IBM Gobi, Mojave, and PowerPC 970 and Beyond

[RalphNumbers]

Originally posted by nuckinfutz
I don't believe this is true. The less pipes you have the harder it is to clock.

I believe you mean the less pipeline stages you have the harder it is to clock. (either that or you're just dead wrong, having less parallel pipes certainly does make things simpler, and generally more clockable)
In any case, longer pipes are a tradeoff for higher clock speeds, but they aren't the only way, or the best way to achieve those speeds.

Originally posted by nuckinfutz
Apple is not the only company that uses G3's. There was a market for a lowcost chip like the G3 at higher mhz. It never materialized. I doubt Apple caused this..that's giving them too much credit.

Apple is the only company that uses G3s in significant enough quantity to drive production to new tech, ans no, there wasn't such a market. Who would buy a high clocked G3 in any significant quantity? Not Apple, Moto and IBM have their own reasons not to be interested in such a thing, the embedded market has nowhere near the performance pressure the desktop market has (and where performance is a problem they're generally better suited by SIMD or DSPs). That leaves... what? Upgrade makers and non-mac PPC computers? Do you realize how low volume they are?

Originally posted by nuckinfutz
The PPC 970 is only %14 larger than a 180nm G4. IBM's fabs should be better than Moto's and the not needing the L3 cache could make the 970 systems basically the same price as the G4. G3's are going to be cheaper of course so I agree with that. IBM moves to 300mm wafers soon and provided yields are good Apple could use 970s amongst a majority of their lineup. Things get even easier when 90nm chips roll out.

Being able to match G4 prices isn't exactly an accomplishment worthy of note, Moto hasn't been doing much better there than anywhere else.
And, unfortunately increasing transistor density also drives up error density, die size is not the only factor (although it is the major one). In any case, low end processors will continue to evolve along with the high end. G3s will be manufactures in smaller CDs too.

Originally posted by nuckinfutz
This makes no sense. They may be the "only" advantages but those are some pretty big additions. If the Market deems those functions as relevant then the fact that the G3 doesn't have them doesn't bode well for it's use.

But these aren't the kinds of features one looks for in a bargain basement and/or mobile processor. Most of those features would see little or no use in the kinds of systems Apple is likely to continue using G3s in anyway.

Originally posted by nuckinfutz
What's the point of technobabble? One only needs to get on OSX running on and iBook versus Powerbook or encode MPEG2 for iDVD to see the the G3 cannot keep up with the G4. We can generate "What ifs" all night long but fact is the G4 feels perceptably faster to the average user IMO.

Sure, you can generate Altivec-heavy cases where the G4 will be niticablly faster, but there are plenty of other cases where it isn't. And even your chosen cases will probably favor the upcoming G3/w Altivec that the article mentioned.

 

[Rincewind42]

Originally posted by dongmin
Even if the processors are comparable in costs (purely speculative on our part), the processor is only a part of the story. The 970 requires a faster bus, ram, etc. which would add considerably more to the overall system cost. On a $999 computer, the extra $100 or so in component costs would obliterate whatever margins it has. It's a consumer machine, costing half to third of the pro laptop, so they need to save wherever they can.

And if Apple released such a machine it would be MUCH faster than the machine it replaced. And as we are talking a consumer machine, it would probably NOT use the high end parts that the Pro machine version uses (I seriously doubt that the 970 must use a 900Mhz DDR bus, I'm sure the bus can be clocked lower). And regardless, if it ends up costing another $100, then the price will go up - and it will still sell because the performance will be much better.

Originally posted by RalphNumbers
Sure, you can generate Altivec-heavy cases where the G4 will be niticablly faster, but there are plenty of other cases where it isn't. And even your chosen cases will probably favor the upcoming G3/w Altivec that the article mentioned.

Agreed - but a G3 with Altivec would almost certainly be called a G4 by Apple (after all, the 7400/7410 are VERY different from the other 74xx chips, but are relatively similar to the 750 line and any G3+Altivec chip IBM may create).

But that still doesn't change the fact that a current G4 system will always feel faster than a comparable G3 system (i.e. iBook 900 vs PB 867). This is completely due to the fact that regardless of what applications use Altivec, there is always one running that really matters - the OS.

