1. If the actual performance of the 970 outstrips our wildest expectations, should Apple restrain itself from putting all the bells and whistles onboard, allowing itself some performance cushion to respond to Intel/AMD as they modify their offerings, when necessary?
Apple should build the best machine it can at each price point. The 970 is fast, will scale in clockrate well, and is very well suited to SMP. The point-to-point FSB means that Apple can scale system performance by adding more processors and doing whatever it can to improve memory bandwidth. If cost reducing the low end means having less than a dual channel DDR400 memory system, then so be it... the chip could still be competitive at low price points with equivalent PC systems.
Quote:
2. Should L3 either embededded in the chipset or otherwise only be availble in the highend?
If it is embedded in the chipset then probably not because it is in Apple's interest to keep the number of chips they have to design and build to a minimum. If it is an external memory chip as suggested earlier then it could be a high-end only feature. If they have to have differing chips for other reasons (e.g. number of FSB ports) then they could also vary the on-chip L3, if they even decide to use that technique.
Quote:
3. If the mid-range (dual or smp) trounces the best of Intel/AMD, what is the God-Box ie. fastest cpu(s) avail/fastest memory/most memory/most bandwidth (just to say you own it) really worth to you.
I'm sure there are people out there who will be really big money to buy an Apple box that is really really really fast. The more it costs the fewer that will buy it so having a range of machines that step up in price and performance is probably a good thing, but is it worth it to Apple to complicate their lineup and do the R&D?
Quote:
4. Given that Apple needs a loss leader bad, does the current line with rediculous price drops serve that purpose?
No. The current machines are too expensive to build. To build a low end machine they should make something like a headless eMac. Old expensive machines don't get cheaper just because new expensive ones arrive on the scene -- the computer business doesn't work that way.
Quote:
5. Should Apple take the first batch of available speeds sell them full spec.ed and reserve the 2.0 GHz and above for Workstation machines and sell that way until the 980 arrives 2H2004 if rumor serves correct, which might actually be a previous question asked a different manner.
I think this will depend on the quantities available and the cost of the high speed chips. If they can get enough to put 2 GHz in the high end, then do so. The workstations can use more processors rather than faster ones.
Maybe Programmer and Amorph or someone can help me understand or establish a better context for what appears to be a discussion on Apple's potential implementation of the 970.
Hey its me, someone...
1. I don't buy that rumor of a 970 X-Station.
2. No L3 Cache
3. A SP 970 will not trounce the fastest Intel machine. A 3 GHz P4 on a quad pumped 200MHz FSB with dual channel PC3200 RAM ist really a fast machine. But a DP 970 will run circles around it
4. Apple will sell the fastest machines they can make. If IBM can deliver 2 GHz 970 in quantities enough to start production of 2GHz PowerMac Apple will immediately do it.
5. Why should Apple drop the price of a product that will sell in quantities they even can not produce?
6. Apple should do this, Apple should do that. It has been written here tenthousand times. Here's mine: I hope Apple will devide the PowerMac line in a midrange (SP, small tower, prosumer market) and a highend (DP, big tower pro market). I take the small tower, please
7. What I missed in the ars articles are some words about bus speeds. Intel claims the P4 has a FSB800 with 6.4GB/sec bandwidth. That's a lie for customers. IBM only give it's 450 MHz bus the same 6.4GB/sec bandwidth after removing the overhead. Ars should have explained that (waiting for article III)...
1. If the actual performance of the 970 outstrips our wildest expectations, should Apple restrain itself from putting all the bells and whistles onboard, allowing itself some performance cushion to respond to Intel/AMD as they modify their offerings, when necessary?
The 970 should mature nicely, move to 90nm promptly, scale up again nicely, be replaced by the 980 promptly, etc. There's currently no reason for Apple to hold back from offering the best machines that they can. They'll have a steady stream of steadily improving CPUs to choose from.
However, what that means is not the fastest machine they can possibly engineer. Apple uses commonly available components whenever possible, to make things easier and less expensive both for themselves and for their users. That means that, for example, Apple might choose to go with DDR333 rather than DDR400 simply because the former is standard, common, and reasonably priced.
Quote:
2. Should L3 either embededded in the chipset or otherwise only be availble in the highend?
The only way to determine this, really, is to consider the implementation options and their associated costs (as Programmer described) and to run actual tests to benchmark the improvements. It might be that the low end has an on-chip cache in the memory controller and the high end doesn't, because the low end has single-channel RAM and the high-end has dual (for example). It should be included when it is affordable and when it is needed.
Currently, it's useful because the bus to the CPU is a bottleneck. That's much less of a problem with the 970.
Quote:
3. If the mid-range (dual or smp) trounces the best of Intel/AMD, what is the God-Box ie. fastest cpu(s) avail/fastest memory/most memory/most bandwidth (just to say you own it) really worth to you.
To me? Nothing. I'm quite happy with the performance of my 450MHz Cube, thanks.
People who work in video and 3D and in scientific research will snap up the fastest thing Apple can make and demand more - really, until a machine can render Pixar movie frames in real time, there's always room for improvement (and even then, they'll be someone clamoring for a machine that can run their high-resolution hydraulic simulation of the Earth's oceans in real time...). They're mostly used to paying for $10,000 Suns and the like.
I should point out something important here: There are two price scales to consider. There's what I'll call a consumer price scale, and a professional price scale.
The consumer price scale is defined by the fact that the machine doesn't earn any money for its owner: It's a luxury item, a pure expense, and so there's a great deal of downward pressure on price on this scale. This is important because one of the reasons PowerMacs are considered exorbitant is that when you measure them on this scale (e.g., for gaming), they put a real dent in your wallet.
