieve the same scale of performance, then they might not put it into the PowerMac. This could make them cheaper to make. If they decide to keep a second chip in some high end configurations, then we actually double our current speed for a lot of things.

Personally, I think Apple will remove the second processor if they go dual core... and that's a good thing. Consider the benefits of a dual core 970 in a mobile environment, with all of the power saving features. :-) Mmmmm.

[Frobozz]

They'll need to double the speed of the bus, double the number of busses, or double the width of the bus to maintain the current balance of memory bandwidth per CPU.

So do you think they can just throw faster memory like DDR2 in the New PowerMacs, or will they have to accompany that with more bandwidth between the memory and the CPUs?

[Frobozz]

At the bottom end I suspect you might be correct, but I wouldn't be surprised to see the top end retaining dual processors, along with the Xserve. The existing dual-core Power 4/5 machines are all multiple CPU for example, although admittedly they are a different market and price range. I think dual core is being seen as a replacement for increasing clock speed, rather than replacing multi-cpu.

I don't know. The more I think about it, the less likely I think it will be for a dual-dual. User tasks won't benefit nearly as much from 2 dual core CPU's, on the dollar, as server applications, databases, and scientific computation would.

Honestly, they may still do it, but I'm not sure how well that would scale.

[Phinius]

A dual-core chip would not only be simpler and cheaper for a dual-processor computer, but it also enables Apple to move into the quad-processor server market. Dell sells a quad-processor server anywhere from about $9,000-$20,000, depending on the speed of the processor. Since Dell is the bellweather for mass market computers, Apple has a lot of potential for underpricing what Dell offers in a quad-machine, while also going after much higher gross margins.

[Pedro Estarque]

I believe that only highly threaded and very CPU hungry applications would benefit from a quad configuration. Maybe FinalCut 5 and other future Pro apps could take advantage of it and cut rendering time significantly, if they were coded properly. But I don't think Photoshop would see much improvement, as I believe the bottleneck here ( as in many other system tasks ) is memory bandwidth. I think they should make a quad Xserver for those who really need it.
I know the Mach Kernel is specially good at SMP but do you think that it can make 4 proc. talk to each other and be 350% faster then a single one on ordinary tasks ? Again, I fell that only special hard coded Apps could really benefit from such a machine. but maybe I'm wrong.

[GFLPraxis]

I believe that only highly threaded and very CPU hungry applications would benefit from a quad configuration. Maybe FinalCut 5 and other future Pro apps could take advantage of it and cut rendering time significantly, if they were coded properly. But I don't think Photoshop would see much improvement, as I believe the bottleneck here ( as in many other system tasks ) is memory bandwidth. I think they should make a quad Xserver for those who really need it.
I know the Mach Kernel is specially good at SMP but do you think that it can make 4 proc. talk to each other and be 350% faster then a single one on ordinary tasks ? Again, I fell that only special hard coded Apps could really benefit from such a machine. but maybe I'm wrong.

I disagree.
If they make Tiger highly threaded...
Imagine. You run Photoshop CS, Final Cut Pro 4, and Maya3D all at the same time.
Since Tiger is threaded, it allocates the programs-
Processor 1: OS X core system processes
Processor 2: Photoshop CS
Processor 3: Final Cut Pro 4
Processor 4: Maya3D

Even if Photoshop, Final Cut Pro, and Maya WEREN'T multithreaded, Tiger itself (being a multithreaded OS) would give each program it's own processor.

Result? Speed boost, baby!

[The Red Wolf]

I seriously doubt Apple would have gone to such lengths to water cool the dual 2.5 without refining the technology for future use. Would a dual core single chip be as hot as two 2.5s? Maybe. But more efficient and lower wattage don't speak to that. What if a refined water cooling system and dual core twin G5s make it into the high end G5? Reasonable?

Single chip, dual core Low end G5 Tower.
Single chip, dual core Mid range G5 Tower.
Twin chip, dual core - water cooled, High end G5 Tower.

Thoughts?

[nuckinfutz]

Like the poster above said. 970MP processors will give Apple much more lattitude in system design.