 

[123]

Re: the G3 is so much better designed than the G4

Originally posted by MacBandit
As far as design is concerned, the G3 is a far better chip than the G4.

What type of G4, G3? Unqualified statement.

That is why the G4 needs Altivec

Ridiculous. The G4 "does not need" Altivec. They took the G3, made some improvements and also added Altivec to the G3, another big (100% orthogonal) improvement.

Originally posted by RalphNumbers

G3s are actually a good deal more easily clockable due to their lower complexity

They also have a short pipeline, sometimes higher speeds can only be acchieved by adding more complexity (= more pl stages).

And of course, the G3 does comparably well to the G4 when hitting L2,

Wrong, the 7400 has superior memory access/caching capabilities (eg. load/store miss folding/merging)

so in practical applications, we can reduce that to:
integer code

This is also totally wrong of course (your reductions/generalizations are all wrong). As mentioned above, the original G4 was an enhanced G3 (memory/cache/fpu/bus) with Altivec. If anything, Integer code is slightly faster on the 7400 and a lot faster on two of them.

Now for the newer revisions. A second Integer Unit has been added as well as more pipeline stages. This allows for higher clocking which will make integer code even faster unless heavy branching occurs. However, according to you, (attention: flawed logic) this is hardly ever the case because integer code is hardly ever vectorized in practical applications and since "branch heavy & vectorizable" reduce to "vectorizable" and eventually even reduce to "not vectorized in practical applications", branch heavy integer code does not exist.

Seriously: An additional integer unit, better fpu, vector units, longer completion queues, better memory/cache management/support and a bus that allows for better multiprocessing is more advanced no matter how hard you try to make yourself look foolish. The design is more complex because the G4 is more powerful/better/faster than the G3 just as the G3 is more complex than a 4-bit adder.

You can always argue that some of the G4's features are not necessary for certain applications/users/products. But the claim was that the G4 is badly designed compared to the G3 and I haven't seen any statement that supports this claim by you or anyone else in this thread.

In any case, longer pipes are a tradeoff for higher clock speeds, but they aren't the only way, or the best way to achieve those speeds.

Now I'm curious, what are better ways to achieve high clock speeds (not talking about performance)?

 

[RalphNumbers]

Re: Re: the G3 is so much better designed than the G4

Originally posted by 123
Wrong, the 7400 has superior memory access/caching capabilities (eg. load/store miss folding/merging)

Keywords here: memory access, not cache. In point of fact the G4 has had some persistent problems with cache access and coherence, especially in MP configs.

Originally posted by 123
This is also totally wrong of course (your reductions/generalizations are all wrong). As mentioned above, the original G4 was an enhanced G3 (memory/cache/fpu/bus) with Altivec. If anything, Integer code is slightly faster on the 7400 and a lot faster on two of them.

On the original G4s it was maybe 10% better integer performance at best at a given clock speed, and it's gotten significantly worse since then because of your favorite "feature", lengthened pipelines.

Originally posted by 123
Now for the newer revisions. A second Integer Unit has been added as well as more pipeline stages. This allows for higher clocking which will make integer code even faster unless heavy branching occurs. However, according to you, (attention: flawed logic) this is hardly ever the case because integer code is hardly ever vectorized in practical applications and since "branch heavy & vectorizable" reduce to "vectorizable" and eventually even reduce to "not vectorized in practical applications", branch heavy integer code does not exist.

Why yes, that is flawed logic, so I suppose it's a good thing I didn't say that. Go back and try actually reading my post this time. (hint: try not to invert things: i.e non-vectorizable=> vectorizable)
Seriously, if your next reply is this incoherent I'm not going to even bother replying.

Originally posted by 123
Seriously: An additional integer unit, better fpu, vector units, longer completion queues, better memory/cache management/support and a bus that allows for better multiprocessing is more advanced no matter how hard you try to make yourself look foolish. The design is more complex because the G4 is more powerful/better/faster than the G3 just as the G3 is more complex than a 4-bit adder.

