[clarkgoble]

If nothing else, it puts both IBM and Motorola on notice that they can't take Apple for granted. That may spur their development cycle. It also tells shareholders that he is prepared if IBM and Motorola can't deliver. I suspect that if he *had* to switch to the x86 there still would be quite a bit of trouble. As many have mentioned, the Darwin team says it hasn't been optimized for x86 - at least not the public distros. So even if Steve-O decided to switch AMD or (more likely) Intel it would still be a long time off and he'd want full working OSX Betas delivered around more than six months before delivery.

So those who see AMD or Intel as the savior over the 970, well even if there was a switch it'd be much further off than the 970 is.

[eddiecatflap]

..motorola...what a joke.

[kupan787]

Originally posted by pliny:
Most if not all of these Intel-AMD in Apple boxes rumors have at their source Steve Jobs himself. He started it by saying there is an OS X x86 build running at 1 Infinite Loop and then said Apple likes to have options. I wonder what his motivation for this is? (I may as well ask why the speed of light is a constant.) What benefit is there to Apple by having encouraged all this speculation?

Link please.

I think you are misquoting something, as Jobs never said they have OS X running on X86. That was MOSR and Marklar (or whatever it was called at the time they posted their story).

However when asked at a shareholder meeting if Apple was considering a move to X86, Jobs basicly laughed and said no. I will try and find the quicktime playback of the meeting, but I can't seem to find it on Apple's site. The meeting was on April 24th, 2002 if I am not mistaken.

Now what this means is that there wont be a move anytime soon. Why? Because if they did move, Apple would have lied to its sahreholders, and would be in such a mess it would be lights out. Lawsuits galore I tell you!

So basicly we have proof that Apple will not be going to X86, yet stupid rumors keep coming up about this, and people jump all over it.

The whole "..and we like to have options.." quote was taken out of context. If you read the whole article (cnet, cnn, I don't remember), and the surrounding sentences you see that he wasn't talking about X86 at all. I think this was discussed to death on AppleInsider months ago, but people seem to forget that, and only remember the options part, and like to say it means something other than what it did.

[Ken_F2]

Hmmm....

[passmaster16]

Here's a writeup in reference to SJ's statement about how Apple will have options after the transition to OS X:

http://www.overclockers.com/tips062/


More here:
http://www.osnews.com/story.php?news_id=1393

[pliny]

Originally posted by kupan787:


Link please.

I got it from eweek back in August.
(you're right about SJ not saying, it's an unnamed source.)

"As Apple Computer Inc. draws up its game plan for the CPUs that will power its future generations of Mac hardware, the company is holding an ace in the hole: a feature-complete version of Mac OS X running atop the x86 architecture.

According to sources, the Cupertino, Calif., Mac maker has been working steadily on maintaining current, PC-compatible builds of its Unix-based OS. The project (code-named Marklar, a reference to the race of aliens on the "South Park" cartoons) has been ongoing inside Apple since the early days of its transition to the Unix-based Mac OS X in the late '90s.

Sources said more than a dozen software engineers are tasked to Marklar, and the company's mainstream Mac OS X team is regularly asked to modify code to address bugs that crop up when compiling the OS for x86. Build numbers keep pace with those of their pre-release PowerPC counterparts; for example, Apple is internally running a complete, x86-compatible version of Jaguar, a k a Mac OS X 10.2, which shipped last week.

Apple did not return calls requesting comment.

But a switch to Intel or Advanced Micro Devices Inc. processors is probably not in the cards for tomorrow's Macs, sources said. Such a move would require a massive revision of Apple's closed hardware architecture and a fundamental rethinking of its business model, which is founded on tight integration between its proprietary system software and hardware. Apple would have to also coax most of its third-party developers to rewrite their applications from the ground up in the company's Cocoa application environment. (Most major vendors have instead tuned their applications to Carbon, a set of Mac OS X-compatible APIs originally culled from the classic Mac OS and rooted in the PowerPC architecture.)

Nevertheless, Marklar has apparently gained strategic relevance in recent months, as Apple's relationship with Motorola has grown strained and Apple looks to alternative chip makers.

Apple has reportedly been dissatisfied with the slow rate of Motorola's PowerPC development after committing to the PowerPC G4 as the centerpiece of its current desktops and professional laptop systems. The Power Mac G4 systems Apple unveiled in August topped off with a dual-1.25GHz system, a disappointing increase from the dual-GHz top model released in January. Meanwhile, users have debated whether the DDR support in the new systems is fully exploited by the G4 processors Motorola was able to provide.

The likeliest solution to the Motorola impasse, sources said: A desktop version of the 64-bit Power4 server chip in the works from IBM, which co-developed the PowerPC platform alongside Motorola and Apple and has provided CPUs for a variety of Macs. Sources told eWEEK that Apple and IBM are collaborating closely to equip the Power4 with the Altivec vector-processing capabilities built into the PowerPC G4. IBM is expected to discuss its new CPU at October's Microprocessor Forum.

As it weighs the future of the Mac as a PowerPC platform, Marklar offers a relatively low-cost way of keeping the company's options open. "It's a hedge," one observer said. "It's a small price to pay to make sure Apple has a fallback plan."

"Steve [Jobs] has said Mac OS X is the OS for the next 15 years," another source said. "Marklar is a way of making sure that's true."

Jobs himself has hinted that Apple won't be constrained by the PowerPC alliance if better options present themselves. The Apple CEO renewed speculation about Apple's hardware future with remarks he made at a July meeting with analysts. "Between Motorola and IBM, the roadmap looks pretty decent," Jobs said. However, he said that after early 2003 (when he forecast the transition to Mac OS X would be complete), the company will re-examine its processor partnerships. "We'll have options, and we like to have options"

At the company's shareholder meeting in April, however, Jobs asserted that Apple has "no plans" for a switch to Intel. When a shareholder argued that a move could be beneficial to the company, Jobs replied, "That is an opinion."

Despite its current PowerPC pedigree, Mac OS X's roots tap Intel hardware. In December 1996, Apple acquired NeXT Software Inc. and its Intel-compatible OpenStep operating system. Under the company's "Rhapsody" OS strategy, it planned to base the next-generation Mac OS on OpenStep, shipping an Intel version to provide a cross-platform development environment. While developer previews of Rhapsody for Intel were released, it was never shipped to customers and quietly left the limelight as Apple's software strategy was refined into today's OS X.

Apple's current efforts come nearly a decade after the company grounded its "Star Trek" program, a collaboration with Novell to develop the Mac's System 7 on Intel microprocessors. While a working prototype was put together in just three months, interest from PC vendors never materialized. In 1993 the project was rolled into Apple's Advanced Technology Group before it fell victim to budget cuts."

eweek

[suhail]

What happened to monopoly laws?