No this doesn't mean it's the end of 2 socket Powermacs. While it's true they can now do SMP with 1 CPU socket I have no doubt that Apple will have a high end configuration with 2 970MP. They won't have the luxury of not doing this because you better believe the dual core Opterons will be working in 2 socket motherboards and possible 4 socket.

As for the FSB we have to remember that the Powermac FSB is largely marketing. I'm not saying this because it's not real but we must look as the real bottleneck. The 1250mhz FSB can support 10GBps (5GBps up/down) of bandwidth. There is no way the dual 2.5 with its PC3200 memory is even getting close. Both CPU must share the total DDR bandwidth amount of 6.4GBps despite the fact that the bus can handle %50+ more bandwidth.

So even if we put both processors ondie and have them share a 1Ghz FSB we won't see a hit usually because not many applications stress the FSB. I'm very willing to see the L2 cache quadruple and the chip to chip communication speed increase with lower latency. That will really help some applications.

I expect Apple will have 2 socket 970MP systems provided IBM yields are sufficient. Don't expect them to be cheap however. I wouldnt' be suprised to see a Quad core system run $3999. This would limit the demand to mainly professionals and ensure that profitability for Apple. I do know there are a lot of people who'd be interested though. 3D animators, Video production pros, audio production, Scientific reseachers etc

Remember these systems aren't shipping until as late as mid 2005. They would not come without OSX 10.4 Tiger. Also remember that Tiger has synced with BSD 5.x and now has the fine grain locking for SMP systems and the kernel now can differentiate between "physical" and "logical" processors. Had Apple not planned to go beyond 2 cpu/cores then these functions in Tiger would not be needed.

As for threaded apps people need to realize why the current SMP systems multitask so well. Remember you will see a speed increase whether you are running one app with multiple threads or multiple single threaded apps. Either way you will derive a benefit from SMP.

[nuckinfutz]

The Red Wolf.

Exactly. I'm suprised at how many writers have commented about how hot the Dual 2.5Ghz chips are without having any empirical evidence regarding the heat max on this cpu. They are basically assuming that the chips are ungodly hot because of the LC system. Well I prefer to think of it as over engineering for the future.

Now that the 970MP system is potentially coming we can see why the liquid colling system will be necessary. Having both cores on the same die gives a wider area to cool which should help. Plus from the eWeek article we see there are enhancements to Powertune as well that will throttle the cores in tandem if things get a bit too hot.

Once 2.5Ghz owners get their computers sometime next week we'll have a better idea on how effective the LC system is and if it's suitable for cooling as many as 4 cores.

[Pedro Estarque]

Processor 1: OS X core system processes
Processor 2: Photoshop CS
Processor 3: Final Cut Pro 4
Processor 4: Maya3D
Result? Speed boost, baby!

Do you really run all this Apps together every day ? Wouldn't you prefer to run one of then but much much faster instead of 3 of them at you current speed?
If I'm editing some movie I won't be doing heavy Photoshoping. Maybe open one file and make some light changes. I would much rather see the end of the progress bar then being able run more at once without sacrificing current speed.

[GFLPraxis]

Do you really run all this Apps together every day ? Wouldn't you prefer to run one of then but much much faster instead of 3 of them at you current speed?
If I'm editing some movie I won't be doing heavy Photoshoping. Maybe open one file and make some light changes. I would much rather see the end of the progress bar then being able run more at once without sacrificing current speed.

I'm just making an example ;)

A better example would be, Maya 3D, Motion, and Final Cut. Or maybe SoundStudio.

What I'm saying is, even if you have a program that isn't multithreaded, the processor will stick each program on its own processor, letting you run four programs each at the speed of only one on a single processor. Whoo!

[nuckinfutz]

Do you really run all this Apps together every day ? Wouldn't you prefer to run one of then but much much faster instead of 3 of them at you current speed?
If I'm editing some movie I won't be doing heavy Photoshoping. Maybe open one file and make some light changes. I would much rather see the end of the progress bar then being able run more at once without sacrificing current speed.

You overestimate the speed increase from shutting your apps down. It really depends on your workflow but I know plenty of people that routinely have multiple apps open so that they can quickly access Photoshop to tweak graphics before importing into Final Cut. Multitasking is the future the preferred OSX method is to have a lot of RAM and just load up your applications.

a single processor running at a high clockrate ala a Pentium 4 is suited for running one app quickly.

