MacBidouille Rumors Followup

[BaghdadBob]

Re: Re: Re: Reality check only suggests problems

Originally posted by fpnc
But for anyone who cares, I'll meet you back here in about another year and we'll compare notes again.

Well, that's pretty close to infinity from now, isn't it? As a note, industry analysts have been predicting the death of Apple for 15 years. Because they would prefer it. It would simplify things.

As my final note in this forum, which has gotten way too "PC Snob Monthly" for me, I just went to chevyrumors.com and the consensus seems to be that there will never be a Corvette that will be faster than a Viper, no matter what. Dang, and it was so close there for a while. I guess Dodge is just that much better than Chevy. If only Chevy was as big and experienced as Dodge was they might stand a chance...*sigh*

 

[jettredmont]

Re: Re: Re: Reality check only suggests problems

Originally posted by fpnc
Yes, "Never" is a long time but I think it's essentially correct because as far as the computer industy is concerned WINTEL is going to be around longer than Apple will, and I don't see any chance of the PCs losing their performance advantages while they are still dominating the business.

Well, Intel can only maintain dominance by outspending on R&D (or just being much more productive with their R&D dollars than the competition). You do not maintain a performance edge via dominance; you maintain that edge via revenues folded over into R&D. Dominance is largely gained by the performance edge.

IBM isn't outspending and outproducing Intel by a wide margin, but it is by a small margin. Which is, of course, where the Power and derivative lines come in, and why IBM is getting pretty darned close to having 90nm manufacturing, etc.

On the long scale, IBM's processor arm has the ability and war chest to out-innovate Intel and regain the performance crown on the desktop, which of course helps Apple because IBM's processor architecture just happens to be the one they co-developed with Apple (and Motorola).

So, I'm not willing to say Apple will never outperform Intel. In fact, the 970 arguably gets us at least to P4 3.2+GHz parity, perhaps slightly beyond, and with more apparent headroom for rapid performance improvements (new processor design vs aging processor design). Given rough parity by EOY and more R&D being spent in the IBM camp than the Intel camp, I'm willing to guess that two years from now Intel will be quite noticably lagging in performance.

Anything can happen between now and then, and cross-platform benchmarks are always open to subjective interpretation, but that's how I see it.

 

[fpnc]

Okay, after some reflection on my "Never" statement, I'll admit that there are no absolutes and anything (given time) is possible. However, let me offer two somewhat fanciful arguments.

First, if you actually believe that sometime in the next five to ten years Apple will introduce a single-processor computer that is clearly shown to broadly out perform the then fastest available PCs in the same product category then you should immediately go out a buy as much Apple stock as possible. Because when that day comes you will stand to make a very good profit on your investment. This doesn't have to happen in three months, one year, or in five years, it just has to happen once for some relatively sustained period of time before you will profit.

Second, as far as the longer term, let me ask the following question:

Fifty years from now what is more likely to have surpassed the PC as the most efficient (powerful, fastest, or best) computing aid:

A.) An Apple Computer (Macintosh Extreme?)
B.) A bio-mechanical brain implant by company XYZ.
C.) A genetically enhanced super human.
D.) Something else (i.e. none of the above).
E.) What's a PC? (response when someone from fifty years in the future is asked this question).

I don't know the answer to this, but I'd suggest that B, C, D, and E are much more likely that A. In fact, the most likely answer may be E.

Of course, none of the above proves anything and in fact both arguments are somewhat silly. But I guess what I'm trying to suggest is that in the high tech industry "Never" doesn't have to extend for very long before in all practical purposes it __really__ is never.

As far as the suggestions of Mac bashing/hating. I don't see that anywhere here. All I see are two divergent viewpoints. One side seems to be satisfied with accepting the performance benchmarks and PPC970 timelines offered by MacBidouille. The other side seems to be somewhat skeptical and is cautioning against building hopes too high. In the case of the latter, I don't see how that qualifies as Mac bashing.

Frankly, I hope Apple does kick WINTEL's butt (that would be good for just about everyone). However, I'm highly skeptical. It will be good, maybe great, but it won't be top of the world.

 

[BaghdadBob]

OK, that said, it's not Apple Intel is up against here.

It's IBM.

And IBM doesn't f*** around like Apple does. I'm cautious too, but IBM is as much the 800 lb. gorilla that Intel is, unlike Motorola, and I guarantee you they would like to humble Intel. And I don't think it will be as hard as all that.