Is this what Microsoft meant when they had their DOJ victory "This sends out a clear message to us and the whole industry" something like that.

I guess the world is changing again!

[mac-at-kearsarge]

Well now that the notion of "Athlon in Apple" is over (for now), as Tuesday has come and gone with no "big announcement".


Yet wait, didn't I read reports that SJ was there this year (at Comdex)? I'm willing to bet that at the last minuet, of the last hour, steve will jump up on stage just as they're about to take everything down and say "Oh wait, and one more thing..." R O T F L M A O !!!!!!!!!!!!

[milhous]

I still don't understand why Apple ignores shows like Comdex or the Consumer Electronics Show. Even if their budget is tight and the economy is slow, the can still just rent a small booth with a G4 demo machine and some product leaflets. I'm sure there are a lot of PeeCee users out there who know nothing about Apple or what Apple is. Even to this day, people still equate Macs as computers for graphic artists. Apple needs to fight this misconception. The switch ads might be helping, but more needs to be done.

[clarkgoble]

Comdex is pretty much dead. I will say a CES presence makes more sense given the iPod and rumors that Apple may be unleashing more consumer products to help it diversify. However in my opinion the utility of most of these conventions is fairly limited for Apple. By and large they just don't have the effect on the industry that they did in the 80's. They are good for certain kinds of companies, but after going to a lot of them, I'm not sure they are worth the effort.

Some more targeted conventions and trade shows may help Apple though. The publishing and graphics conventions would be great, especially if Apple gets some hardware than can show a real significant improvement over equivalent PCs.

[vmpaul]

Totally naive technology question:

Does a 64-bit sytem just double the speed of current 32-bit systems or is it more expotential?

Sorry, I haven't seen this discussed on any board before.

[clarkgoble]

Most applications don't need or use the 64-bit aspects of the chip. By and large you can ignore that. What counts more are the other aspects of the chip.

The kind of applications that need 64bit are largely certain limited kinds of scientific or mathematical applications and then databases. Even in science floating point counts far more, although some applications may need the larger memory size. (Although there are enough ways around that problem that I don't think it is key)

The big problem is that some 64 bit chips actually run 32 bit operations slower. That was one plus for the Hammer - it could run existing Linux and Windows programs purportedly faster than equivalent 64 bit offerings from Intel. To take advantage of the chip you really needed a recompile.

I could be wrong, but I don't believe that the 970 suffers these problems. The Power4 platform was always moving towards 64 bit anyway. I wouldn't be surprised if many aspects of OSX are already 64bit ready and only need a recompile and a little tweaking. Your existing apps likely won't be optimized though and probably won't see much benefit. Which is not to say that the improved speed of the 970 won't be significant. Just that the 64 bit nature of the chip is somewhat beside the point.

[Superchicken]

the extra 32 bits... is so that while they can say they have 2Ghz on us... we can say we have 32 bits on them.. and continue our pissing contest fair and square

[clarkgoble]

OK, I know this rumor has more or less died already. But here is an interview with AMD that pretty well finishes it off.

http://www.eweek.com/article2/0,3959,741273,00.asp

To quote:


First of all, I have no indication that Apple is even considering what we make. I've heard rumors going around. But you know it would be interesting because at some point in time if Apple is going to do a 64-bit version, they're going to face the decision, what do they do for it? I cannot picture Apple putting an Itanium in their stuff. So I think if they're going to do that they're going to figure out some way to get a PowerPC version of that. Or they're going to have to consider one of the alternatives we offer and see there's many more than that.

[mac-at-kearsarge]

THANK YOU DARKGOBLE!!! I have friend here at school (mega PC user)who keeps "trying" to convince me Apple is going to AMD clawhammer. I've tried to explain it to him (wouldn't listen), reffered him to this thread (didn't want to read what we "little Mac Freaks" had to say. So hopefully this will get the message across that IT AIN'T GONNNA HAPPEN!!!!!

[passmaster16]

While I think it's highly unlikely they'll ever switch to AMD, I think that article means nothing. Seriously guys, Ruiz isn't allowed to talk about Apple either way. He admitted that. So the question asked by eWeek was one that would never get a honest answer. Until he comes out and says "No, we have no relationship with Apple, and we are not going to make chips for them" then I will believe this Apple/AMD talk is offically over. In the meantime, I'm just going to sit back and see what happens.

[kupan787]

Originally posted by passmaster16:
While I think it's highly unlikely they'll ever switch to AMD, I think that article means nothing. Seriously guys, Ruiz isn't allowed to talk about Apple either way. He admitted that. So the question asked by eWeek was one that would never get a honest answer. Until he comes out and says "No, we have no relationship with Apple, and we are not going to make chips for them" then I will believe this Apple/AMD talk is offically over. In the meantime, I'm just going to sit back and see what happens.

Why would he not be allowed to come out and say they aren't workign with Apple? Ruix said this, but it didn't make sense to me. There is no NDA or anything if they have no agreement, so he should be able to come out and say "We are not making Apple any chips".

Maybe he doesn't want to say that, because he wants peopel to think AMD isn't ogign to die off in the next few years, as they might have a big backer pretty soon. Who knows.

[Avon]

Man, this thread needs a serious visit from Captain Obvious. What's the only chip in the world that is 64 bit PPC and 32 bit capable? The IBM PPC 970.

Why in the world would Apple switch to x86? iBooks would melt, software developers would cry, and macs will go to the dark side.

[clarkgoble]

There actually were good reasons to examine the x86. The problem is that for both IBM and Motorola the main market for the PPC isn't the desktop but embedded systems. What some forget is that even if Apple went to AMD they'd still be a fairly small client relative to the overall sales of AMD. Of course given AMD's financial troubles, I think that their President would love Apple as a client. The interview whose link I posted seems to indicate this. However there are oft mentioned difficulties and performance issues with moving to an AMD x86 chip.

The difficulty with the "obvious" choice is that there is no guarantee that IBM will keep the 970 competitive with Intel's offerings. In a way IBM would like to be competitive but it really doesn't lose much if they aren't. On the other hand because of the competition of AMD and Intel, there was that drive to improve the chips faster.

What many worry is that even though OSX is a far superior OS to either Linux or XP, it might continue to lose market share due to uncompetitive hardware.

Now I both agree and disagree with this reasoning. I agree in that when the 970 comes out, it will at best be in the middle of the pack in terms of equivalent PC desktops. That means that on price/performance ratios Apple will still be hurting. And make no mistake, the first and second quarters of 2003 will really hurt in terms of performance comparisons.

My disagreement arises in that for the vast majority of computer uses, current speeds are sufficient. That is why so many businesses are dramatically slowing their computer purchases. They don't need to upgrade. What they have works quite well.