The professional price scale is defined by the fact that the machine earns money for its owner: It's a production tool. Given this, there is very little price pressure, because at most professional rates, a $1000 price difference for a somewhat faster machine can pay for itself in a week or two. Here the price/performance equation is tilted heavily toward absolute performance (including maintainability and reliability, not just speed). Viewed on a professional scale, the PowerMac prices aren't quite so bad, and if the top end machine offers, say, 40% more performance for $1000 more, that'll look like a pretty good deal: 40% more work means 40% more billable hours, which means the machine pays for itself quickly and then earns the owner that much more income. This is why Apple (and others) offer "Ultimate" configurations whose appeal seems incomprehensible to consumers. This is the scale upon which the PowerMac and the PowerBook are priced.
Quote:
4. Given that Apple needs a loss leader bad, does the current line with rediculous price drops serve that purpose?
Apple does not need a loss leader at all. What you seem to be proposing would fall more into the low end (consumer) price structure that you claim not to be interested in. The "headless eMac," as Programmer opined, or even just the eMac/iMac as they currently are.
The discussion has centered on PowerMacs partly because it's just really nice to be able to explore high-end performance after the last three years, and largely because PowerMacs are almost universally expected to be the first recipients of the 970 technology.
Quote:
5. Should Apple take the first batch of available speeds sell them full spec.ed and reserve the 2.0 GHz and above for Workstation machines and sell that way until the 980 arrives 2H2004 if rumor serves correct, which might actually be a previous question asked a different manner.
The PowerMac is the Apple workstation, unless there's a tectonic shift within Apple. The high-end workstation market (Sun, SGI, HP, IBM, etc.) is slowly dying, so there's not much point going there. High-performance computing is rapidly adopting the Beowulf model of clusters of inexpensive machines, and I think that this is where Apple will head. It plays to their strengths. As for the parts they include, Apple will use whatever it can get in sufficient quantities to meet the sales and cost projections for its models, and it's impossible to say exactly what meets those qualifications without access to a lot of data that nobody here will ever be privy to (unless one of the top Apple execs is a lurker here ).
If anything, Apple will be slow to update the Xserve, or to introduce any companions. The server market likes proven architectures, and they'd probably prefer to see the 970 and its boards battletested and debugged in workstations for a while before they'd consider buying a server based on that technology. The server market is conservative, for good reason: Servers have to be reliable above all.
Your diagram completely misses the important link -- off to the left of the system ASIC should be the connection to RAM. This will likely be no more than a 128-bit wide DDR400 connection.
Look at my diagram again. There are two connections to "PC3200" going out the bottom.
Quote:
Impressive, to be sure, but still probably less than 6 GB/sec. ~16 GB/sec worth of demands on a 6 GB/sec. Hmmmm, might there be room for about 10 GB/sec of improvement?
If Apple delivers a 6.4 GB/s main memory solution, I don't think an inline L3 cache will help much, especially on uniprocessor systems.
For dual systems, there will a need for L3 cache, but I think Apple (and IBM) will live without it. For quad systems, if a beast ever ships, will definitely need one.
Quote:
It is very likely that Apple will say that a 5x improvement (in processor bandwidth) is "enough for now" and just ship exactly that.
That's what I'm thinking. Heck, Intel doesn't even do it for their dual-Xeon systems.
Look at my diagram again. There are two connections to "PC3200" going out the bottom.
Heh, and I'm complaining about you not reading. Sheesh. I misinterpreted your diagram as being dual processor. Okay, add a second processor on the left of your diagram and you'll see what I mean. Sorry.
We "know" that Apple is going to ship a dual, according to rumours and the last couple years of G4s. These will be the interesting machines, IMO.
You might not buy it, but lots of the Pro market would. High-end PowerMac / Xstation, it doesn't really matter -- the same principle applies. Apple has had an "Ultimate" configuration on their store for a long time, but it has never really had "an edge".
Quote:
2. No L3 Cache
If they do an Xstation there probably will be. If they don't then maybe, maybe not.
Quote:
3. A SP 970 will not trounce the fastest Intel machine. A 3 GHz P4 on a quad pumped 200MHz FSB with dual channel PC3200 RAM ist really a fast machine. But a DP 970 will run circles around it
"trounce" is probably optimistic, but they aren't going to be far apart.
Quote:
5. Why should Apple drop the price of a product that will sell in quantities they even can not produce?
Amen to that. On the other hand, they will probably hold the price points where they are, otherwise the barrier to entry for a lot of people is just too high.
Quote:
7. What I missed in the ars articles are some words about bus speeds. Intel claims the P4 has a FSB800 with 6.4GB/sec bandwidth. That's a lie for customers. IBM only give it's 450 MHz bus the same 6.4GB/sec bandwidth after removing the overhead. Ars should have explained that (waiting for article III)...
I don't know anything about that Intel bus -- is it a packet oriented bus as well, or is it a pipelined address+data bus like previous FSBs? In that case its not a lie and it has the advantage of being able to exceed 3.2 GB/sec in either direction.
What a lot of people should know is that chip for chip (ie. single processor) at say 1.8 GHz. the PPC 970 will definitely keep up with anything that will be out there from Intel. I've been lurking on the Ars site and from what I can tell, all the comparisons are being done comparing a single 970 to a single Intel CPU. The bottom line is that even in a single config, the 970 should be more than competitive.
Anything that's AltiVec enabled should literally embarrass the competition with that bandwidth. In a dual config or perhaps a very *slight* chance a quad-config the Intel offerings start to look really pathetic. Then there is the heat and all the other nasty things the Intel side has to deal with ;-) And this is only the beginning of a new line of processors...