SMP systems don't run as fast generally but they multitask so much better so they can be faster when you've loaded your system up with apps and processes.

[Pedro Estarque]

I'm just saying that the more Proc. you have the harder it is to spread the threads in a efficient way. One info might be in the first Proc.'s cache and is being needed in the fourth. I have 202 threads running right now, its not like you could split 202 ÷ 4 :
1º 970 -> 51
2º 970 -> 51
3º 970 -> 50
4º 970 -> 50

And make it run 4x faster.
Otherwise Apple could just gather some PPC 604e and make a 256 processor machine that would kill anything for a really nice price.

[Pedro Estarque]

You overestimate the speed increase from shutting your apps down.
Photoshop benefits a lot from free RAM. So running it alone leaves more memory to it and the speed is quite noticeable.
I run PS next to safari, mail and itunes every day. When I start Dreamweaver, Toast , iMovie etc I can see some sluggishness in PS. Specially in the first few minutes after switching from other apps. It takes more time to switch from app to app. UNIX is great, but RAM is a finite resource and the more Apps running the more you'll be hitting the HD. There's no magic. OS X does a much better job doing lots of things at the same time then OS 9 did, but memory intensive applications such as PS will run faster if running alone.

[Frobozz]

I disagree.
If they make Tiger highly threaded...
Imagine. You run Photoshop CS, Final Cut Pro 4, and Maya3D all at the same time.
Since Tiger is threaded, it allocates the programs-
Processor 1: OS X core system processes
Processor 2: Photoshop CS
Processor 3: Final Cut Pro 4
Processor 4: Maya3D

Even if Photoshop, Final Cut Pro, and Maya WEREN'T multithreaded, Tiger itself (being a multithreaded OS) would give each program it's own processor.

Result? Speed boost, baby!

Well, I don't know. Maybe I'm wrong, but unless you are a very fast individual I don't think you can use all 4 simultaneously. :-) Even then, if you set them up to render something all at the same time, you'd probably be better off just having one app open and letting 4 logical processors render frames: cpu 1 does frame 1, cpu 2 does frame 2, etc.

[JoePike]

I definitely think that Steve and the kids down in Cupertino will not shy away from putting dual dual-core processors in a powermac. They did it with the new graphics cards, didn't they? The 6800 provides the dual-DVI necessary for the 30" display, and you can plug in two of those at once if you want, right? This is a company that has a history of trailblazing and leading the way by "thinking big". Steve-o also likes turning heads, and a computer with a total of 6+ ghz of computing power would certainly turn some heads.

-Joe

[nuckinfutz]

I'm just saying that the more Proc. you have the harder it is to spread the threads in a efficient way. One info might be in the first Proc.'s cache and is being needed in the fourth. I have 202 threads running right now, its not like you could split 202 ÷ 4 :

That article already combats that. With ondie cores the cpus can snoop each others cache without traversing the FSB. The latency and bandwidth improve greatly. Code Cracking allows for the instructions to benefit from this increase in bandwidth and latency between cores by cracking the insruction in have and processing simultaneously. The kernel handles the threading, there is no need to split the threads evenly amongst 4 cores. Some threads are more demanding than others. The kernel handles these in in the best way possbile. Remember Tiger will have improved SMP performance with fine grain locking. Targetting and managing multiple threads will be improved over what we have today.


Photoshop benefits a lot from free RAM. So running it alone leaves more memory to it and the speed is quite noticeable.

Yes it does. Photoshop is a poster child for how NOT to do your memory management. We can only hope that they rewrite the next memory management in PS to support 64bit memory and get rid of the 2GB RAM limit. Adobe rolled their own scheme and it's getting a bit long in the tooth.

[~Shard~]

Will everybody please stop repeating this nonsense about this being the end of dual processor Macs.

Why it is nonsese? Do you know for a fact there will always be DP PowerMacs? Sound pretty confident in yourself...

A dual-core is nothing but two processors on one chip, there are still two processor cores in the computer. The OS and all programms will see two processors, you will even have a somewhat better performance than with a two processor computer.