And would you then use a G3 whenever you needed to add in 4 bits?
I wouldn't, but given all of your blanket statements, you might just be that dumb.

Originally posted by 123
You can always argue that some of the G4's features are not necessary for certain applications/users/products. But the claim was that the G4 is badly designed compared to the G3 and I haven't seen any statement that supports this claim by you or anyone else in this thread.

Yes, "the G4's features are not necessary for certain applications/users/products". When we're talking about using it for those "applications/users/products", wasting resources on implementing those features is pretty much the definition of bad design. By your logic I could make my car a better design by bolting the kitchen sink to it's hood, because I might someday want to wash my hands on the road.

Originally posted by 123
Now I'm curious, what are better ways to achieve high clock speeds (not talking about performance)?

Narrower gates, thinner oxides, shorter max channel lengths, shorter wiring paths, etc...
(and the above all have the advantage of improving performance as well as clock speeds, unlike long pipelines which are basically a marketing gimmick to increase clock speeds without significant improvements in performance, design, or fab tech)






God, replying to this has reminded me why I hate web forums. There's always someone who spends too much time reading documents online he only half understands, then assuming he's suddenly a qualified microelectronic engineer, and arguing by throwing buzzwords around.

 

[pyrotoaster]

Consider what we know...

I'm going to just jump in here and try and bring a different perspective (also, I have no time to read 53 existing posts).

I wrote a little article-post over at MacUnderground called "Long live the G3?" My thoughts boil down to this.

1. It's obvious that Apple has a bigger iBook update in the works (BTW, arn: you might want to lower that iBook "Buy Now!" rating just a bit).
The "new" iBooks don't even have AirPort Extreme! The eMac has AirPort Extreme, and the iBook doesn't. It's crazy.

2. It's not at all obvious what kind of processor Apple's going to put into these new iBooks.
We do know it won't be a current G3. And we also know that it will cast a decent sized shadow over the power in the current 12" AlBook (if not other current Powerbooks as well).

Apple's waiting to put the 970 in the Powerbooks (my guess is that it will be a simultaneous announcement with 970 Powermacs at the WWDC) before it puts whatever faster processor it chooses into the iBook (maybe also at WWDC, maybe at that MWNY-ish thing, maybe later this year).

What's clear is that either this new iBook processor either is still months away from production, or is actually being kept secret (how Apple would pull that off, I don't know ).

And that's what I think (well, I also think the eMac is ugly, but I didn't want to bring that up).

 

[wallinbl]

Re: Consider what we know...

Originally posted by pyrotoaster
The "new" iBooks don't even have AirPort Extreme! The eMac has AirPort Extreme, and the iBook doesn't. It's crazy.

Not all the PowerBooks have Airport Extreme. That's why the iBook didn't get it.

 

[pyrotoaster]

Originally posted by wallinbl
Not all the PowerBooks have Airport Extreme. That's why the iBook didn't get it.

Yes, I forgot to mention the only remaining Titanium Powerbook. It doesn't have AE because it hasn't been updated since last November.

The remaining TiBook model will be replaced with an AE-capable 970 Powerbook at the WWDC.

The iBook is a different story.

 

[Vonnie]

I wonder if they won't do a seperate announcement on their hardware. Not at WWDC, but a seperate one, somewhat like the Music presentation. A Portable presentation, featuring new powerbooks and ibooks.

Maybe they'll show XServes and Powermacs with PPC970 at WWDC. Isn't this the first year Apple has an enterprise track? An PPC970 XServe announcement would fit in rather well.

 

[nuckinfutz]

Add an Altivec equivalant unit to a G3 chip and the perceived advantage will vanish.

Wrong. You still wouldn't have MERSI SMP support nor would you have the better FPU

Even if the processors are comparable in costs (purely speculative on our part), the processor is only a part of the story. The 970 requires a faster bus, ram, etc. which would add considerably more to the overall system cost

The 970 will use the same memory that the current G4's use perhaps a little faster and in Dual Channel. No huge difference here. The Motherboard will most likely cost more to make initially but lack of L3 cache is a benefit. In many ways it's a potential wash.