Especially because IBM's OS customer here is a lot more open to change than MS is. MS is going to have to be willing to stop using that "look! You can still use all your old DOS programs" crap and revolutionize their OS around a new architecture before Intel has it's hands undone and can even start to be on the same playing field as IBM. IBM is in a situation where Apple is asking for new architecture for a new OS, not a tired framework for a crap-piled-upon-crap operating system that just won't die.

You just need to take into account who the real players are here. And although the statement has been made that Apple loyalists are easily pleased (not true) they are used to paying considerably more for their computers.

So who has the real barriers to development here? Intel, making processors based on the decades-old x86 architecture, built for a decades-old OS that is afraid of change, for customers who expect to be able to buy an entire system for under $1000? Or IBM, arguably just as capable if not more at chipmaking as Intel, working with Apple who thrives on change, has adopted a completely new architechture in the last ten years, has built its (superior in its foundation) OS from the ground up in the last five years, and whose customers expect to pay through the ******* to enjoy their products?

I think the arguments are compellingly in Apple's favor for the near future. And I think one could argue the other way as well. But to discount with any level of security that there is a good chance Big Blue Apple is on it's way to catching up and passing Wintel with a fully overhauled OS and chipset, that is folly as much as expecting 286% performance from one 1.4 970 over a 3.0 P4.

That's all.

 

[skunk]

Originally posted by BaghdadBob
... whose customers expect to pay through the ******* to enjoy their products?

Puhleeze, "through the nose" sounds a lot less unpleasant!

 

[Kai]

fpnc:

About your reality check thing:
Well, I'm certainly (like you) someone who warns of expectations too high (afterall, only low expectations are never disappointing in the end, they're either met or surpassed, which is even greater! ;-), but:

1) IBM apparently just kicked Intels butt with the new 1.7GHz Power4. They certainly have the potential to do so also on the Desktop! They got the Knowhow, they got the patents (SOI, copper, FCBGA etc, all IBM! Hell, Intel and AMD both license certain patents from IBM afaik!), they got the manufacturing processes, they got the production facilities and they now (could) have the mass-market (or atleast a part of it in the beginning) to finance their R&D like Intel has for the past 25 years!
Motorola - never! IBM - most certainly!

2) People used to say similar things about AMD using the same logic: "AMD is not big enough, they could never afford the R&D to make faster chips than big bad Intel" etc.. Now flashback to 1999, enter the Athlon! ;-)
Dirk Meyer and his team came (from Alpha), saw and conquered (the performance crown)! From what i know they even built the K7 in just a few weeks from scratch!

So I wouldn't rule Apple owning the performance crown some day completely out! Others did it before and looking at IBM i would certainly say it'd be possible Apple does it!

 

[anonmac]

fnpc(sic)...only conspiracy is you.

fpnc said "And please, anonmac, no more talk about WINTEL conspiracies to make the Mac look bad. "

i never said anything about conspiracies...

just that the site that headlines their articles "Macs Blown away by PC's again" HAS links where you can buy the machine that supposedly 'blew away' the macs.

how do you explain the graphs that adobe used being SOOOOO blatantly wrong that they had to remove them after a couple days? if the guy testing (charlie white) can't even MAKE A GRAPH CORRECTLY, you still weant to believe his testing methodology is valid / believeable? if you are not completely closed minded, i ask you this: Would someone who sells a certain product have a vested interest in seeing it do better in a review? OF COURSE. would they rig the test? only if they were unscrupulous.

and yes, i have print references as to speed,like the aberdeen report that showed a dual 1.25 was much faster running concurrent tasks than a similar pc. also-webpage where (xinet.com) the xserve is shown 'blowing the doors off' a comparable dual dell box. do a search there and you will find its true.
or don't and continue to live in ignorance, just saying things like 'i don't have time to look up things'.

HAVE ANY OF YOU ****EVER**** BEEN TO BAREFEATS.COM? they do have the benchmarks i'm talking about. also see xinet.com, and ALMOST EVERY APPLE KEYNOTE, where the photoshop-off is held. (admittedly, like charlie white, they stand to gain from better benchmarks for THEIR product.)

 

[3.1416]

Re: Re: Re: Reality check only suggests problems

Originally posted by fpnc
I'm not saying that Apple will be going out of business, or that Apple won't be able to produce compelling machines, but the single-processor performance advantage will always be tipped towards the PC.