You don't have to tell me this, I already recognized and appreciated it in my post. ;)

If you think that the wide public and maybe yourself would perceive a dual-core Powermac as somethink less powerful than a dual-processor Powermac, you should for once accept that the opposite is the case and also rest assured that Apple marketing would try its best to correct that impression.

I don't. You should read posts a little more closely, I never once even suggested this. In fact, I completely agree with you, and since I used to focus on microprocessor ansd ASIC design, I probably know the relative power and performance attributes even more in depth than you. :p A word of advice - people don't like it when you put words in their mouth, so I don't know why you were stating all of this by quoting my post... :confused:

[maxvamp]

While I know that many if not most apps out there are MULTI-THREADED, I will concede that there is a practical limit of usefulness for multiprocessing. After all, will a 16 way system really buy you much more power than a 4 way when doing Photoshop?

I doubt it.

My prediction is this:

The chips will not only be dual core, but will be SMT enabled. In the Powermacs, there is more of a need for space. In the XServe, there is more of a need for raw power.

I suspect, that the PowerMacs will be reintroduced as a single chip, single or dual core processor, and some extra drive bays / PCI (x?) slots thrown in. On the more server centric side, I could see the XServe being a dual chip single or dual core proc running at a lower speed and maintaining the same form factor that they have now.

This is my opinion based upon nothing.

Max.

[AidenShaw]

If they make Tiger highly threaded...
Imagine. You run Photoshop CS, Final Cut Pro 4, and Maya3D all at the same time.
Since Tiger is threaded, it allocates the programs-
Processor 1: OS X core system processes
Processor 2: Photoshop CS
Processor 3: Final Cut Pro 4
Processor 4: Maya3D

Even if Photoshop, Final Cut Pro, and Maya WEREN'T multithreaded, Tiger itself (being a multithreaded OS) would give each program it's own processor.

Result? Speed boost, baby!

Dedicating threads to CPUs works until you have as many threads as CPUs. Dynamic scheduling works much better - as threads request CPU they are scheduled on a free one. If FCP has 4 threads requesting CPU, it should get all 4 CPUs rather than being restricted to Processor 3 as you suggest.

Right now my XP system has 61 processes and 1133 threads....

[gerrycurl]

I can't wait for dual cores!

[G4-power]

The Red Wolf.

Exactly. I'm suprised at how many writers have commented about how hot the Dual 2.5Ghz chips are without having any empirical evidence regarding the heat max on this cpu. They are basically assuming that the chips are ungodly hot because of the LC system. Well I prefer to think of it as over engineering for the future.
...
Once 2.5Ghz owners get their computers sometime next week we'll have a better idea on how effective the LC system is and if it's suitable for cooling as many as 4 cores.

Yes, indeed. When the multiple forum threads about Rev. B PowerMacs were up, I remember some wattages for PPC 970 and 970fx. IIRC, 970 fx was about 24 watts, and 970 at about 70? ( I don't remember if they were typical or max or whatever) Anways, they didn't actually need LC in the original 2x2 GHz model, and let's say if the wattages were 25 with the fx and, say 40 (@2GHz) with the original chip, a 0.5 GHz boost on speed won't take wattage over 40 with the fx. So it won't need the LC either. But it's okay for me, and I think others too, it makes the computer way quieter, I hope.

As for the 970MP, PowerTune on it? That would help keeping heat generation even lower. So at 3 GHz a single chip dual-core machine wouldn't be a problem, and it shouldn't be at dual chip dual-core a problem either. Just waiting for next Spring to see what'll happen, and hopefully a G5 for my use too! Just hope we'll get a Rev. C or at least speedbump before December. I'd say the deadline is in March, another 12 months without upgrading isn't taking Apple far.
Just my opinion, what do you think?

[virividox]

6 more months from announcement till release and another 3 months till shipping

[Amdahl]

Very good, you have recognized the meaning of Amdahl's Law.

But I have to disagree on the SMT prediction. SMT (known as Hyper-Threading on Pentium 4) makes sense if you have a CPU with very few registers, like x86. PPC has at least 4 times as many registers, so the benefit of SMT would be close to zero. It usually has only a few percentage point benefit on Pentium 4/HT.