I expect Apple to take advantage of the fact that this thing was designed with SMP configs in mind. Not only that, it's not taking into consideration any of the other system enhancements we are likely to get. Imagine if Apple finds a way to simply throw more processors at a task. I know.. all the complexity of such a system... But I keep thinking about the "ants and the elephant" scenario... The ants eventually win out. And I think this is exactly what Apple is banking on. They had 6-way configured Macs years ago, I see no reason why something similar couldn't be brought to the table today since the CPUs are obviously better designed for it. Now, you or I and most consumers might not need such a machine, but I can see a machine like that selling to the professional sectors as well as science, research, medicine etc....
Apple has never before had a processor that they were using in their professional Macs which could be scaled in an essentially unlimited fashion. Previous PowerPCs used a shared bus with limited bandwidth and cache potential. Before that the 680x0 family was not multi-processor capable and never really scaled to the huge clock speeds of their day.
With the 970 and OS X's multi-processor capabilities and new technologies like RapidIO or HyperTransport they have the capability to scale up to many processors (e.g. 16), and larger clusters beyond that. They may choose not to pursue this since its not their traditional market, but for the first time it is an option that is open to Apple and there is a clear and obvious path to get there (i.e. they don't have to adopt a different processor or a radically new architecture). Nonetheless the potential for the 970 is there, and we'll probably see at least IBM exploit it. They could build a machine that costs $5000, $10000, $15000, etc ... and have it perform like you'd expect for that amount of money.
Look out Sun, SGI, HP -- Apple might come barking up your tree soon, and this time you'd better think twice before laughing about it. If that's not bad enough (for them), we might find Apple teamed up with IBM on more than just the processor.
Ed M brings up an excellent point. I can see it now:
(begin dream sequence)
They bring back the Photoshop/Render/whatever shoot out, and show the Mac doing well against a 3GHz Pentium. Then Steve says, "oh, I forgot to tell you...that is the new iMac. Our Towers are so much faster, it isn't even worth compairing".
(end dream sequence).
Most of the talk has centered on the highend systems, but I think the 970 will actually help the lowend just as much, if not more. Finally, the low end wont be artificially constrained by the high end. A single 970 in an iMac would be fantastic. If the IBM chips do scale as well as many think, it would allow apple to start spacing out their products better, and add models, too: very highend workstations (2-4 or more processors at highest speed), towers (1-2 processors, mid to high speed), intermediate machine (1 processor, mid to high speed), and finally the i/eMac (1 processor, up to mid speed). Right now there is no way to have a high end, and no space for a mid-level machine.
Ed M brings up an excellent point. I can see it now:
(begin dream sequence)
They bring back the Photoshop/Render/whatever shoot out, and show the Mac doing well against a 3GHz Pentium. Then Steve says, "oh, I forgot to tell you...that is the new iMac. Our Towers are so much faster, it isn't even worth compairing".
(end dream sequence).
Well, I could see him doing such a thing - though I'd think that a well selected bakeoff with a single 1.8 970 would crush a 3GHz P4 just due to Altivec.
Programmer I think is on the right path here. It's not OS X or the technical details of 970 that leads me to think that Apple will soon be gunning for SGI and HP, rather the nature of the software acquistions Apple made last year. In the end for Apple, the winning point won't be over systems that compare well with Wintel, but systems that don't exist in the Wintel world. I can image the CAD/CAE boys would be interested to see a high-performance 64 bit, 4-way system with hardware accelerated X11. I wonder if Apple's talked to MSC or Dassault...
Look out Sun, SGI, HP -- Apple might come barking up your tree soon, and this time you'd better think twice before laughing about it. If that's not bad enough (for them), we might find Apple teamed up with IBM on more than just the processor.
Heh heh! Therein lies the truth. However I think "might" may be a slight understatement. I really believe they already have. No I don't have proof, or insider information, just an underlying gut feeling.
Fascinating discussion, guys! What really intrigues me about the impending 970-based towers is that for the first time in years, Apple's design decisions will most likely be the limiting factor in performance. No more hiding behind Moto's ineptitude--although I don't think Apple actually were doing that, but now Apple is forced to lay their hand on the table.
My only thoughts on this is that it would be in Apple's best interest to make at least one "God system" regardless of it's sales potential. I hate car analogies but in this case it fits, Apple needs a Dodge Viper or Chevy Corvette of a tower; they've been making Neons for so long that few associate Apple with performance. Build a God system, hype it with lots of PR, and suddenly Apple's reputation gets a crash-makeover. Reviewers benchmark everything from Maya to Doom 3 on this God system and Wintel users begin to get penis envy, then they price it out and see it's out of their reach, but hey, that midrange dual 970 system is within reach and even it's faster than my 3 GHz PIV system.
By "God system", I mean either quad CPU or dual CPU w/ heavily optimized muthaboard that can actually feed the 970's monster bus.
The main question for me is, will Apple actually do this? The Xstation rumors sort of fit in with this, but if the God system is too expensive, then it's PR impact seems limited. Nobody cares how Doom 3 runs on a $10000 system, but on a $4000 system gamers will start to notice, and same goes for more casual users of FCP and Bryce type apps.
Some good points in general from Thai, Prog' and JD and Co.
I can see quite an alliance between IBM and Apple. From staving off Intel in server markets to creative markets.
Linux may do cluster rendering. But I can Apple taking a sizeable chunk of 3D creative markets with Panther and 970s...and X-serves and X-stations.
Apple will have the option of single to dual to quad and octo based systems.