I just don't think that's a universal truth. Wintel has always been dominant over the last 10 years but there have been numerous occasions when Macs had the performance edge, most recently when it was the G3 against the Pentium 2. Motorola screwed up big time with the G4, but IBM isn't Motorola.

 

[macrumors12345]

Re: Re: Re: Reality check only suggests problems

Originally posted by fpnc
Yes, "Never" is a long time but I think it's essentially correct because as far as the computer industy is concerned WINTEL is going to be around longer than Apple will, and I don't see any chance of the PCs losing their performance advantages while they are still dominating the business. I'm not saying that Apple will be going out of business, or that Apple won't be able to produce compelling machines, but the single-processor performance advantage will always be tipped towards the PC.

That doesn't really make too much sense. In the worst case scenario, PowerPC would never catch up with or exceed x86, despite the fact that the POWER architecture that it is derived from is currently matching or beating even the mighty Itanic (see the recently released 1.7 Ghz Power 4+...this will match up very nicely with Itanic, even when Intel cranks the clock to 1.5 Ghz at the end of the year), and has a bright future ahead of it (Power5, etc.). But fine, assume that PPC for some reason always lags x86. Even in that worst case scenario, what would probably happen is that Apple would port OS X to x86. Remember, this is years in the future, so nobody is writing Carbonized apps anymore to support OS 9, so changing chip architectures, while not trivial, would be relatively easy for developers (certainly MUCH easier than going from x86 to Itanic would be for the Wintel world). At that point, Apple and Windows would be running on the same hardware, so it would be strange to assert at that point that Apple's hardware is intrinisically slower than PC hardware. A does not equal A? I don't think so. So I would really caution you against making the statement that Apple will NEVER be faster than PCs in single threaded, scalar integer benchmarks.

 

[macrumors12345]

Re: fpnc:

Originally posted by Kai

1) IBM apparently just kicked Intels butt with the new 1.7GHz Power4. They certainly have the potential to do so also on the Desktop!

Yes, another thing you (fnpc) should consider is that the only reason that the P4 core will be able to match or beat the 970 (on scalar integer calculations) is that Intel is moving to the 90 nm manufacturing process more aggressively than IBM (or anyone else) is. The P4 has clearly maxed out at 3.0 or 3.2 Ghz on the 130 nm process. However, it is pretty clear that the 970 will easily scale to 2.5 Ghz, if not beyond, on the 130 nm process (this comes both from the IBM press release and from the fact that the Power4+, a chip that is intrinsically less capable of high clock speeds than the 970, has already reached 1.7 Ghz on 130 nm). And a 2.5 Ghz 970 should outperform a 3.0 Ghz P4 by a significant margin in scalar integer code (and a larger margin for floating point code, and totally blow away the P4 on vector code and in terms of scalability - P4 can't even be used for MP).

Now, I'm not saying that it somehow shouldn't "count" that Intel is planning to move to 90 nm ahead of other companies. However, to assume that IBM will not move to 90 nm as well anytime in the forseeable future, or to assume that Intel will always move to smaller processes earlier, is a real leap of faith. But your claim that the P4 core will be able to keep ahead of the 970 core is specifically based on the assumption of Intel always being on a smaller process. That isn't something that I would be so certain about.

 

[skunk]

Further information from MacBidouille (with a health warning): the 970 mother-boards will have 6 PCI slots in addition to the AGP, and maybe a 5.1 sound card. The bus will be rated at 200MHz and will support 3200DDR. According to an internal source at IBM, the 970 will surprise the public greatly. This information came through before the fabled benchmarks, and from a different source.

Make of it what you will.....

 

[Snowy_River]

Originally posted by macrumors12345
...which makes it a little strange that he actually has access to such advanced, pre-production machines...

Well, I don't know about that. I've know of people of varying levels of intelligence that have access to some pretty cutting edge stuff in other industries, so just because he's 'stupid' or has approached presenting this information is 'stupid' means anything about whether or not he would have access to such a machine.

Very funny. That could be pretty much impossible to do given that nobody has yet found any credible evidence that this mythical Bryce 6 beta exists (several people have searched Hotline, where the tester claimed it was available - there is no sign at all of a Bryce 6 beta).

Well, as I noted, there has been further evidence of a beta version of Bryce6 in existence. Whether or not it is currently available on Hotline (and who can say whether or not it was available on Hotline?), is another issue.

Really? In what field? In the journals in my field (economics) we are are rarely able to amass such convincing empirical evidence against a particular hypothesis (though certainly not for lack of trying), but any hypothesis which had as much evidence going against it as these benchmarks do would certainly be dismissed by any academic who was not a complete crank.