While I know that many if not most apps out there are MULTI-THREADED, I will concede that there is a practical limit of usefulness for multiprocessing. After all, will a 16 way system really buy you much more power than a 4 way when doing Photoshop?

I doubt it.

My prediction is this:

The chips will not only be dual core, but will be SMT enabled. In the Powermacs, there is more of a need for space. In the XServe, there is more of a need for raw power.

I suspect, that the PowerMacs will be reintroduced as a single chip, single or dual core processor, and some extra drive bays / PCI (x?) slots thrown in. On the more server centric side, I could see the XServe being a dual chip single or dual core proc running at a lower speed and maintaining the same form factor that they have now.

This is my opinion based upon nothing.

Max.

[nuckinfutz]

But I have to disagree on the SMT prediction. SMT (known as Hyper-Threading on Pentium 4) makes sense if you have a CPU with very few registers, like x86. PPC has at least 4 times as many registers, so the benefit of SMT would be close to zero. It usually has only a few percentage point benefit on Pentium 4/HT.

If this is true then how do you explain the SMT in the POWER5 which is still PowerPC ISA? While I agree with you the 970MP isn't going to have SMT I think we see SMT on the next major revision(G6 ?). But not for another 12-18 months.

[wdlove]

[QUOTE=nuckinfutz]If this is true then how do you explain the SMT in the POWER5 which is still PowerPC ISA? While I agree with you the 970MP isn't going to have SMP I think we see SMP on the next major revision(G6 ?). But not for another 12-18 months.[/QUOTE/)

If it is going to be that time frame for the Dual Core, that is a long wait. Then the problem of not wanting to get Rev. A of a brand new Mac. Just think that if we buy now we will get to enjoy a G5.

[manu chao]

Why it is nonsese? Do you know for a fact there will always be DP PowerMacs? Sound pretty confident in yourself...

Simply because a Mac having one chip with two processors (dual-core) is a two processor Mac for me, or a DP Mac to use your expression. So any Mac having a dual-core chip is a DP Mac by definition. That's my main point, the rest is just an explanation why I think that is the way you have to look at it.

One can certainly argue whether a chip with two processor cores counts as one processor (which happens to have two cores) or as two processors. But since it acts as if it had two processors (roughly the same processing power as two individual processors) I see no point in not calling it a dual-processor computer.



I don't. You should read posts a little more closely, I never once even suggested this.

Sorry, that part about the perception was a response to this post:
Yeah,
The trick will be selling these machines with one physical processor to a customer base that has come to believe that dual procs are the way to go for best performance.

I could start a big fight with people claiming that hybrid cars will replace all the gasoline driven cars in the medium term. Not that we might not be driving all hybrid cars in 30 years from now but any hybrid car (as it is conceived today) uses a gasoline (or diesel) engine and is therefore also a gasoline driven car.
The correct way to say this would be that "hybrids will replace all the CONVENTIONAL gasoline driven cars". Naturally, most people discussing this will understand that the first statement implied the word 'conventional' but to people not knowing exactly what a hybrid car is, the first statement could be misleading.

Maybe this helps you to understand my logic.

[Amdahl]

If this is true then how do you explain the SMT in the POWER5 which is still PowerPC ISA? While I agree with you the 970MP isn't going to have SMP I think we see SMP on the next major revision(G6 ?). But not for another 12-18 months.

SMT.

It is very unlikely it will occur in the 970MP, because the power4 did not have it; they wouldn't do a retrofit, so we agree there.

But I think it also quite unlikely to appear in a 'G6' either for a few reasons.

The POWER5 goes into systems (envisioned) to have way more memory bandwidth than a desktop system.

The POWER5 sells for a lot more, and is allowed to suck a lot more power for the intended use.

As I said previously, the benefit of SMT is not great on Power PC, thus, this next point really drives it home:

According to http://arstechnica.com/cpu/003/mpf-2003/mpf-2003-1.html, "The rest of the big changes to the core are related to the addition of SMT, which IBM claims increased the size of each core by 24%. (This increase in die size is another reason why an SMT-capable POWER5 derivative for the Mac is a ways off.)"