That's alot of solutions for alot of markets.
I see no reason why the Towers can't be single to all dual. With dual to quad systems defining a 'Workstation' segment that Apple never really had with current towers. The towers can have a slight price cut and a new entray X-station line above. They'll get their cream there. And with motherboard production in Taiwanese houses then I can see Apple being serious about (Hello, REALISTIC!) pricing, aggressive price points for growth and competing with Wintel. They said so themselves. Mhz and price matter. That's from them! And with 2% and shrinking...I'm sure they're aware they really need to pull their finger out and work it to that holy grail 10%. How far are Towers from Cube like sales? No very at their current rate of sales decline. Guess what, they eventually canned the Cube. But seeing as Freddy Andy said that Apple were now focused on the tower problem...the 970 is a given.
With cpus like the 970, 980 etc and a G3+, there is no reason why mid and top range consumer machines like 12 Powerbook and iMac can't have single 1.4-1.8 970s leaving the dual 970s for the tower line.
I would have guessed that the 'custom' runs of current 'high end' G4s must be quite pricey to include in Apple's machines. It's all artificially stifling.
The 970 might be cheaper to produce. Like others have said, it's going to give Apple the first chance to really, no, I mean REALLY diff' their machines.
From Doom III to Lightwave to Photoshop to servers to consumer. Watch out Intel. You know when Apple's high end becomes the low end every six months?
For the 1st time, Apple will have a 1.8 970 (equiv' of at least a 3.6 P4 I'd vouch...) that will become 'low end' after the 970+ die shrink...and be faster than Intel's low end...or high end by the looks of things! That's one hell of a turnaround! Apple's eMacs, iMacs and low end towers could end up trouncing the 'cheap' Dell market in performance. And with the 980 pending...there will be no end to Wintel's world of hurt.
In the next year, I'll be quite happy to start talking mhz myths...
JYD - glad to see you apparently have a real life now based on your infrequent contributions.
Your "God System" was the Macintosh Anniversary model, and I think your right - we're going to see a new one. Your price point may be low, but it will still accomplish the goal.
Ehh I don't really think the 20th anniversary Mac was a God system. It was a system for rich fanboys who wanted something stylin. IIRC, it only had a 200MHz 603e, not that great a processor in the day.
Apple ought to build something of a 9600/350 or Power Tower Pro 275(G3...I don't think it ever shipped) stature. Those were some serious f'ing machines
I leave the word to Donatello which proposes his analysis of the situation concerning the sources of Darwin and on the influence of materials used in the manufacture of Power4 and PPC 970 to us.
In connection with Darwin, it is interesting to note that the sources are not updated any more since February. Now APPLE diffuses only the sources corresponding to the official versions of MacOSX. For example, Darwin 6.6 for OS X 10.2.6.
Previously, APPLE diffused also the sources of the versions under development on their waiters CVS. Ainsi last May, one had already access to the sources of the future Jaguar. APPLE took however time to remove the parts of the code written to support products not left.
I imagine that if they added the support of PowerPC 970 in Darwin. It is not a small cleaning which it is necessary to subject the code to hide these modifications.
One can then imagine that with the support of PowerPC 970, it had become impossible to hide these modifications. And it east can be why APPLE is limited since February to diffuse the sources of the official versions.
The tools for developers underwent the same treatment. One has the impression that APPLE stopped working above since December.
With my opinion, the exit of Macintosh containing PowerPC 970 is certain. But APPLE wants to preserve the effect of advertisement.
Another thing, I had confirmation that it was the material with low K, leSiLK, which was at the base of new Power4+ with 1,7 GHz. They must leave in July. With same material, PowerPC 970 should turn to 2,3 GHz and more.
Approximately, PowerPC 970 it is finer engraved Power4+. There is not same the exigeances reliability for this type of processor. But more one processor is engraved fine more it is fast. The problem it is that one reaches others then limit physical. And it is then necessary to facilitate the passage of the electrons by using materials with low K For IBM, it is SiLK for engraving in 0,13 um. And still new material for engraving in 0,09 um.
As PowerPC 970 is engraved finer than Power4+, it will profit more this new material. And it should reach in an immediate future 2,5 GHz as IBM announced one moment. At the end of 2003, at the beginning of 2004, it should profit from engraving with 0,09 um. And at the end of 2004, IBM will add le"strained silicon to it ". The goal is always to accelerate the mobility of the electrons. Power5+ will reach 3 then GHz. And PowerPC 970/980 peutêtre 4,5 GHz. If the reports/ratios are maintienent.
Ehh I don't really think the 20th anniversary Mac was a God system. It was a system for rich fanboys who wanted something stylin. IIRC, it only had a 200MHz 603e, not that great a processor in the day.
Apple ought to build something of a 9600/350 or Power Tower Pro 275(G3...I don't think it ever shipped) stature. Those were some serious f'ing machines
I agree! I meant that Apple is willing to spend money on a high profile system with a small market (= small or negative return) for the prestege.
And at the end of 2004, IBM will add le"strained silicon to it ". The goal is always to accelerate the mobility of the electrons. Power5+ will reach 3 then GHz. And PowerPC 970/980 peutêtre 4,5 GHz. If the reports/ratios are maintienent.
If the 970/980 would reach 4.5ghz by the end of 2004, that would mean more than doubling the performance in about a year. I doubt very much that we will see this kind of speed increase in the first year. Heck, I mean just going from 1.8ghz this july/august to about 3 ghz in a year would be very impressive. But 4.5 ghz ???