Physics. And what I was objecting to was not saying that the majority of evidence suggests that these benchmarks are fake. I was objecting to the statement that they have been 'proven' fake. In Physics, the Standard Model of elementary particles is considered one of the most reinforced theories in existence (with some of the predicted values being verified out to dozens of decimal places). However, it is not considered 'proven'. Indeed, there are physicists who devote themselves to finding something that definitively disproves the Standard Model.

Now, if you had said, from the beginning, "Clearly, the majority of evidence leads to the conclusion that these are fake benchmarks", I might have agreed with you. (In the beginning, before more of the evidence came out, I might have still drawn some exception, but, now I would certainly agree with this statement.)

If you want to 'prove' it, present me with some evidence that cannot be explained by some other, however unlikely, set of events. If you want to argue it, then make it clear that that is what you are doing.

 

[Cubeboy]

Re: fpnc:

Originally posted by Kai
About your reality check thing:
Well, I'm certainly (like you) someone who warns of expectations too high (afterall, only low expectations are never disappointing in the end, they're either met or surpassed, which is even greater! ;-), but:

1) IBM apparently just kicked Intels butt with the new 1.7GHz Power4. They certainly have the potential to do so also on the Desktop! They got the Knowhow, they got the patents (SOI, copper, FCBGA etc, all IBM! Hell, Intel and AMD both license certain patents from IBM afaik!), they got the manufacturing processes, they got the production facilities and they now (could) have the mass-market (or atleast a part of it in the beginning) to finance their R&D like Intel has for the past 25 years!
Motorola - never! IBM - most certainly!

2) People used to say similar things about AMD using the same logic: "AMD is not big enough, they could never afford the R&D to make faster chips than big bad Intel" etc.. Now flashback to 1999, enter the Athlon! ;-)
Dirk Meyer and his team came (from Alpha), saw and conquered (the performance crown)! From what i know they even built the K7 in just a few weeks from scratch!

So I wouldn't rule Apple owning the performance crown some day completely out! Others did it before and looking at IBM i would certainly say it'd be possible Apple does it!

What are you comparing the Power4 with, of course it's going to cream any desktop processor in server/scientific apps but it's not going to be much against Intel's or Alpha's server chips, namely Itanium 2 and 3. Also, Alpha now works for Intel developing Itanium chips, when Compaq let go of Alpha, Intel acquired all but one member of the entire team. PPC970 could well take back the performance crown but I wouldn't hold my breath on it.

 

[Snowy_River]

Re: fpnc:

Originally posted by Kai
...People used to say similar things about AMD using the same logic: "AMD is not big enough, they could never afford the R&D to make faster chips than big bad Intel" etc.. Now flashback to 1999, enter the Athlon! ;-)
Dirk Meyer and his team came (from Alpha), saw and conquered (the performance crown)! From what i know they even built the K7 in just a few weeks from scratch!

I'm glad that someone pointed this out. It goes to the fact that there is more involved than just market dominance and $ spent on R&D. There is also skill of the engineers involved, benefit from design decisions (push the MHz higher vs. making more performance per clock cycle), and there is a certain degree of luck involved. Arguably, from what we've heard, IBM has had a fair amount of luck with the production of the 970 having many fewer problems than expected.

 

[macrumors12345]

Originally posted by Snowy_River
Whether or not it is currently available on Hotline (and who can say whether or not it was available on Hotline?), is another issue.

Who is to say whether it is available on Hotline? Well, as MacBd reported it, the tester specifically said it was available on Hotline! Who did you think was making the claim? This isn't something that other people were randomly coming up with.

 

[anonmac]

scaling the wall...top500.org

the list of top500 supercomputers is available at top500.org for anyone who wants to know where the power really is.

out of the top40 supercomputers:
21 are ibm
5 are alpha
6 are intel
2 are nec (notably #1, the earth simulator)
and 2 are cray.

so, to answer the question of whether ibm can scale the wall of intel performance, i would say 'wall scaled, game over.'

NOTE (AND CHECK THE FACTS) THIS:

#12 is a 3,328 processor IBM box that runs at 375mhz (power **3** not the newer power 4)

#15 is an intel box (when i say box here i refer cluster) with 9,632 processors at an undisclosed speed.

if the power3 runs at 375, and has LOTS less processors, why IS IT ALREADY FASTER than what intel cobbled together? (i must assume they are not using p3, as that is about where 375 mhz would be.)

so to all who say IBM will never do it...you couldn't be more wrong, they already have, and the facts are right there, waiting for you to look at them, and then disregard the testing as 'well, in the real world, we all know...'(fill in the ellipses with whatever garbage you are spouting today that you really do not understand.)