24% more chip size is not worth the limited benefits in a desktop.

[macsrus]

In addition, the PowerPC 970MP will be a significantly larger package than the existing PowerPC 970FX. Compared to the latter, which sports a die size of 66.2 square millimeters, the PowerPC 970MP will spread out over 154 square millimeters

Did anyone notice that the die size is more than double?

I expect that the cache size increase will take up some of the extra space...
But just maybe IBM will add an on die memory controller at the same time...

They did for the Power5
The POWER5 also supports DDR2, with an on-chip memory controller.

Maybe these will be added to the PPC too....

[~Shard~]

... ... Maybe this helps you to understand my logic.

Thanks for explaining yourself a little better, I slightly misunderstood your position as well. :cool:

[zelmo]

Powermac : Dual-Core-Dual-Processor @ 3 GHz :cool:
iMac / Powerbook : Dual-Core-Single-Processor :cool:
iBook / eMac : Actual G5 :cool:

????

Yes! Bring them up! Go IBM! Go apple! :p :p

I was thinking exactly the same thing, AmigoMac. :cool:

[nuckinfutz]

The POWER5 goes into systems (envisioned) to have way more memory bandwidth than a desktop system.

The POWER5 sells for a lot more, and is allowed to suck a lot more power for the intended use.

As I said previously, the benefit of SMT is not great on Power PC, thus, this next point really drives it home:

Thanks for the links. I love Stokes' writing on CPUs. Actually after reading this article I'm more convinced now that both dual core and SMT will be on the next major revision and here's why.

The rest of the big changes to the core are related to the addition of SMT, which IBM claims increased the size of each core by 24%. (This increase in die size is another reason why an SMT-capable POWER5 derivative for the Mac is a ways off.)

That automatically rules out the 154mm 970MP. Adding SMT puts the 970MP at nigh 185mm. No go.

Since increasing execution unit utilization is one of the main goals of SMT, the increase in issue bandwidth utilization as described above is going to be key, especially for the POWER5. I say this because in my first articles on the G5 I suggested that the POWER4/970's group-based dispatch scheme and issue queue configuration probably constrains issue flexibility and therefore execution unit utilization in some peculiar ways under certain worst-case scenarios (i.e., one execution unit of a pair is overloaded while the other is starved, due to a combination of poor instruction ordering on the part of the programmer/compiler and the group dispatch limitations)

Another gotcha in the 970. I expect IBM to rectify this not that it's really bad but you don't want one execution unit starved while the other is gorged. I'm figuring that IBM fixing this is a priority.

At this point, I could talk about the need for SMT in an Apple system, but I'll just leave off that sort of commentary for now and observe only that Apple's long-standing and ongoing affinity for SMP designs has resulted in two things: 1) a huge potential for wasted execution resources on the current crop of non-SMT-capable G5s and 2) a body of natively-developed and -ported applications that have been subjected to years of pressure to use multithreading wherever possible in order to wring the best performance out of Apple hardware. I think both of these factors will converge to make a SMT a significant improvement for the Mac platform.

Makes sense. Apple has been actively promoting the use of threading(more than one session at WWDC 2004 on threading). While the 970x processors do not efficiently feed the execution units I'm sure this will not happen with the mythical G6.

I see the plans going like this.

970MP systems ship Q2 2005. There are two refreshes.

IBM works on POWER5 derivative which replaces current execution units in 970x CPU, is also a dual core with SMT and I'm guessing adds an ondie memory controller. This is all at 65nm so we're about 100-120mm squared for the whole shebang.

Apple wants this because

1. SMT will help in Xserve systems
2. Dual Cores are standard now.
3. Ondie Mem controllers allow them to reduce the complexity of the system controllers. 980MP systems will have hypertransport 2.0 links betwen them.
4. Thread prioritization is right up Apples alley being heavy in multi media.

I won't be suprised to see Apple really hammer threading even more for WWDC 2005. It's amazing but large apps like Maya still don't support SMP.

I'm pretty jazzed about the POWER5 queing 10 instructions and dispatching 2 per cycle. IBM need only add this same function to the 980 and get the Altivec unit back to G4+ levels and we'll be sitting pretty by 2006.