However, where this fails, is the tradition place of Macs in illustration, publishing and other media markets. Those are the very markets that do still need more speed. For rendering, for video, and even for complex Illustrator or Photoshop work, speed still makes a difference. Those have also traditionally been markets where the Mac has been strong. I don't think Apple could fall too far behind and still keep up. Take one market: publishing. It wouldn't be that big a move if you are already moving from Quark to Adobe to move from OS9 to XP instead of OSX. If you throw in the speed and price difference. . .

So Apple does need fast hardware. And really, it needs it last week not in six months.

[Nebrie]

Originally posted by Scotttheking:
OH PLEASE LET IT BE DELL SHIPPING ATHLON SYSTEMS.
Company went all dell, and I'm fighting for an athlon system.

I wouldn't be surprised if there is a low cost render system coming. Apple boxes, AMD chips.
We shall see.
Until then, this stays in powermac

Look around, AMD chips no longer have any real price advantage over Intel chips.

[clarkgoble]

One other thing, to the guy who said nothing that the head of AMD could be trusted. I don't think that is entirely true. For one, what would Apple and AMD have to lose by releasing the information? Second, he doesn't really use the "weasel" language you'd expect. If he was going to weasel he could say something that would hint at a possible relationship while still remaining well within his NDA. That would even help AMD as they are desperate for a large client.

While Apple is small relative to the number of chips AMD ships, they have a lot of influence that is greater than their size. Further it is a market that could conceivably grow. Likely not beyond 15%, but 15% of the entire computer market is nothing to sneeze at.

So given the way the interview went, I think AMD can officially be considered a non-event. Perhaps Apple might look at AMD if IBM can't produce sufficient yields of the 970. There's even a small chance Apple might still come up with either a dual system (Hammer and 970) if that is achievable. But I don't think we'll see a real x86 OSX anytime soon.

Further even if they went x86, I'm not sure it would be AMD. If you are going x86, why not go Intel? It is just as reasonable as AMD and further you know that Intel, unlike AMD, will keep pushing the desktop.

[Avon]

I agree apple needs faster processors, or at least more slower ones. But switching to x86 is not a good solution.

x86 is an overly complex ciscs solution over a risc core. Intell and amd keep dumping tons of money into this technology to keep it at the top of the pack.

Yes it works. It works fast. But its far form elegant, and I still think its days are numbered. My main concern is power consumption. The power requirements for an underclocked ppc 970 are amazing. This will keep the mac at the top of the laptop market. A 64 bit processor running at over 1 ghz in a laptop.

If IBM can keep developing the PPC 970, it will satisfy most peoples needs. I am also a big fan of multiple processors. Mac OS X handles them very niceley indeed. When the 970 comes out I really hope apple keeps putting multiple processors in the powermacs.

[clarkgoble]

In this day and age debates about CISC vs. RISC are largely pointless. All x86 styled systems have a lot of RISC to their nature.

The real issue is how fast the systems go and how much faster particular designs can go in terms of clock rate.

Intel surprised everyone with how they've pushed the x86 basic design. (Recognizing that they did so by adopting a lot of RISC design decisions) They really ought to be commended for what they've done. At this point their desktop 32 bit chips are cheaper and faster than most other equivalents. About the best one can say is that AMD made a very good chip as well which is sold cheaper for certain situations.

With 64 bit things are different and AMD may win the 64 bit desktop war, since Intel's offerings don't really make sense for the desktop. However one could simultaneously argue that 64 bits for the desktop really isn't needed and that the bigger issue is speed.

IBM appears to have a great chip in the 970 that is much more along their traditional RISC design. But while it looks like it can increase in speed nicely, it will still run slower than what Intel has. (In terms of actual calculations, that is) This really isn't a RISC vs. CISC debate though and relates to many other design decisions along with the fact that it is only coming out now while Intel's recent offerings came out earlier.

As for "elegance" I think that is a value too often raised higher than it ought. At this stage I see no reason why the x86 platform won't just keep getting faster and faster. Even at this stage a lot of stuff is effectively emulated by RISC circuitry, so who cares what is going on internally? The only thing I can think of that might be a problem with x86 based chips are the many rules on registers and (with AMD) the far fewer registers than the 970.

I'll add that the 970 will likely be behind the curve in terms of speed when released. That means Apple will almost certainly continue supplying dual processor systems. Unfortunately that only helps with multiprocessing but not really the speed of your main thread. So it only helps to a point. (As all dual G4 users can attest)

[Vanquish]

All apps would become useless when Apple would switch to x86, right ? (espcially Carbon) But couldn't Apple develop a "Patching-application" (iPatch ) that patches existing application to run on x86 (or something else) ?

(Maybe this sounds extremly stupid, but I know nothing about this stuff, I'm just a simple designer)

[TimmyDee51]

Originally posted by clarkgoble:
The only thing I can think of that might be a problem with x86 based chips are the many rules on registers and (with AMD) the far fewer registers than the 970.

I do think that elegance of design is a big issue and I think you've struck at the heart of it there. It's those same registers that will end up being the real problem in the x86 architecture. Who knows what kind of instructions processors will need to take in the future and I think IBM has realized this. Emulation can only get you so far. As we have seen with 680x0 and Classic in Apple, it is at best a stopgap measure, and that's just emulation at the software level. I can't imagine what problems emulation at the processor level will lead to.

In any case, the 970 will be a good chip, raw numbers or not. There's a reason why everyone's wetting their pants about the little brother of the Power4.

[Metzen]

Originally posted by clarkgoble:
There actually were good reasons to examine the x86. The problem is that for both IBM and Motorola the main market for the PPC isn't the desktop but embedded systems.

The PPC 970 isn't targetted towards embedded systems.

Originally posted by clarkgoble:
However there are oft mentioned difficulties and performance issues with moving to an AMD x86 chip.

I'd say any x86 chip, not just AMD.

Originally posted by clarkgoble:
The difficulty with the "obvious" choice is that there is no guarantee that IBM will keep the 970 competitive with Intel's offerings. In a way IBM would like to be competitive but it really doesn't lose much if they aren't. On the other hand because of the competition of AMD and Intel, there was that drive to improve the chips faster.

IBM/Motorola has as much to loose as Intel or AMD. Supply and demand dictate, lower demand, higher price, more profit. I see the PPC business as a very lucrative one.

Originally posted by clarkgoble:
Now I both agree and disagree with this reasoning. I agree in that when the 970 comes out, it will at best be in the middle of the pack in terms of equivalent PC desktops. That means that on price/performance ratios Apple will still be hurting. And make no mistake, the first and second quarters of 2003 will really hurt in terms of performance comparisons.

Intel doesn't plan to increase there speeds much because of the lack of competition from AMD. They intend to gouge the customer for all they can as they finesse their manufacturing process to get the most bang for their buck. A recent Intel roadmap put the P4 at 3.2GHz by mid-year 2003. If the PPC 970 comes out at 1.8GHz mid-year next year, I'd say that the gap has been closed quite well, especially when you take into account the PPC efficency vs. x86.