I believe you mean the less pipeline stages you have the harder it is to clock. (either that or you're just dead wrong, having less parallel pipes certainly does make things simpler, and generally more clockable)
In any case, longer pipes are a tradeoff for higher clock speeds, but they aren't the only way, or the best way to achieve those speeds.

Ralph I can't be dead wrong because looking at the changes from the Pentium 3 to 4 and AMD K6-K7 and G4-G4+ will show that in EVERY case the addition of more Pipline stages increased the clockrate. I know it lowers the IPC but CPU's are always a tradeoff in Design.

God, replying to this has reminded me why I hate web forums. There's always someone who spends too much time reading documents online he only half understands, then assuming he's suddenly a qualified microelectronic engineer, and arguing by throwing buzzwords around.

Ralph. Save it. You got your ass handed to you. I've read post from people alot smarter than you. If your an engineer ..congrats but Webforums are based on common sense ie just because it CAN be done...doesn't mean it should. Arstechinca is the forum you want if you wish to have people kiss your a** and praise your intellect . Cheers .

I'm sorry but it doesn't take a PHD to realize the the G3 is not a superior Design overall than the G4. Some Mac user have become accustomed to believe there's some sort of Conspiracy. There is not. The G4 was the next evolutionary step base primarily on the G3 but with enhancements.

I look forward to seeing IBM's Gobi and Sahara and what applications they can be applied in.

 

[RalphNumbers]

Originally posted by nuckinfutz
Ralph I can't be dead wrong

Sure you can.

Originally posted by nuckinfutz
because looking at the changes from the Pentium 3 to 4 and AMD K6-K7 and G4-G4+ will show that in EVERY case the addition of more Pipline stages increased the clockrate. I know it lowers the IPC but CPU's are always a tradeoff in Design.

Yeah... and this has what to do with your previous post? I never argued that more pipeline stages don't make for faster clocks. (in point of fact I corrected you to say that they did, when you made the erroneous statement "the less pipes you have the harder it is to clock")

Originally posted by nuckinfutz
I'm sorry but it doesn't take a PHD to realize the the G3 is not a superior Design overall than the G4.

But appearantly, it does take some kind of thought to recognise that the G3 is a cleaner design.

Originally posted by nuckinfutz
Some Mac user have become accustomed to believe there's some sort of Conspiracy. There is not. The G4 was the next evolutionary step base primarily on the G3 but with enhancements.

It's not a conspiracy, it's just Motorola's utter failure to adapt to the desktop processor market. Those "enhancements" were far too expensive, in terms of cost, design/manufacturing complexity, and development problems&delays, to justify themselves.

 

[123]

Re: Mr. chief engineer

Originally posted by RalphNumbers
Keywords here: memory access, not cache.

You're actually wrong again, load instructions in general, eg. L2 access in case of a dL1 miss.

On the original G4s it was maybe 10% better integer performance at best at a given clock speed, and it's gotten significantly worse since then because of your favorite "feature", lengthened pipelines.

1) You forgot about the simple/complex integer units, bus speed increase, on-chip L2 cache with lower latency etc.
2) I'm not talking about IPC, I'm talking about performance. The reason why pipelines are lenghtened is because the processor can be clocked higher which along with good branch prediction leads to higher performance. Comparing two chips at the same clock speed is interesting but doesn't say much about their actual performance and even less about the quality of their design.

Why yes, that is flawed logic, so I suppose it's a good thing I didn't say that. Go back and try actually reading my post this time. (hint: try not to invert things: i.e non-vectorizable=> vectorizable)
Seriously, if your next reply is this incoherent I'm not going to even bother replying.

Sorry, my bad. Anyway, that doesen't change the fact that your reductions are all wrong:

1) Just because there isn't a lot of vectorized code out there, that doesn't mean that vectorized code does not exist at all. Especially time critical code and Apple's own code IS vectorized.

2) Just because Apple builds some G4 products without L3 cache you can't just neglect the G4's support for it when talking processor design.

3) Since the G4 handles cache misses much better than the G3 (and because of 2), it is not permitted to reduce "integer code with a working set of less than 512K..." to "integer code..."