[Sun Baked]

The marginal increase in size and heat on a Power5-UL would probably mean that they'd drop right into the current PowerMac G5 and provide a 35-40% performace boost (according to IBM).

The dual-core make a bit more sense for dual processor XServe G5s, DP iMac G5s, and a low-end PowerMac (which Apple seems to be moving towards, like the PB12).

IBM did say dual cores were the way of the future, though you "may" have to turn the clock down to keep them within the current heat/power envelope.

Using the 90nm tech, a 1.8/2.0 GHz PPC970MP would definitely make sense in a consumer machine -- especially since you kill all the extra stuff (Power Supply and support chips) needed for the second CPU module.

But then you hit marketing perception -- how would a dual 2.0GHz PPC970MP (4 core) at 90nm stack up to a dual 3.0GHz Power5-UL when there are still 3.x-4.xGHz Pentiums out there. When people still see MHz as the guide to which machine is faster.

Apple/IBM WILL need to compete against the coming dual x86 core CPUs (at lower clock speeds), but the MHz war is still winding down -- they may or may not need to sell both at the same time.

---

While there are technical benefits to going either way (fast single core, lower clock dual core) -- the choice for Apple may come down to which is easier to market/sell in the current MHz marketing climate.

[FFTT]

I am SO grateful for these forums. Honestly!

I went by the local Apple store yesterday trying to make up my mind
about finally getting a replacement for my beige G3 tower.

Although I really did like the PowerBooks. I felt that spending so much
for G4 technology this late in the game was setting myself up for disappointment in the all too near future.

The 2.5 G5 dually seemed to be a much safer long term bet, but spending
$3400 for the CPU alone sent me back to the think about it stage.
Adding a 20" display and a Protools LE interface and I would be in rather deep.

Reading this thread now makes me very glad to wait just a bit longer.

This is not just a CPU speed increase, this is a MAJOR difference.

[tych0]

Hey all

To those that say that multiple cores won't bring speed-ups to some applications, it might be good to point out that much of OSX already distributes some of its API calls on multiple processors. For example, Core Audio is multithreaded. Cocoa apps and games that call that library, for example will benefit from having multiple processors, even if they weren't written to take advantage of them.
As for Photoshop, is _has_ run well on 4-way systems in the past (4x Daystar Mac clone, back in the day), so it shouldn't be too much of a stretch to get it working very well on those systems.
Not what I would REALLY like is a return of the Cube... single-processor, dual core, maybe an inch larger to accomodate recent graphics cards... At the price point of a headless iMac... I can dream, can't I? :D

My 0.02$ worth...

[wdlove]

Though with the G5 being a new technology the recommendation was to wait for Rev. B. It seems that there is possibility dual core is due to be released at MWSF '05. If so that means that the Rev. B, would mean waiting a least another year. :eek:

[jared_kipe]

Dual G5 with one CPUs, or Quad G5 with two CPUs.

[macsrus]

Though with the G5 being a new technology the recommendation was to wait for Rev. B. It seems that there is possibility dual core is due to be released at MWSF '05. If so that means that the Rev. B, would mean waiting a least another year. :eek:

I can only hope to see a quad processor in the xserve running at 3 GHZ or better by the end of 2 Quarter next year....
If so I can see us buying between 2000 to 2500 of em...

[macsrus]

Note that IBM/Apple needs to double the amount of memory bandwidth per chip in order to maintain performance.

If the dual-core chips shares one memory bus (for example, if the dual-core chip (or 2) are dropped into the current G5 architecture) then the memory bandwidth *per CPU* will be halved.

That's gotta hurt.

They'll need to double the speed of the bus, double the number of busses, or double the width of the bus to maintain the current balance of memory bandwidth per CPU.

There is a very good chance that the dual core will get a memory controller on die with it

[wdlove]

I can only hope to see a quad processor in the xserve running at 3 GHZ or better by the end of 2 Quarter next year....
If so I can see us buying between 2000 to 2500 of em...

You must be working for a rather large business. So you think that the dual core technology will come out on the Xserve first?