This has the ring of truth about it. We might see 2.5 GHz out of a 0.13 micron 970, and it'll easily make 2 GHz at or near introduction. The absence of substantial changes to the OS and associated tools from Apple is also what we would expect leading up to Panther. The speeds of future processors are highly speculative (and optimistic), but IBM's process technology is some of the best in the world. Happy times ahead for Apple and its fans, I think.
Comments
1. If the actual performance of the 970 outstrips our wildest expectations, should Apple restrain itself from putting all the bells and whistles onboard, allowing itself some performance cushion to respond to Intel/AMD as they modify their offerings, when necessary?
Apple should build the best machine it can at each price point. The 970 is fast, will scale in clockrate well, and is very well suited to SMP. The point-to-point FSB means that Apple can scale system performance by adding more processors and doing whatever it can to improve memory bandwidth. If cost reducing the low end means having less than a dual channel DDR400 memory system, then so be it... the chip could still be competitive at low price points with equivalent PC systems.
2. Should L3 either embededded in the chipset or otherwise only be availble in the highend?
If it is embedded in the chipset then probably not because it is in Apple's interest to keep the number of chips they have to design and build to a minimum. If it is an external memory chip as suggested earlier then it could be a high-end only feature. If they have to have differing chips for other reasons (e.g. number of FSB ports) then they could also vary the on-chip L3, if they even decide to use that technique.
3. If the mid-range (dual or smp) trounces the best of Intel/AMD, what is the God-Box ie. fastest cpu(s) avail/fastest memory/most memory/most bandwidth (just to say you own it) really worth to you.
I'm sure there are people out there who will be really big money to buy an Apple box that is really really really fast. The more it costs the fewer that will buy it so having a range of machines that step up in price and performance is probably a good thing, but is it worth it to Apple to complicate their lineup and do the R&D?
4. Given that Apple needs a loss leader bad, does the current line with rediculous price drops serve that purpose?
No. The current machines are too expensive to build. To build a low end machine they should make something like a headless eMac. Old expensive machines don't get cheaper just because new expensive ones arrive on the scene -- the computer business doesn't work that way.
5. Should Apple take the first batch of available speeds sell them full spec.ed and reserve the 2.0 GHz and above for Workstation machines and sell that way until the 980 arrives 2H2004 if rumor serves correct, which might actually be a previous question asked a different manner.
I think this will depend on the quantities available and the cost of the high speed chips. If they can get enough to put 2 GHz in the high end, then do so. The workstations can use more processors rather than faster ones.
Originally posted by ArkAngel
Maybe Programmer and Amorph or someone can help me understand or establish a better context for what appears to be a discussion on Apple's potential implementation of the 970.
Hey its me, someone...
1. I don't buy that rumor of a 970 X-Station.
2. No L3 Cache
3. A SP 970 will not trounce the fastest Intel machine. A 3 GHz P4 on a quad pumped 200MHz FSB with dual channel PC3200 RAM ist really a fast machine. But a DP 970 will run circles around it
4. Apple will sell the fastest machines they can make. If IBM can deliver 2 GHz 970 in quantities enough to start production of 2GHz PowerMac Apple will immediately do it.
5. Why should Apple drop the price of a product that will sell in quantities they even can not produce?
6. Apple should do this, Apple should do that. It has been written here tenthousand times. Here's mine: I hope Apple will devide the PowerMac line in a midrange (SP, small tower, prosumer market) and a highend (DP, big tower pro market). I take the small tower, please
7. What I missed in the ars articles are some words about bus speeds. Intel claims the P4 has a FSB800 with 6.4GB/sec bandwidth. That's a lie for customers. IBM only give it's 450 MHz bus the same 6.4GB/sec bandwidth after removing the overhead. Ars should have explained that (waiting for article III)...
8. Why does Programmer always beat me?
Originally posted by ArkAngel
1. If the actual performance of the 970 outstrips our wildest expectations, should Apple restrain itself from putting all the bells and whistles onboard, allowing itself some performance cushion to respond to Intel/AMD as they modify their offerings, when necessary?
The 970 should mature nicely, move to 90nm promptly, scale up again nicely, be replaced by the 980 promptly, etc. There's currently no reason for Apple to hold back from offering the best machines that they can. They'll have a steady stream of steadily improving CPUs to choose from.
However, what that means is not the fastest machine they can possibly engineer. Apple uses commonly available components whenever possible, to make things easier and less expensive both for themselves and for their users. That means that, for example, Apple might choose to go with DDR333 rather than DDR400 simply because the former is standard, common, and reasonably priced.
2. Should L3 either embededded in the chipset or otherwise only be availble in the highend?
The only way to determine this, really, is to consider the implementation options and their associated costs (as Programmer described) and to run actual tests to benchmark the improvements. It might be that the low end has an on-chip cache in the memory controller and the high end doesn't, because the low end has single-channel RAM and the high-end has dual (for example). It should be included when it is affordable and when it is needed.
Currently, it's useful because the bus to the CPU is a bottleneck. That's much less of a problem with the 970.
3. If the mid-range (dual or smp) trounces the best of Intel/AMD, what is the God-Box ie. fastest cpu(s) avail/fastest memory/most memory/most bandwidth (just to say you own it) really worth to you.
To me? Nothing. I'm quite happy with the performance of my 450MHz Cube, thanks.
People who work in video and 3D and in scientific research will snap up the fastest thing Apple can make and demand more - really, until a machine can render Pixar movie frames in real time, there's always room for improvement (and even then, they'll be someone clamoring for a machine that can run their high-resolution hydraulic simulation of the Earth's oceans in real time...). They're mostly used to paying for $10,000 Suns and the like.