 

[macrumors12345]

Re: Re: fpnc:

Originally posted by Cubeboy
What are you comparing the Power4 with, of course it's going to cream any desktop processor in server/scientific apps but it's not going to be much against Intel's or Alpha's server chips, namely Itanium 2 and 3.

What on earth are you talking about?

Here are the respective SPEC scores of the fastest available version of each chip. Keep in mind that these scores correspond to the performance of A SINGLE CORE (i.e. what you would see if there were only a single user on the server running a single thread):

SPECint2000:
Power4+ @ 1.7 Ghz: 1113
Alpha 21364 @ 1.15 Ghz: 877
Itanium 2 @ 1 Ghz: 807

SPECfp2000:
Power4+ @ 1.7 Ghz: 1699
Alpha 21364 @ 1.15 Ghz: 1482
Itanium 2 @ 1.0 Ghz: 1431

Clearly the Power4+ core has a significant performance margin over both Alpha and Itanic at this time. However, more importantly, the Power4+ packs TWO CORES per chip, whereas Alpha and Itanic each have only one. This means that in the server market each Power4+ chip has MORE THAN TWICE as much processing power as each Itanic or Alpha chip.

In terms of the future, Intel is claiming that the process shrink of Itanium 2, which supposedly will ship towards the end of the year, will improve performance by 30 to 50%. If that is true, then the 1.5 Ghz Itanium 2 will match the integer performance of a single 1.7 Ghz Power4+ core and beat the floating point performance by 10 to 20%. However, Itanic will still only be single core, so the Intanic 2 chip will still have substantially less processing power than the Power4+ chip. Furthermore, IBM claims that the Power5 chip, due to ship by next year, will quadruple the performance of the Power4+ (300% increase). If the numbers are anywhere close to that, then clearly Power will maintain or even widen its lead on Itanic. So it is quite unlikely that Itanic is going to be beating the POWER architecture in the area of server class chips anytime in the near future. As for Alpha, well, HP is sending that chip to its grave (for marketing reasons, of course, not because it is a crappy design...quite the opposite, in fact, as the benchmarks show that it is still capable of competing with the best of them).

These numbers are all publicly available on the web. You really ought to use Google before you post strange claims like "Power cannot compete at all with Itanic and Alpha."

 

[anonmac]

hear hear macrumors12345

we seem to think along the same lines when it comes to power vs intel/alpha/ et al.

nice post. i too think some ***RESEARCH*** should be done by the wintelites before posting how superior they are to the rest of the world (at least to themselves.)

i guess if you don't have real references, you just make stuff up "hey in 2050 the intel pentium 17 (still 32 bit) will kick the power4+ ass!"

yup, maybe by then ;^>

 

[jettredmont]

Re: Re: fpnc:

Originally posted by Cubeboy
What are you comparing the Power4 with, of course it's going to cream any desktop processor in server/scientific apps but it's not going to be much against Intel's or Alpha's server chips, namely Itanium 2 and 3. Also, Alpha now works for Intel developing Itanium chips, when Compaq let go of Alpha, Intel acquired all but one member of the entire team. PPC970 could well take back the performance crown but I wouldn't hold my breath on it.

You are confusing Power4 (specifically Power4+, the reduced-process and increased-frequency version of the Power4) with the 970.

The PPC 970 is a desktop-oriented "lite" version of the Power4, and, no, would not blow away an Itanium 2 or an Alpha (although it happily will not require its own power substation and two tons of insertion force to put it on the motherboard). Intel had planned on bringing out a desktop-oriented version of Itanium 2 this fall (Deerfield), but those plans appear to have been scrapped. If you have an Itanium 2 on your desktop, yes, you will have more power than your Mac-loving neighbors.

The Power4+/1.7GHz blows away the fastest Alpha and Itanium 2 systems quite nicely.

 

[Snowy_River]

Originally posted by macrumors12345
Who is to say whether it is available on Hotline? Well, as MacBd reported it, the tester specifically said it was available on Hotline! Who did you think was making the claim? This isn't something that other people were randomly coming up with.