4) Even though most code is not vectorized, it is wrong to assume that it CANNOT BE vectorized (is non-vectorizable) and to conclude that all integer code therefore must be full of (like every other instruction or so) unpredictable control/flow dependencies / hazards -> stalls.

Of course, the final conclusion isn't right either and in case of the 7400 this whole thing doesn't even matter since its pipeline is the same as the 750's.

And would you then use a G3 whenever you needed to add in 4 bits?
I wouldn't, but given all of your blanket statements, you might just be that dumb.

What has this to do with anything?

Anyway, YES, if I have to add 4-bit numbers, I don't use a 74F283, but a MC68000 processor (in a TI-89 calculator) and I'm sure a whole lot of people do since it can also add 8-bit numbers, you know. To me this makes sense because I'm not adding 4-bit numbers all day long but do other stuff as well, but I might as well "just be that dumb".

Since I am "that dumb", I also listen to mp3s while compiling and occasionally add a new CD to my collection, obviously you don't.

Yes, "the G4's features are not necessary for certain applications/users/products". When we're talking about using it for those "applications/users/products", wasting resources on implementing those features is pretty much the definition of bad design. By your logic I could make my car a better design by bolting the kitchen sink to it's hood, because I might someday want to wash my hands on the road.

Congrats, what a good analogy. BTW, I'm not saying that it is a BETTER design, I'm saying it is at least NOT WORSE. The G4 is, however, better suited for today's tasks.

And yes, IF we WERE "talking about using it for those \"applications/users/products\"" ONLY, the design WOULD be bad. However, since we ARE NOT talking about this, it IS irrelevant. We are talking about a processor that is being used for all sorts of applications by all sorts of users, and most of them actually do use the G4's features.

You've still failed to point out why the 7400's (or any other G4's) design is worse than the 750's. Just because Mr. RalphNumbers doesn't need AltiVec all day long, never uses the FPU and doesn't need MERSI cache coherency, it certainly doesn't make a processor badly designed.

Narrower gates, thinner oxides, shorter max channel lengths, shorter wiring paths, etc...
(and the above all have the advantage of improving performance as well as clock speeds,
unlike long pipelines which are basically a marketing gimmick to increase clock speeds without significant improvements in performance, design, or fab tech)

Yes, and all of this can be done with longer pipelines as well.

marketing gimmick

The only "marketing gimmick" I know is Apple's "megahertz myth" myth. Being able to process many instructions concurrently and increasing throughput is a "significant improvement" in design. There are issues with pipelining (and especially with very deep pipelines) but that doesn't mean the idea is bad in general.

God, replying to this has reminded me [bla bla...]

Sorry, didn't know you are the world number one all time super genious. But then again, I probably would have known right from the beginning if you actually had added some substance to your posts. Instead, you go like "bah bah, the G3 is a much cleaner design because it is. And the G4 is a very badly designed chip because it is." Doesn't sound like engineer talk to me.

 

[MacBandit]

Sorry but both of you sound like babbling fools. Neither of you are coming across as being the one that is correct and instead are just making ass's out of yourselves. Just agree to disagree and end it already.

 

[joelc]

Yeah...I don't really understand most of what is being debated here, but at this point I don't care much. The G3 is either better or worse than the G4. That's about what you two have accomplished. I read most of it hoping to learn something, perhaps the answer to the question of which is better, or maybe what merits each has. By now, all I've learned is that nothing is going to settle this question except the release of a great new chip making the debate outdated and pointless. So I'm still waiting anxiously for the 970 or the 750GX or something. I'm sure they will both be fast. Does anyone dispute THAT? Let me say this: for whatever problems the G4 has, it seems pretty fast to me, and I don't give a hoot whether it needs more or fewer pipelines or better cache access or more integer units. I don't even know if I'm talking about real things. It's pretty fast. Let's settle our Motorola vendettas in private, eh?

 

[jeffosx]

Classic and 970?

I have seen alot of talk about how easy OSX will be to run on the 970 but will OS9 (or Classic) need alot of work to get going on the new chip? Does this next gen mark the end of OS9 altogether?