I should point out something important here: There are two price scales to consider. There's what I'll call a consumer price scale, and a professional price scale.
The consumer price scale is defined by the fact that the machine doesn't earn any money for its owner: It's a luxury item, a pure expense, and so there's a great deal of downward pressure on price on this scale. This is important because one of the reasons PowerMacs are considered exorbitant is that when you measure them on this scale (e.g., for gaming), they put a real dent in your wallet.
The professional price scale is defined by the fact that the machine earns money for its owner: It's a production tool. Given this, there is very little price pressure, because at most professional rates, a $1000 price difference for a somewhat faster machine can pay for itself in a week or two. Here the price/performance equation is tilted heavily toward absolute performance (including maintainability and reliability, not just speed). Viewed on a professional scale, the PowerMac prices aren't quite so bad, and if the top end machine offers, say, 40% more performance for $1000 more, that'll look like a pretty good deal: 40% more work means 40% more billable hours, which means the machine pays for itself quickly and then earns the owner that much more income. This is why Apple (and others) offer "Ultimate" configurations whose appeal seems incomprehensible to consumers. This is the scale upon which the PowerMac and the PowerBook are priced.
4. Given that Apple needs a loss leader bad, does the current line with rediculous price drops serve that purpose?
Apple does not need a loss leader at all. What you seem to be proposing would fall more into the low end (consumer) price structure that you claim not to be interested in.
The discussion has centered on PowerMacs partly because it's just really nice to be able to explore high-end performance after the last three years, and largely because PowerMacs are almost universally expected to be the first recipients of the 970 technology.
5. Should Apple take the first batch of available speeds sell them full spec.ed and reserve the 2.0 GHz and above for Workstation machines and sell that way until the 980 arrives 2H2004 if rumor serves correct, which might actually be a previous question asked a different manner.
The PowerMac is the Apple workstation, unless there's a tectonic shift within Apple. The high-end workstation market (Sun, SGI, HP, IBM, etc.) is slowly dying, so there's not much point going there. High-performance computing is rapidly adopting the Beowulf model of clusters of inexpensive machines, and I think that this is where Apple will head. It plays to their strengths. As for the parts they include, Apple will use whatever it can get in sufficient quantities to meet the sales and cost projections for its models, and it's impossible to say exactly what meets those qualifications without access to a lot of data that nobody here will ever be privy to (unless one of the top Apple execs is a lurker here
If anything, Apple will be slow to update the Xserve, or to introduce any companions. The server market likes proven architectures, and they'd probably prefer to see the 970 and its boards battletested and debugged in workstations for a while before they'd consider buying a server based on that technology. The server market is conservative, for good reason: Servers have to be reliable above all.
Originally posted by Programmer
Did you even bother to read what I wrote?
Yes, I did. Maybe I've missed something.
Your diagram completely misses the important link -- off to the left of the system ASIC should be the connection to RAM. This will likely be no more than a 128-bit wide DDR400 connection.
Look at my diagram again. There are two connections to "PC3200" going out the bottom.
Impressive, to be sure, but still probably less than 6 GB/sec. ~16 GB/sec worth of demands on a 6 GB/sec. Hmmmm, might there be room for about 10 GB/sec of improvement?
If Apple delivers a 6.4 GB/s main memory solution, I don't think an inline L3 cache will help much, especially on uniprocessor systems.
For dual systems, there will a need for L3 cache, but I think Apple (and IBM) will live without it. For quad systems, if a beast ever ships, will definitely need one.
It is very likely that Apple will say that a 5x improvement (in processor bandwidth) is "enough for now" and just ship exactly that.
That's what I'm thinking. Heck, Intel doesn't even do it for their dual-Xeon systems.
Originally posted by THT
Yes, I did. Maybe I've missed something.
Look at my diagram again. There are two connections to "PC3200" going out the bottom.
Heh, and I'm complaining about you not reading. Sheesh. I misinterpreted your diagram as being dual processor. Okay, add a second processor on the left of your diagram and you'll see what I mean. Sorry.
We "know" that Apple is going to ship a dual, according to rumours and the last couple years of G4s. These will be the interesting machines, IMO.
Originally posted by someone
1. I don't buy that rumor of a 970 X-Station.
You might not buy it, but lots of the Pro market would.
2. No L3 Cache
If they do an Xstation there probably will be. If they don't then maybe, maybe not.
3. A SP 970 will not trounce the fastest Intel machine. A 3 GHz P4 on a quad pumped 200MHz FSB with dual channel PC3200 RAM ist really a fast machine. But a DP 970 will run circles around it
"trounce" is probably optimistic, but they aren't going to be far apart.
5. Why should Apple drop the price of a product that will sell in quantities they even can not produce?
Amen to that. On the other hand, they will probably hold the price points where they are, otherwise the barrier to entry for a lot of people is just too high.
7. What I missed in the ars articles are some words about bus speeds. Intel claims the P4 has a FSB800 with 6.4GB/sec bandwidth. That's a lie for customers. IBM only give it's 450 MHz bus the same 6.4GB/sec bandwidth after removing the overhead. Ars should have explained that (waiting for article III)...
I don't know anything about that Intel bus -- is it a packet oriented bus as well, or is it a pipelined address+data bus like previous FSBs? In that case its not a lie and it has the advantage of being able to exceed 3.2 GB/sec in either direction.
8. Why does Programmer always beat me?
Walk softly and carry a big stick.
Being a fast typist helps too.
What a lot of people should know is that chip for chip (ie. single processor) at say 1.8 GHz. the PPC 970 will definitely keep up with anything that will be out there from Intel. I've been lurking on the Ars site and from what I can tell, all the comparisons are being done comparing a single 970 to a single Intel CPU. The bottom line is that even in a single config, the 970 should be more than competitive.