Sorry, you missed my point. It's possible that the beta of Bryce6 was available on Hotline, but isn't currently available.

Anyway, I think that this discussion has outlived its natural life by a significant amount. Shall we call an end to it? Peace?

 

[BaghdadBob]

No. I'm going to extend it by almost two hours, so there.

And hydrogen SUCKS! HA!

 

[Snowy_River]

Originally posted by BaghdadBob
No. I'm going to extend it by almost two hours, so there.

And hydrogen SUCKS! HA!

Ah, these boards are a never ending source of entertainment...

 

[matznentosh]

Originally posted by skunk
Further information from MacBidouille (with a health warning): the 970 mother-boards will have 6 PCI slots in addition to the AGP, and maybe a 5.1 sound card. The bus will be rated at 200MHz and will support 3200DDR. According to an internal source at IBM, the 970 will surprise the public greatly. This information came through before the fabled benchmarks, and from a different source.

Make of it what you will.....

Am I wrong or is a 200 mhz bus much slower than what everyone has been talking about? I've seen 800 mhz thrown around a lot, with the suggestion that the high bus speed would be a considerable part of the extra processing power. So unless the bus is quad pumped or something, it won't be that much faster than a 167 mhz bus.

As I've said before, I still think a conservative estimate is more likely, that a 970 Macintosh would be 1.5 times as fast as a G4 at the same clock speed.

 

[Dave Marsh]

Bus speed?

An excellent question. I had the same thought last night and read intently as a number of the board's better techs talked around this issue, but no one nailed it for the less tech savvy like me.

There was talk that the 3200 memory mentioned was DDR400, which is consistent with what others have said earlier was needed for this CPU, which was comforting. Then there was confusion over whether the 200MHz mentioned related to the memory or the PCI bus, but most settled on the side of the CPU bus.

If I understand properly, the 970 doesn't use a traditional bus, but rather has two one-way links for memory access. But, even then, two concurrent paths sharing the DDR400 over a 200MHz bus, for an equivalent (?) 400Mhz shared path, would still appear far slower than the talked about "half the CPU speed bus" discussed earlier (900MHz for 1.8GHz CPU). Also, I'm not sure what "double pumped" or "quad pumped" means, although a number of people here throw around those terms frequently...perhaps that will have some bearing on the answer to this question.

I really doubt that Apple would cripple their new CPU, after years of suffering the effects of the bus starved G4s (which are really pretty good chips, by the way), but we'll have to wait for the techs to join in to clear this up.

I'm also hoping MacBidouille returns with the promised board/system specs on the 15th, although I'm a bit anxious on that after all the grief they took from some of the "experts" on the boards following their last posting. I'm keeping my fingers crossed, though.

 

[jettredmont]

Originally posted by matznentosh
Am I wrong or is a 200 mhz bus much slower than what everyone has been talking about? I've seen 800 mhz thrown around a lot, with the suggestion that the high bus speed would be a considerable part of the extra processing power. So unless the bus is quad pumped or something, it won't be that much faster than a 167 mhz bus.

The 200MHz bus speed refers to the memory bus, using DDR-400 memory, not the FSB, or bus from the CPU to the system controller.

The 970 at 1.8GHz has a 900MHz FSB (two half-width unidirectional buses). Discounting overhead, that 900MHz bus can transfer 6.4Gb/s of data, which is equivalent to an 800MHz no-overhead bus. Note that the latest P4s also have an 800MHz bus, although that is "raw" speed, not discounted for bus overhead (which I believe is less significant with the P4 anyway).

Contrary to what some people have said here and elsewhere, we really don't know what the bus speed is on any other PC 970 chip, just that the 1.8GHz model will be set at 900MHz. It is more likely that the 900MHz will go roughly across the entire 970 line than that the FSB speed will be exactly half the processor speed (FSB speeds don't scale as easily as processor speeds, and the main thing to consider here is that a faster FSB means a redesigned system controller chip, which is fairly expensive for speed-bumping).

The 200MHz memory bus is pure supposition right now, as that is completely in Apple's domain, and as you know Apple doesn't give details out about what it is planning. However, the choices there are a 200MHz bus using DDR-400 memory (which is essentially what top-of-the-line PCs will be using soon), a 167MHz bus using DDR-333 memory, a 133MHz bus using DDR-266 memory, or a 100MHz bus using DDR-200. If memory serves, the current top-of-the-line Macs use DDR-333 on a 167MHz memory bus (which is the same frequency as the FSB to the G4).