Originally posted by clarkgoble:
So Apple does need fast hardware. And really, it needs it last week not in six months.

This will be said for a long time while Intel is a GHz+ ahead of Apple.

But, I still contend that a 2x1.25GHz G4 is still sufficient to keep up with the P4.

Originally posted by clarkgoble:
In this day and age debates about CISC vs. RISC are largely pointless. All x86 styled systems have a lot of RISC to their nature.

The real issue is how fast the systems go and how much faster particular designs can go in terms of clock rate.

1.3GHz Power4 trounces 3.0GHz P4. 800MHz Itanium kicks the 2.13GHz AMD Athlon. Clockrate is mostly irrelevant.

Originally posted by clarkgoble:
IBM appears to have a great chip in the 970 that is much more along their traditional RISC design. But while it looks like it can increase in speed nicely, it will still run slower than what Intel has. (In terms of actual calculations, that is) This really isn't a RISC vs. CISC debate though and relates to many other design decisions along with the fact that it is only coming out now while Intel's recent offerings came out earlier.

Not really... Sounds like you only care about clock speed. In terms of actual calculations, I have no doubts the PPC970 @ 1.8GHz will beat a 3.0GHz P4. But then again, this is all based on the type of calculations. At RC5, we see a 1.0GHz G4 beating a 3.0GHz P4. Again, clock, which you like to stress, is not something to get too worked up about.

Originally posted by clarkgoble:
I'll add that the 970 will likely be behind the curve in terms of speed when released. That means Apple will almost certainly continue supplying dual processor systems. Unfortunately that only helps with multiprocessing but not really the speed of your main thread. So it only helps to a point. (As all dual G4 users can attest)

I beg to differ. The 970 will out perform the P4 in terms of performance (not MHz... yet).

The last time IBM put any real muscle improving the PPC was with the initial G3. At that time, the G3 was roughly two times faster than the PII.

[clarkgoble]

The PPC 970 isn't targetted towards embedded systems.

Yes, I know that. But that's not my point. The point is that the volume of chips Apple buys makes them a smaller market for the chip makers. For instance the G3, while used by Apple, is also targeted at other markets. Motorola, for all their incompetence, also realizes that the embedded market is a bit safer. Especially given financial worries with Apple the prior five years. (Which is partially why I suspect they didn't invest enough R&D)

Don't get me wrong. I'm excited about the 970. However lets not deceive ourselves in how we view Apple's place in the marketplace.

I see the PPC business as a very lucrative one.

In general or with Apple in particular? After all lots of things use PPCs other than Macintoshes. I think that if Apple can turn around the desktop PPC market could become lucrative. However there is a bit of a catch-22. Apple is somewhat limited by speed, but speed is determined in part by R&D and that is dictated largely by Apple's success.

Intel doesn't plan to increase there speeds much because of the lack of competition from AMD.

That is true now that AMD has stated they aren't focusing on the desktop as much. And even then, according to discussions I've seen, it sounds like more they are scaling back their R&D. But I'm not sure I'd say there is a lack of competition from AMD. Unless you mean marketing competition. But that has just really changed the last six months. (IMO) Intel won the battle. But you are right, that does mean that the rapid advance in x86 chips will slow down. Further the fact that for most uses computers are "fast enough" means that not as many are selling which changes the market somewhat. All this may change when we fully pull out of the recession.

A recent Intel roadmap put the P4 at 3.2GHz by mid-year 2003. If the PPC 970 comes out at 1.8GHz mid-year next year, I'd say that the gap has been closed quite well, especially when you take into account the PPC efficency vs. x86.

We'll see. It'll be interesting to see how full 970 systems compare to P4 systems next summer. For the Mac, a lot will depend upon Apple's support chips. I've read that those are behind schedule. So we'll see what happens.

But, I still contend that a 2x1.25GHz G4 is still sufficient to keep up with the P4.

It really depends upon your application. The problem is that for applications that aren't multithreaded well a 2x1.25GHz performs slightly worse than a 1.25GHz system.

1.3GHz Power4 trounces 3.0GHz P4. 800MHz Itanium kicks the 2.13GHz AMD Athlon. Clockrate is mostly irrelevant.

By "how much faster particular designs can go in terms of clock rate," I meant how much a particular architecture can speed up. I didn't mean in terms of comparison between different architectures. In other words, a 2 GHz P4 is slower than a 3 GHz P4. If the P4 tops out at 3.8 GHz then that is a limit. The issue is, for a given architecture, what kind of speed improvements can we expect with its design and with the manufacturing processes. Obviously even that isn't everything as support chips and motherboard design impact performance heavily.

The 970 will out perform the P4 in terms of performance (not MHz... yet).

From what basis do you argue this? I wish it were true, but I just don't see the evidence.

[Metzen]

Originally posted by clarkgoble:
The PPC 970 isn't targetted towards embedded systems.

Yes, I know that. But that's not my point. The point is that the volume of chips Apple buys makes them a smaller market for the chip makers. For instance the G3, while used by Apple, is also targeted at other markets. Motorola, for all their incompetence, also realizes that the embedded market is a bit safer. Especially given financial worries with Apple the prior five years. (Which is partially why I suspect they didn't invest enough R&D)

Don't get me wrong. I'm excited about the 970. However lets not deceive ourselves in how we view Apple's place in the marketplace.

Motorola's G3 topped out at around 500MHz. The G3 currently being used by Apple is from IBM, not Motorola.

Originally posted by clarkgoble:
1.3GHz Power4 trounces 3.0GHz P4. 800MHz Itanium kicks the 2.13GHz AMD Athlon. Clockrate is mostly irrelevant.

By "how much faster particular designs can go in terms of clock rate," I meant how much a particular architecture can speed up. I didn't mean in terms of comparison between different architectures. In other words, a 2 GHz P4 is slower than a 3 GHz P4. If the P4 tops out at 3.8 GHz then that is a limit. The issue is, for a given architecture, what kind of speed improvements can we expect with its design and with the manufacturing processes. Obviously even that isn't everything as support chips and motherboard design impact performance heavily.

When your comparing architectures, don't specify that "particular designs" as that really can mean anything from AMD vs Apple or AMD vs Intel. But you caught my confusion, thanks for clearing it up

Originally posted by clarkgoble:
I see the PPC business as a very lucrative one.

In general or with Apple in particular? After all lots of things use PPCs other than Macintoshes. I think that if Apple can turn around the desktop PPC market could become lucrative. However there is a bit of a catch-22. Apple is somewhat limited by speed, but speed is determined in part by R&D and that is dictated largely by Apple's success.