Cheers

 

[Snowy_River]

Originally posted by Rincewind42
...
But that still doesn't change the fact that a current G4 system will always feel faster than a comparable G3 system (i.e. iBook 900 vs PB 867). This is completely due to the fact that regardless of what applications use Altivec, there is always one running that really matters - the OS.

Well, whether or not OS X uses Altivec, I think that there are those that would disagree with the statement "a current G4 system will always feel faster than a comparable G3 system". Case in point, here is a CPU test comparing several PB G4 systems to several iBook G3 systems (before the upgrade to 900MHz).

The rest of the tests are shown here (about half way down the page):

http://www.cube-zone.com/

 

[barkmonster]

G3 better for integer, G4 better for floating point.

I found an old review of 2 cpu upgrades being tested in 1 mac (comparable results with all factors being equal).

G3 500 vs G5 350

Yes, it's only the 750 and 7400 PowerPC chips, both have the same amount of backside cache, L2 cache and both have the same number of pipeline stages.

Simple dividing the Mhz by the CPU score in each of the test results It showed that a G4 cpu is 11.7% faster at floating point calculations than a G3, integer performance was a different story with the G3 being 2.5% faster than the G4.

I know this doesn't compare to how modern versions of the G3 and G4 cpus would perform but iBook vs Powerbook tests have proven how weak the G4 is when altivec isn't an option (check out the 700Mhz iBook thrashing a dual 800Mhz G4 in the Bryce test), this also exposes how much the extra 3 pipeline stages and the smaller L2 cache have reduced the performance compared with the original 7400 and 7410 series.

The G3 with Altivec is going to be a better overall chip than any current G4, even if it doesn't support SMP, it would be a screamer of a cpu in the iMac, eMac and Powerbook and a perfect replacement for the G4 in those models with the SMP capable 970 replacing the G4 in the towers and Xserves.

I don't agree with the 'feel faster' stuff, it's apple's bloatware OS that's dragging down the performance of non G4 models. If OS X needs altivec just to provide something close to the smooth response of OS 9 then they shouldn't have thought so big in the first place and offered an entirely 2D option complete with outline dragging instead of whole window dragging and got rid of the shadows and transparency in favour of a small window border like OS 9.

It would make OS X look ugly but it could be an option for non G4 macs, they could even cripple it so the option doesn't even show up on G4s.

Just a thought.

 

[Rincewind42]

Originally posted by jeffosx
I have seen alot of talk about how easy OSX will be to run on the 970 but will OS9 (or Classic) need alot of work to get going on the new chip? Does this next gen mark the end of OS9 altogether?

MacOS 9 as the primary operating system will never run on a PowerPC 970 based system. Even if the modifications are simple enough to be done to MacOS 9, Apple has ceased development of MacOS 9 and thus would never do it.

MacOS 9 as classic is just another 32-bit process to MacOS X, and thus would run just fine, unmodified, on any system that Apple would likely release in the near to medium future (never can comment on where we will be 5-10 years from now!)

Originally posted by Snowy_River
Well, whether or not OS X uses Altivec, I think that there are those that would disagree with the statement "a current G4 system will always feel faster than a comparable G3 system". Case in point, here is a CPU test comparing several PB G4 systems to several iBook G3 systems (before the upgrade to 900MHz).

Since I'm not in the mood to drive across town to verify any of this, I will say that some of the numbers look a little suspect. If you look at the G4 867 numbers between the Average Reported and the Apple Store results there is a HUGE difference. I presume the Average Reported is based on a large number of web submissions vs the single[small#] testing that the Apple Store results are based on, and thus am more likely to trust those numbers. That given, extrapolating the iBook 800 number (about 78?) I would guess that the iBook 900 would score about 88. Looking at the 12" PB it already scores what looks to be a little over 90. The Apple Store results show a huge inversion of this, but I can't comment on the conditions of those tests. And anyway, I own a 1Ghz PB and I KNOW it isn't nearly twice as fast as the 867 (which the Apple Store test indicates) but is decently faster (as the Average Reported indicates).

 

[Rincewind42]

Re: G3 better for integer, G4 better for floating point.