Anything that's AltiVec enabled should literally embarrass the competition with that bandwidth. In a dual config or perhaps a very *slight* chance a quad-config the Intel offerings start to look really pathetic. Then there is the heat and all the other nasty things the Intel side has to deal with ;-) And this is only the beginning of a new line of processors...
I expect Apple to take advantage of the fact that this thing was designed with SMP configs in mind. Not only that, it's not taking into consideration any of the other system enhancements we are likely to get. Imagine if Apple finds a way to simply throw more processors at a task. I know.. all the complexity of such a system... But I keep thinking about the "ants and the elephant" scenario... The ants eventually win out. And I think this is exactly what Apple is banking on. They had 6-way configured Macs years ago, I see no reason why something similar couldn't be brought to the table today since the CPUs are obviously better designed for it. Now, you or I and most consumers might not need such a machine, but I can see a machine like that selling to the professional sectors as well as science, research, medicine etc....
--
Ed
Originally posted by Ed M.
Programmer....
<<good comments omitted for brevity>>
Well said.
Apple has never before had a processor that they were using in their professional Macs which could be scaled in an essentially unlimited fashion. Previous PowerPCs used a shared bus with limited bandwidth and cache potential. Before that the 680x0 family was not multi-processor capable and never really scaled to the huge clock speeds of their day.
With the 970 and OS X's multi-processor capabilities and new technologies like RapidIO or HyperTransport they have the capability to scale up to many processors (e.g. 16), and larger clusters beyond that. They may choose not to pursue this since its not their traditional market, but for the first time it is an option that is open to Apple and there is a clear and obvious path to get there (i.e. they don't have to adopt a different processor or a radically new architecture). Nonetheless the potential for the 970 is there, and we'll probably see at least IBM exploit it. They could build a machine that costs $5000, $10000, $15000, etc ... and have it perform like you'd expect for that amount of money.
Look out Sun, SGI, HP -- Apple might come barking up your tree soon, and this time you'd better think twice before laughing about it. If that's not bad enough (for them), we might find Apple teamed up with IBM on more than just the processor.
(begin dream sequence)
They bring back the Photoshop/Render/whatever shoot out, and show the Mac doing well against a 3GHz Pentium. Then Steve says, "oh, I forgot to tell you...that is the new iMac. Our Towers are so much faster, it isn't even worth compairing".
(end dream sequence).
Most of the talk has centered on the highend systems, but I think the 970 will actually help the lowend just as much, if not more. Finally, the low end wont be artificially constrained by the high end. A single 970 in an iMac would be fantastic. If the IBM chips do scale as well as many think, it would allow apple to start spacing out their products better, and add models, too: very highend workstations (2-4 or more processors at highest speed), towers (1-2 processors, mid to high speed), intermediate machine (1 processor, mid to high speed), and finally the i/eMac (1 processor, up to mid speed). Right now there is no way to have a high end, and no space for a mid-level machine.
[edit: delete unnecessary items
Originally posted by Thai Moof
Ed M brings up an excellent point. I can see it now:
(begin dream sequence)
They bring back the Photoshop/Render/whatever shoot out, and show the Mac doing well against a 3GHz Pentium. Then Steve says, "oh, I forgot to tell you...that is the new iMac. Our Towers are so much faster, it isn't even worth compairing".
(end dream sequence).
Well, I could see him doing such a thing - though I'd think that a well selected bakeoff with a single 1.8 970 would crush a 3GHz P4 just due to Altivec.
Programmer I think is on the right path here. It's not OS X or the technical details of 970 that leads me to think that Apple will soon be gunning for SGI and HP, rather the nature of the software acquistions Apple made last year. In the end for Apple, the winning point won't be over systems that compare well with Wintel, but systems that don't exist in the Wintel world. I can image the CAD/CAE boys would be interested to see a high-performance 64 bit, 4-way system with hardware accelerated X11. I wonder if Apple's talked to MSC or Dassault...
Originally posted by Programmer
Look out Sun, SGI, HP -- Apple might come barking up your tree soon, and this time you'd better think twice before laughing about it. If that's not bad enough (for them), we might find Apple teamed up with IBM on more than just the processor.
Heh heh! Therein lies the truth. However I think "might" may be a slight understatement. I really believe they already have. No I don't have proof, or insider information, just an underlying gut feeling.
My only thoughts on this is that it would be in Apple's best interest to make at least one "God system" regardless of it's sales potential. I hate car analogies but in this case it fits, Apple needs a Dodge Viper or Chevy Corvette of a tower; they've been making Neons for so long that few associate Apple with performance. Build a God system, hype it with lots of PR, and suddenly Apple's reputation gets a crash-makeover. Reviewers benchmark everything from Maya to Doom 3 on this God system and Wintel users begin to get penis envy, then they price it out and see it's out of their reach, but hey, that midrange dual 970 system is within reach and even it's faster than my 3 GHz PIV system.
By "God system", I mean either quad CPU or dual CPU w/ heavily optimized muthaboard that can actually feed the 970's monster bus.
The main question for me is, will Apple actually do this? The Xstation rumors sort of fit in with this, but if the God system is too expensive, then it's PR impact seems limited. Nobody cares how Doom 3 runs on a $10000 system, but on a $4000 system gamers will start to notice, and same goes for more casual users of FCP and Bryce type apps.
I can see quite an alliance between IBM and Apple. From staving off Intel in server markets to creative markets.