Cost and Volume. Everything is based around cost and volume. If the volume it high, cost can be low, if the volume is low, cost will be high. The G4 is definitely not cheaper than a equivalent MHz PIII or Athlon... Why? Volume.

Originally posted by clarkgoble:
The 970 will out perform the P4 in terms of performance (not MHz... yet).

From what basis do you argue this? I wish it were true, but I just don't see the evidence.

The 970 is essentially a derivative of a Power4, and the Power4 spanks the P4...

From what do you make your claim?

[clarkgoble]

The 970 is essentially a derivative of a Power4, and the Power4 spanks the P4... From what do you make your claim?

For SPECint2000 the 1.3 GHz Power4 with a single core gets a 804. A 2.8 GHz P4 gets a 970. With the latest P4 that should improve considerably. IBM is predicting that the 970 at 1.8GHz is 937. That's comparable to the previous generation P4 at 2.8 GHz.

For SPECfp2000 the P4 gets a 938 while the 970 gets an estimated 1051 and the Power4 a 1202.

This tells me both that the Power4 can't be used to predict the speed of the 970 and that the Power4 isn't "spanked" by the P4, although it does do floating point considerably faster. The 970, however, is expected to do floating point on par with the P4.

http://www.ricford.net/ppcperf.html

Now one can argue about effectiveness of compilers and so forth. Admittedly Intel helps produce some very nice compilers for their chips. Gcc for the G4 is still a little behind, from what I've heard. Further we don't know how well the first generation 970 compilers will work.

The general theme though, even according to IBM, is that the chip is about on par with the Pentium. However the Pentium is out now while the 970 is six months from now. Even keeping in mind that Intel is going to slow down their releases, this means that the 970 has to play catchup.

Motorola's G3 topped out at around 500MHz. The G3 currently being used by Apple is from IBM, not Motorola.

Yes. I don't see how this addresses my point. Both IBM and Motorola don't target the G3 primarily at the desktop. IBM did impressive work with producing a scalable version of the architecture. But Apple wasn't driving it all.

[Amorph]

Originally posted by clarkgoble:

Now one can argue about effectiveness of compilers and so forth. Admittedly Intel helps produce some very nice compilers for their chips. Gcc for the G4 is still a little behind, from what I've heard. Further we don't know how well the first generation 970 compilers will work.[/qb]

This is true, and since the (pitiful) SPEC marks floating around for the MPC7455 were compiled with GCC 2.95 - hardly an optimal compiler - I'd like to see SPEC marks for the P4 compiled with MSVC++ set to default settings and running under Windows XP. That will probably do a lousy job of benchmarking the processor's capabilities, but it will give a better measure of the performance available to the applications and games that people actually use. And it'll be somewhat fairer to the poor G4.

[qb]The general theme though, even according to IBM, is that the chip is about on par with the Pentium. However the Pentium is out now while the 970 is six months from now. Even keeping in mind that Intel is going to slow down their releases, this means that the 970 has to play catchup.[/qb]

Only if you compare single core to single core. The PPC 970 should be cheaper (since it's not that big, and since Apple will probably get sweet prices), and it will be much cooler and much more SMP friendly. If Apple keeps up the all-SMP strategy, and extends it to four- and eight- processor systems, they can soundly beat the P4. Single-threaded apps might run better on the P4, but then multi-threaded apps are becoming common on OS X, and the SMP system could run two (or four, or eight, or 32) such applications each on their own processor - multitasking heaven.

[qb]Yes. I don't see how this addresses my point. Both IBM and Motorola don't target the G3 primarily at the desktop. IBM did impressive work with producing a scalable version of the architecture. But Apple wasn't driving it all. [/B]

Apple drove the development of the original G3, as I recollect. It was intended as a replacement for the 603, and designed to get around the limitations of Apple's budget motherboards (read: really slow busses). It usurped the role of PowerMac processor only when Apple found out exactly how well it got around those limitations, and so the first PowerMac G3s shipped with a motherboard originally intended for the successor to the Performas! (Does "Gossamer" sound like the name of a high performance board to anyone?) Of course, since it was a collaboration with Mot, there was an eye toward using it in the embedded market eventually. That's how Mot does things.

Note that Apple has done a fair job of turning the PPC's late focus on embedded systems into a strength: All of their most-loved systems now could be considered embedded uses, even though their PCs, because of the design constraints. The only laggard is the tower line, which doesn't play to the strengths of an embedded architecture. But even there, the G4 is doing better than some people give it credit for. Not that I'll mind the shift to the 970 one little bit.

[Metzen]

Originally posted by clarkgoble:
The 970 is essentially a derivative of a Power4, and the Power4 spanks the P4... From what do you make your claim?

For SPECint2000 the 1.3 GHz Power4 with a single core gets a 804. A 2.8 GHz P4 gets a 970. With the latest P4 that should improve considerably. IBM is predicting that the 970 at 1.8GHz is 937. That's comparable to the previous generation P4 at 2.8 GHz.

For SPECfp2000 the P4 gets a 938 while the 970 gets an estimated 1051 and the Power4 a 1202.

Ok, my computer crashed yesterday while I was finishing my reply, so I'm going to do this again...

If you can choose your benchmark, I'll get to choose mine...

RC5 -- 18MKeys/s for PPC970 @ 1.8GHz
RC5 -- 5.5MKeys/s for Pentium 4 @ 2.8GHz

Spanked? You betcha. As it is the PowerPC is a more effiecent processor, and the PPC970 is expected to scale extremely well. The G4 scores for RC5 are right around the same as the PPC970, does that mean the 970 will be as fast(slow?) as the G4... You bet it won't.

Originally posted by clarkgoble:
The general theme though, even according to IBM, is that the chip is about on par with the Pentium.

Where does IBM say that?

Originally posted by clarkgoble:
Motorola's G3 topped out at around 500MHz. The G3 currently being used by Apple is from IBM, not Motorola.

Yes. I don't see how this addresses my point. Both IBM and Motorola don't target the G3 primarily at the desktop. IBM did impressive work with producing a scalable version of the architecture. But Apple wasn't driving it all.

As I had read it, it appeared to me that you were implying that the G3 was being supplied to Apple from Motorola, not IBM who has other targets besides embedded systems.

[Ken_F2]

This is true, and since the (pitiful) SPEC marks floating around for the MPC7455 were compiled with GCC 2.95 - hardly an optimal compiler -

IBM's 970 results obviously weren't obtained with GCC 2.95.
[quote]There is virtually no performance difference on the PC under Windows between the Intel 6.0 and latest MS VC++ compilers. A few days ago, Intel released its 7.0 compiler which it says offers improved performance on P4 systems, but that is too new for there to be any results or benchmarks comparisons as yet.

That will probably do a lousy job of benchmarking the processor's capabilities, but it will give a better measure of the performance available to the applications and games that people actually use. And it'll be somewhat fairer to the poor G4.