Originally posted by barkmonster
I know this doesn't compare to how modern versions of the G3 and G4 cpus would perform but iBook vs Powerbook tests have proven how weak the G4 is when altivec isn't an option (check out the 700Mhz iBook thrashing a dual 800Mhz G4 in the Bryce test), this also exposes how much the extra 3 pipeline stages and the smaller L2 cache have reduced the performance compared with the original 7400 and 7410 series.

Agreed, the extra pipeline stages do make the G4 slower than the G3 at the same Mhz rating (and at close ratings) in Integer tests - in general at least. But the debate isn't as simple as that. 7400/7410 G4s have 2 integer units, just like the 750. 745x G4s actually have 4 integer units, although they are split as 3 simple+1 complex integer unit. So there are cases where the 7450 can be much faster or pathelogically slower than the 7400/750 in integer code. I would probably guess that Bryce is a slightly pathelogical case . But that aside, given inside knowledge I wouldn't call the DP800G4 vs 700G3 example a 'thrashing' - Bryce doesn't use multiple CPUs for rendering, thus the G4 was playing with one hand tied behind it's back . And I wouldn't make the statement that Altivec isn't an option - I don't know the operations in Bryce, but there are probably some that could be enhanced in that way - but back in 2000 when it was probably being developed, the G4 was still a rather new chip and only being used in one model. I probabaly wouldn't have done any altivec work for it either .

The G3 with Altivec is going to be a better overall chip than any current G4, even if it doesn't support SMP, it would be a screamer of a cpu in the iMac, eMac and Powerbook and a perfect replacement for the G4 in those models with the SMP capable 970 replacing the G4 in the towers and Xserves.

I wouldn't be so bold as to make that prediction. One of the biggest reasons, aside from Motorola, why the G4 hasn't clocked as high as it was initially predicted was the Altivec unit. That is also the reason why in the 745x chips the unit was rearranged, and in the 970 chips the Altivec unit is also the unit with the longest pipelines. The biggest reason to keep the Altivec unit is because many of those last 10% optimizations can take advantage of it for 200% or more of a boost. But adding it to the G3 may have to force IBM to add pipeline stages to it in order to get the speeds that everyone here claims they will make.

I don't agree with the 'feel faster' stuff, it's apple's bloatware OS that's dragging down the performance of non G4 models. If OS X needs altivec just to provide something close to the smooth response of OS 9 then they shouldn't have thought so big in the first place and offered an entirely 2D option complete with outline dragging instead of whole window dragging and got rid of the shadows and transparency in favour of a small window border like OS 9.

There are many definitions of speed - but I'll go with yours for the moment . What you are basically saying is that OS X shouldn't take advantage of a processor option that is increasingly becoming more available (and will almost certainly be ubiquitous on it's computer models by the end of 2004) because it is trying to offer a more grandiose interface? OS X uses altivec when available - so what, OS 9 does too, just not as much (mostly due to development dollars). And if OS X did offer an OS 9 interface, then without Altivec it would be fast, and with Altivec it would be incredibly fast. You can't tell me you want only fast over incredibly fast .

And if we shift gears a moment to the definition of speed being how much you can get done in a period of time, then I most certainly want OS X doing as much as it can in the Altivec unit as possible - becuase it is that much faster than doing it in the integer/floating point units that it isn't funny. And if you still question the usage of altivec, then you shouldn't be surprised to learn that many of OS X's subsystems (lets use CoreAudio, the source of all sound on your computer) are required to provide data in a way that it can be convienently processed via Altivec. I can't see a more ringing endorsement for the technology from Apple, or a bigger sign for things to come. Not to mention that they also hired the man that wrote the first (and as far as I know only) Altivec tutorial out there to do Altivec work for them .

 

[barkmonster]

I'm not suggesting an OS 9 style theme in OS X, just the aqua theme as it is now, with quartz toned down to reduce cpu overhead.

If anything, the alpha channels and transparency effects could be the fastest part of the process (especially seeing as the 3D hardware can handle that on most macs with a radeon or better), it's the PDF based start of the quartz graphics system that makes OS X slow on a lot of models.