Linux may do cluster rendering. But I can Apple taking a sizeable chunk of 3D creative markets with Panther and 970s...and X-serves and X-stations.
Apple will have the option of single to dual to quad and octo based systems.
That's alot of solutions for alot of markets.
I see no reason why the Towers can't be single to all dual. With dual to quad systems defining a 'Workstation' segment that Apple never really had with current towers. The towers can have a slight price cut and a new entray X-station line above. They'll get their cream there. And with motherboard production in Taiwanese houses then I can see Apple being serious about (Hello, REALISTIC!) pricing, aggressive price points for growth and competing with Wintel. They said so themselves. Mhz and price matter. That's from them! And with 2% and shrinking...I'm sure they're aware they really need to pull their finger out and work it to that holy grail 10%. How far are Towers from Cube like sales? No very at their current rate of sales decline. Guess what, they eventually canned the Cube. But seeing as Freddy Andy said that Apple were now focused on the tower problem...the 970 is a given.
With cpus like the 970, 980 etc and a G3+, there is no reason why mid and top range consumer machines like 12 Powerbook and iMac can't have single 1.4-1.8 970s leaving the dual 970s for the tower line.
I would have guessed that the 'custom' runs of current 'high end' G4s must be quite pricey to include in Apple's machines. It's all artificially stifling.
The 970 might be cheaper to produce. Like others have said, it's going to give Apple the first chance to really, no, I mean REALLY diff' their machines.
From Doom III to Lightwave to Photoshop to servers to consumer. Watch out Intel. You know when Apple's high end becomes the low end every six months?
For the 1st time, Apple will have a 1.8 970 (equiv' of at least a 3.6 P4 I'd vouch...) that will become 'low end' after the 970+ die shrink...and be faster than Intel's low end...or high end by the looks of things! That's one hell of a turnaround! Apple's eMacs, iMacs and low end towers could end up trouncing the 'cheap' Dell market in performance. And with the 980 pending...there will be no end to Wintel's world of hurt.
In the next year, I'll be quite happy to start talking mhz myths...
Options? Hell yeah...
Lemon Bon Bon
Barto
Your "God System" was the Macintosh Anniversary model, and I think your right - we're going to see a new one. Your price point may be low, but it will still accomplish the goal.
Apple ought to build something of a 9600/350 or Power Tower Pro 275(G3...I don't think it ever shipped) stature. Those were some serious f'ing machines
Power4+ with 1,7 Ghz for July - Yoc - 11:06:52
I leave the word to Donatello which proposes his analysis of the situation concerning the sources of Darwin and on the influence of materials used in the manufacture of Power4 and PPC 970 to us.
In connection with Darwin, it is interesting to note that the sources are not updated any more since February. Now APPLE diffuses only the sources corresponding to the official versions of MacOSX. For example, Darwin 6.6 for OS X 10.2.6.
Previously, APPLE diffused also the sources of the versions under development on their waiters CVS. Ainsi last May, one had already access to the sources of the future Jaguar. APPLE took however time to remove the parts of the code written to support products not left.
I imagine that if they added the support of PowerPC 970 in Darwin. It is not a small cleaning which it is necessary to subject the code to hide these modifications.
One can then imagine that with the support of PowerPC 970, it had become impossible to hide these modifications. And it east can be why APPLE is limited since February to diffuse the sources of the official versions.
The tools for developers underwent the same treatment. One has the impression that APPLE stopped working above since December.
With my opinion, the exit of Macintosh containing PowerPC 970 is certain. But APPLE wants to preserve the effect of advertisement.
Another thing, I had confirmation that it was the material with low K, leSiLK, which was at the base of new Power4+ with 1,7 GHz. They must leave in July. With same material, PowerPC 970 should turn to 2,3 GHz and more.
Approximately, PowerPC 970 it is finer engraved Power4+. There is not same the exigeances reliability for this type of processor. But more one processor is engraved fine more it is fast. The problem it is that one reaches others then limit physical. And it is then necessary to facilitate the passage of the electrons by using materials with low K For IBM, it is SiLK for engraving in 0,13 um. And still new material for engraving in 0,09 um.
As PowerPC 970 is engraved finer than Power4+, it will profit more this new material. And it should reach in an immediate future 2,5 GHz as IBM announced one moment. At the end of 2003, at the beginning of 2004, it should profit from engraving with 0,09 um. And at the end of 2004, IBM will add le"strained silicon to it ". The goal is always to accelerate the mobility of the electrons. Power5+ will reach 3 then GHz. And PowerPC 970/980 peutêtre 4,5 GHz. If the reports/ratios are maintienent.
Here it is all that I know,
Donatello.
Originally posted by crayz
Ehh I don't really think the 20th anniversary Mac was a God system. It was a system for rich fanboys who wanted something stylin. IIRC, it only had a 200MHz 603e, not that great a processor in the day.
Apple ought to build something of a 9600/350 or Power Tower Pro 275(G3...I don't think it ever shipped) stature. Those were some serious f'ing machines
I agree! I meant that Apple is willing to spend money on a high profile system with a small market (= small or negative return) for the prestege.
And at the end of 2004, IBM will add le"strained silicon to it ". The goal is always to accelerate the mobility of the electrons. Power5+ will reach 3 then GHz. And PowerPC 970/980 peutêtre 4,5 GHz. If the reports/ratios are maintienent.
If the 970/980 would reach 4.5ghz by the end of 2004, that would mean more than doubling the performance in about a year. I doubt very much that we will see this kind of speed increase in the first year. Heck, I mean just going from 1.8ghz this july/august to about 3 ghz in a year would be very impressive. But 4.5 ghz ???