There isn't much difference (+/- 2%). Console, cross-platform tests show that version 6.0 of Intel's compiler for Windows offers code that is comparable in performance to version 6.0 of the Intel compiler for UNIX/Linux. Not to forget, Intel and Microsoft do work together, and have a cross-licensing agreement, which may explain some of the similarities. Certainly, we know from tests that Intel's 6.0 compiler is no faster than the latest MS VC++ compiler under Windows used for 90% of commercial applications. Intel's newest 7.0 compiler, released a few days ago, may be faster for P4 systems, but there aren't any SPEC results that use that compiler yet.

Single-threaded apps might run better on the P4, but then multi-threaded apps are becoming common on OS X, and the SMP system could run two (or four, or eight, or 32) such applications each on their own processor - multitasking heaven

Don't forget Intel's Hyperthreading (SMT) implementation, which uses DEC Alpha-inspired technology to improve performance in multhreaded applications by up to 40%. With Hyperthreading, one processor appears as two to the OS. Some examples of multithreaded results and the improvement on a single P4 3.06 with SMT enabled compared to the same test with SMT disabled:

3DS Max 4.26 (Aceshardware) - 16% improvement with SMT
Adobe Photoshop 7.0 (Hardocp) - 26% improvement with SMT
Adobe Premiere (Aceshardware) - 10.6% improvement with SMT
DIVX 5.0 encoding (Aceshardware) - 20% improvement with SMT
Kribibench (Aceshardware) - 32% improvement with SMT
GoGoMP3 2.39 MP3 encoding (x86-secret.com) - 25% improvement with SMT
Newtek Lightwave 7.5 (Aceshardware) - 23% improvement with SMT
Sciencemark 2.0 Moldyn (Aceshardware) - 20.5% improvement with SMT

In effect, SMT transforms a 3.06GHz P4 into a 3.6GHz P4 for multithreaded applications. Intel has announced they will be moving to a 800MHz bus early next year (from 533 now), which should extend performance further.

[Amorph]

Originally posted by Ken_F2:
IBM's 970 results obviously weren't obtained with GCC 2.95.

Do we know what compiler was used? Just out of curiosity... If it was GCC 2.95, look out.

Not to forget, Intel and Microsoft do work together, and have a cross-licensing agreement, which may explain some of the similarities.
Certainly, we know from tests that Intel's 6.0 compiler is no faster than the latest MS VC++ compiler under Windows used for 90% of commercial applications.

Fascinating. So MS finally gave up trying to write their own. Good idea. Their compilers sucked.

Last I'd heard, MS was working with HP on compiler tech. That was a while ago, though. Oh, and MS had shot down Intel's attempt to make Windows run faster on Intel processors, as the antitrust trial revealed, so I had assumed that MS had no interest in using the fastest available compiler. They want people to keep upgrading, after all.

Don't forget Intel's Hyperthreading (SMT) implementation, which uses DEC Alpha-inspired technology to improve performance in multhreaded applications by up to 40%.

I haven't. But would you rather have one processor pretending to be two, or two processors? As far as power dissipation goes, you can have 5 970s running for the power cost of one of these hyperthreaded monsters. Also, there's a penalty for running a single-threaded app on a HT processor.

It makes sense as a way to provide some of the benefits of MP while keeping a simpler, cheaper single-processor mobo design, but since Apple's already gone dual, and since the 970 will be an SMP monster, there's no reason for them not to continue with true SMP.

In effect, SMT transforms a 3.06GHz P4 into a 3.6GHz P4 for multithreaded applications. Intel has announced they will be moving to a 800MHz bus early next year (from 533 now), which should extend performance further.

Is that 200MHz quad-pumped, 64-bit, or 400MHz double-pumped, and narrow (16 or 32 bit)? I doubt they actually have an 800MHz clock on a bus.

I'm more interested in measured throughput and latency numbers than bus clocks.

[Ken_F2]

The general theme though, even according to IBM, is that the chip is about on par with the Pentium.

Where does IBM say that?

IBM said so in its 970 presentation; it said the 970 would offer "competitive performance." And IBM backed up that assertion by quoting projected SPEC results, which are recognized by the microprocessor industry (including IBM) as THE MEANS to compare processor performance across different platforms. IBM makes extensive use of SPEC benchmarks in its marketing literature for the POWER4.

The RC5 tests measure performance in one specific task that makes specific use of one specific type of processor operation--that happens to do much better under the PowerPC than many other processors. But SPEC is a SUITE of 12 to 14 different unix tests designed to measure many different aspects of processor performance to reach an OVERALL assessment of performance (for floating point and integer respectively). The SPEC suite test certainly isn't perfect, but most in the microprocessor industry consider it a good measure of system performance. If you'd like to see a sample breakdown of the SPECint2000 suite, with gzip, gcc, perlbmk, vpr, parser, etc tests, see this page.

According to IBM's SPEC results, which it used to qualify the 970's system performance, the 970 performs integer operations at slightly less speed than a P4 2.8GHz, and performs floating point operations at about 15% faster than a P4 2.8GHz. Those same results make it more than two times as fast as any PowerPC processor to date (from SPEC results), although I have not seen results for the latest PowerPC processors with GCC 3.0. IBM has one of the best compilers in the industry with its own Visual Age 6.0, so no doubt that also helped to improve their SPEC results.

That's not to say that Apple's 970 systems can't or won't outperform their Intel counterparts. Certainly, if the 970 can perform at anywhere near the level of the latest P4 (3.2GHz P4 with 800MHz bus in April, or 3.66GHz PentiumV with 800MHz bus in October), and Apple continues its use of dual processor machines, then they will certainly outperform single processor Intel desktops.

[kupan787]

Originally posted by Ken_F2:
In effect, SMT transforms a 3.06GHz P4 into a 3.6GHz P4 for multithreaded applications. Intel has announced they will be moving to a 800MHz bus early next year (from 533 now), which should extend performance further.

Ya, but I read in a thread over on AppleInsider that one of these P4's with hyperthreading consumes 106 watts (!!). That is insane.

So the 970 uses, what like 20 watts at 1.8 GHz? So I could run 5 970's or one P4 3.06 GHz with hyperthreading.

Hmm...which would you choose?

[Ken_F2]

Ya, but I read in a thread over on AppleInsider that one of these P4's with hyperthreading consumes 106 watts (!!). That is insane.
So the 970 uses, what like 20 watts at 1.8 GHz? So I could run 5 970's or one P4 3.06 GHz with hyperthreading.

Yes, with Hyperthreading, more of the P4 is active, so it obviously produces more heat. But you won't find that 100+W figure on Intel's web site; officially, Intel claims 81.8W for the P4 3.06 with Hyperthreading enabled, but this is only for a 90% cpu load. People have somehow extrapolated to obtain the 100+W figure at 100% load. The "typical" or "average" heat dissipation is closer to 50W.