PDF is like running a toned down postscript RIP to render the whole screen every time something changes on screen to warrant redrawing it. Drawing directly like OS 9 and every os before it would elimate this bottleneck, infact with the GPU handling all the windows as textures, they could offload the entire graphics system on to the video card. I don't see PDF acceleration high on the list of future GPU features, it made sense for ATI and Nvidia to work on speeding up video playback but rendering PDF quickly isn't something a lot of people outside of the mac or desktop publishing community have any need for. I also believe altivec enhancements to this area of the quartz system are the most likely reason for far snappier performance on a G4 over a G3.

I also think a compromise where windows and their content are only rendered using PDF if the application is specifically written that way but everything else is just fired directly to the GPU as a texture might be a good idea if it's at all practical.

As far as the G3 vs G4, I think the bryce results are made even more in favour of the G3 when you concider OS X will have allocated 100% of 1 CPU to bryce with the other one being used for the OS where as the G3 has to do the whole lot with 1 CPU.

 

[Rincewind42]

Originally posted by barkmonster
If anything, the alpha channels and transparency effects could be the fastest part of the process (especially seeing as the 3D hardware can handle that on most macs with a radeon or better), it's the PDF based start[sic] of the quartz graphics system that makes OS X slow on a lot of models.

This is a misconception. MacOS X use a PDF imaging model, which is a lot different from actually rendering PDF. The difference is that there is no translation from PDF -> Pixels involved in drawing. Instead, you have mathematical primatives that correspond directly to primatives found in the PDF model. Drawing in Quartz isn't necessarily slower than Quickdraw, it just has more features (like antialiasing, floating point coordinates, and subpixel imaging).

Regardless of this, most of the images on screen in MacOS X are still bitmaps. All the controls, background patterns, just ab out anything but text that is drawn by the OS on behalf of the application are actually stored as bitmaps. So in terms of drawing speed, there probably isn't much that can be done to make the actual drawing operations any faster, at least that I know of, but there are things to be done to eliminate redundant drawing.

As far as the G3 vs G4, I think the bryce results are made even more in favour of the G3 when you concider OS X will have allocated 100% of 1 CPU to bryce with the other one being used for the OS where as the G3 has to do the whole lot with 1 CPU.

Good point. I would guess the CPU usage of Bryce on the DP800 vs the iBook 700 would be about 10% (of total), maybe 15%. If only because on the iBook everything does have to be done on 1 CPU (and thus would probably max about 95% of that CPU) while in the DP some other operations (sound, asych IO, screen updates) could be pushed off to the 2nd CPU, giving Bryce 100-110% of the available CPU resources (depending on it's operations during render)

 

[shycoy]

yeah...

i'll take one of each please, as quick as you can ! ! !
All these things sounds great but I need to see them now...COME ON APPLE give us some new toys...

 

[visor]

has anyone...

already posted that there will be no new g3's in reasonable future?
The only thing that is good about g3 is the power consumption - i've read a typical use ov <4watt in an IBM spec of the 750fx.

for the rest - it's way to slow. i've compared my ibook700 against a amd duron1300, and the ibook isn't even half as fast, making the rumours about superior die design quite unbelivable.
I didn't test anything special, just webbrowsing really flash and things like that.

 

[MacBandit]

Re: has anyone...

Originally posted by visor
already posted that there will be no new g3's in reasonable future?
The only thing that is good about g3 is the power consumption - i've read a typical use ov <4watt in an IBM spec of the 750fx.

for the rest - it's way to slow. i've compared my ibook700 against a amd duron1300, and the ibook isn't even half as fast, making the rumours about superior die design quite unbelivable.
I didn't test anything special, just webbrowsing really flash and things like that.

Web browsing and flash are the worst permformance test you can perform. You could probably run the smae tests on a Dual 1.4 and find that the duron beats them due to one thing or another. I couldn't tell you the technical details of it but I know that flash performance is definitely lacking on Macs though it doesn't hinder them in any way. Also web browsing is so browser dependant that you could never get fair comparrison. Try a photoshop test or QuakeIIIArena or something that actual tests the whole machine.