IBM quoted average power consumption of 42 Watts at 1.8GHz; see slide 14 of the linked presentation. Maximum power consumption will obviously be greater (50W or 60W?), but IBM hasn't provided those figures yet. IBM did say they would have a low power 1.2GHz version that would offer average (not maximum) heat dissipation of 19W. That's still a good deal more heat than put out by current low-end G4 processors.

Certainly, the 970 is cooler than the P4, but it nowhere near 1/4 as cool, and probably not half as cool. It is a large processor, about the size of the P4 with nearly as many transistors. We won't know just how much "cooler" the IBM 970 runs until IBM provides maximum dissipation numbers for 100% cpu load. When IBM finally does release maximum dissipation figures, the 1.8GHz 970 will probably put out comparable heat to the "Northwood" P4 2.0GHz (albeit with performance at least as good as a P4 2.8 in integer, and a P4 3.2GHz in floating point). Certainly, it would seem possible for IBM could release "hotter" processors at even faster speeds than 1.8GHz.

[Amorph]

Originally posted by Ken_F2:

IBM quoted average power consumption of 42 Watts at 1.8GHz; see slide 14 of the linked presentation. Maximum power consumption will obviously be greater (50W or 60W?), but IBM hasn't provided those figures yet. IBM did say they would have a low power 1.2GHz version that would offer average (not maximum) heat dissipation of 19W. That's still a good deal more heat than put out by current low-end G4 processors.

Ah, yes. I had misremembered.

mea maxima culpa on that one.

Still, it sounds like two 970s is a viable alternative to one HT P4.

[Metzen]

Originally posted by Ken_F2:
Where does IBM say that?
IBM said so in its 970 presentation; it said the 970 would offer "competitive performance." And IBM backed up that assertion by quoting projected SPEC results, which are recognized by the microprocessor industry (including IBM) as THE MEANS to compare processor performance across different platforms. IBM makes extensive use of SPEC benchmarks in its marketing literature for the POWER4.

That doesn't mean IBM is comparing the PPC970 to x86, as I highly believe they would do. By competitive performance, I'm sure they're referring to workstation machines ala Sun.

x86? Seems laughable...

Originally posted by Ken_F2:
The RC5 tests measure performance in one specific task that makes specific use of one specific type of processor operation--that happens to do much better under the PowerPC than many other processors.

Hold on... Hold on... Let's play the word game:

The SPECint tests measure performance in one specific task that makes specific use of one specific type of processor operation--that happens to do much better under the x86 than many other processors.

You truly believe that a Power4 is less powerful than a P4 because of SPEC scores? Quite laughable...

[Simon]

Originally posted by Ken_F2:
We won't know just how much "cooler" the IBM 970 runs until IBM provides maximum dissipation numbers for 100% cpu load.

Sorry Ken, but I don't like your agumentation and I believe it's deceiving. Each and every time you see an advantage of the P4 you compare it to the 970 numbers you have found somewhere which are supposedly from prelimenary IBM material or from some IBM presentation. When the 970 seems to show behaviour that is more mature compared to a P4 you start saying that we can't judge until we see a 970 for real and that the P4 is already out there, yada yada...

So now, what is it? Are we gonna do a little chit chat on stuff we are just pulling out of somewhere or are we going to wait until we have both chips side by side? Tell me and then stick to it.

Didn't want to sound nasty but for fairness sake we should stay consequential here.

[brainchild2b]

lets get something very clear. Using the IBM 970 will most likely cause us the same headaches we have right now with motorola clock speeds. If you read at ars technica the IBM chip uses registers differently and will top out at a much lower megahertz than the current pentium technology. Even IBM will admit this but since for the time being their slower proccessor will beat a faster pentium no one seems concerned. Eventually we will see the same thing as the motorola issue with the IBM chip a year or two down the road and will have the same stupid problem all over again.

[Ken_F2]

That doesn't mean IBM is comparing the PPC970 to x86, as I highly believe they would do. By competitive performance, I'm sure they're referring to workstation machines ala Sun.

Sun doesn't have any workstations that are competitive performance-wise. The only reason people buy Sun workstations at all anymore is to preserve an investment in legacy software and [legacy] human resources. Because of their outdated Ultrasparc hardware, SUN has been forced to release its Solaris OS and software for Intel.

SUN holds about 17% workstation market share, compared to about 28% for Dell (Intel), and about 25% for HP (mostly Intel; 2001 figures). IBM's best-selling workstations use Intel processors with unix, while it's best performing and most expensive workstations use POWER4 processors. On a related note, some 88.4% of servers sold use AMD/Intel processors, although only a minority use windows.

The SPECint tests measure performance in one specific task that makes specific use of one specific type of processor operation--that happens to do much better under the x86 than many other processors.

No, once again, SPECint is a suite of 12 different tests that measures 12 different aspects of UNIX performance, with the results then weighted to establish a total score. SPECfp uses 14 different tests to assess UNIX floating point performance. The SPEC aren't synthetic tests, they are tests of actual unix operations (compiling, gzip, etc) that people perform every day. SPEC is very good representation of system performance under UNIX. The people here slamming SPEC obviously don't understand it; most appear to be speaking out of ignorance.

IBM's has one of the best compilers in the business in Visual Age 6.0, which it uses for all its SPEC submissions, so the whole "the compiler is crap" argument goes out the window. With it's upcoming 1.6 and 1.8GHZ POWER4 processors, IBM will regain the lead over Itanium2 in its SPECfp scores, and possibly retake the SPECint crown from the Athlon and P4 as well.

Sorry Ken, but I don't like your agumentation and I believe it's deceiving. Each and every time you see an advantage of the P4 you compare it to the 970 numbers you have found somewhere which are supposedly from prelimenary IBM material or from some IBM presentation. When the 970 seems to show behaviour that is more mature compared to a P4 you start saying that we can't judge until we see a 970 for real and that the P4 is already out there, yada yada...

You don't make any sense. Supposedly from some presentation? If you want to learn more about the 970, download the presentation from IBM's site right here. This is the official presentation IBM gave on the 970 at the Microprocessor Forum, in PDF format.

On slide 14, IBM quotes average power consumption of 42 watts. Intel quotes average power consumption of about 50 watts. Intel also quotes "maximum power consumption" (but it isn't really maximum, because it's 90%) of 81.8 watts at 3.06GHz. IBM doesn't provide that information in their official presentation, and until they do, I don't see how we can compare that data.

[TimmyDee51]

I'm not so sure we will have the same problem, though. The 970 is a direct cousin of the Power4, so as the Power4 grows, so too does the 970. Plus, IBM has shown in the past that they can create architectures that are far more scalable than comparable Motorola ones. Take the G3 as an example. Moto's G3 topped out at 500 MHz, where IBM has taken them to 1 GHz. If we follow that 100% increase in clock speed for the same architecture (not to mention the L2 cache advances), the 970 should be well off for a while.

[Ken_F2]

lets get something very clear. Using the IBM 970 will most likely cause us the same headaches we have right now with motorola clock speeds.

IBM has said the 970 will hit at least 1.8GHz on .13 micron. Who's to say that it won't hit 2.5GHz to 3.0GHz on .09 micron in 2004/2005? And then perhaps there will be a 990, based on the POWER5 that is coming in 2005?

So long as Apple continues its use of dual processors, and there is no reason to think they won't, then the 970 should make the Mac platform competitive for quite some time. For if a 1.8GHz 970 can deliver 90% of the performance of a P4 3.6GHz with Hyperthreading, then imagine what two 970s can do. Plus, Altivec--which isn't figured in to the above--will help even more for certain tasks. It might take a 5.0GHz PentiumV to match a pair of 970s, or a 6.0GHz PentiumV to match a pair of 1.8GHz 970s when Altivec is used.

[kupan787]

Originally posted by Ken_F2:
Intel claims 81.8W for the P4 3.06 with Hyperthreading enabled, but this is only for a 90% cpu load. People have somehow extrapolated to obtain the 100+W figure at 100% load. The "typical" or "average" heat dissipation is closer to 50W.

IBM quoted average power consumption of 42 Watts at 1.8GHz; see slide 14 of the linked presentation. Maximum power consumption will obviously be greater (50W or 60W?)

So when Intel claims 81.8W, you say it is more closer to 50W. but when IBM claims 42W, you say it will be closer to 60W. Um ok, to me it looks like you are giving a big break to the Pentium.

Why not go by the number they both give us. Of course they will be off one way or the other (heh, thats marketing), but at least lets not extrapolate our owns numbers without testing first. And since you can't buy either yet (or can you buy a P4 with HT now?), we should wait and see.

BTW what does a similar AMD disipate? What are the projections for the opteron (that is their 64 bit processor, right?)

[Metzen]

Originally posted by Ken_F2:
IBM has said the 970 will hit at least 1.8GHz on .13 micron.

I hate it when people quote the introductory speed of the PPC970 with it's top out speed. It's running at 1.3GHz right now, and is to be introduced at 1.4-1.8GHz. The scaling happens after that.

Get it? It'll scale a few GHz before it hits it's EOL.

It won't "at least hit 1.8GHz", it will hit it and than surpass it.

Signficant quote from article:

The chip is expected to debut at 1.4 GHz and 1.8 GHz, and it will eventually be clocked down to 1.2 GHz and clocked up to 2 GHz

[Avon]

STOP THE MADNESS!

Just give me my PPC 970!

[Ken_F2]

So when Intel claims 81.8W, you say it is more closer to 50W. but when IBM claims 42W, you say it will be closer to 60W. Um ok, to me it looks like you are giving a big break to the Pentium.

No, Intel claims 81.8W for "maximum dissipation," as in worse case scenario; this is what any manufacturer producing a P4 heatsink must be able to tolerate/support. Intel also claims a lower figure for typical use, just as AMD does. IBM claims 42W for typical use, but they don't provide a maximum figure. If you can't understand the difference between typical use, and a worse case scenario (two different figures from Intel and AMD!), then I don't know what to tell you. Intel and AMD give us two numbers, IBM gives us one. You should compare apples to apples when possible.

Of course, the problem does arise with the definition of "typical use." Just what exactly constitutes "typical use?" For the Athlon 2700+, AMD quotes 62W for typical use (higher than Intel's "typical" figure), and 68.3W for "maximum" consumption (lower than Intel's "maximum" figure).

AMD has not quoted heat dissipation figures for Hammer as yet, although it is expected to run cooler than Athlon. A "Hammer" at 2.00GHz will offer comparable performance to an Athlon at 2600-2700MHz, so obviously the former will run cooler at a given processor performance rating.

It won't "at least hit 1.8GHz", it will hit it and than surpass it.

That's not what IBM's presentation says. See page 15 of their official 970 presentation. It's possible the 970 could exceed 1.8GHz on the .13 process, but IBM has said nothing to that effect. And if you want technical information on why it may not, read the article over on Arstechnica.

[UnixMac]

interesting thread, very informative.... as an asside, on Screen Savers tonight, Patrick Norton overclocked a P4 3GHZ to 4.1GHz.. not bad..

[kupan787]

Originally posted by Ken_F2:
No, Intel claims 81.8W for "maximum dissipation," as in worse case scenario; this is what any manufacturer producing a P4 heatsink must be able to tolerate/support.

If you can't understand the difference between typical use, and a worse case scenario (two different figures from Intel and AMD!), then I don't know what to tell you.

Wrong.

Maybe it is you that can't tell the difference from typical and maximum. The 81.2W is Intels typical claim. From http://www.aceshardware.com/read.jsp?id=50000320

As far as I know, no x86 CPU has dissipated so much heat as the 3.06 GHz Pentium 4 with HyperThreading. The maximum power dissipation will never occur in the real world, as this can only be reached with special "thermal viruses." Luckily, the Pentium 4 has a very efficient heatspreader and a good clamping mechanism. We could not measure temperatures higher than 57�C with a decent cooler.

Nevertheless, the enormous amount of electrical current that the 3 GHz P4 requires will bring many motherboards and PSUs to their knees. This is probably the reason why we won't see the 3.2 GHz version arriving soon, as Intel will lower the power dissipation of its cores before moving to higher speeds. But right now, the current 3 GHz parts require 1.55V core voltage, and it is the most demanding processor of all.

Pentium 4 3.06 GHz Typical 81 W Max +/- 105 W

So what I was saying is valid. The typical for a P4 3.06 with HT is 81 W, and the typical for a 970 at 1.8 is 42W.

[clarkgoble]

I believe that the 970 was designed with scalability in mind. While IBM predicts speeds up to 1.8 GHz, I'd assume that is with the initial production. All other productions have gone to higher clock rates. I'm not sure what in the ARS Technica article you're referring to that suggests limited scalability.

I should add that IBM has stated that it was designed from day one to be very effective SMP. It will handle 4-way SMP easily and purportedly 8-way although I doubt Apple would ever produce such a system. However if the cost per chip is low enough I could see a 4 way 970 system which would be quite interesting for Photoshop or one of those new graphic companies Apple has bought.

Adobe is coming up to beta testing for the next round of upgrades on many of their products. So hopefully one of the things they've done is improved multithreading.

Getting back to speed. I've seen specs for machines running 970 with clock speeds in excess of the stated speeds. These are OS/400 systems, but it should give an idea of what's ahead for the Mac.

http://www.midrangeserver.com/tfh/tf...y01-fig01.html