This is just the writer's being sloppy (which does seem to be rather common these days). As is clearly stated in the PPC 970MP user manual, the processor bus is double-pumped.
Well, no offense, but after the whole "my cordless phone works at several GHz" argument, I tend to think that that statement actually applies to you...
Similar maybe, but the complexity differs by orders of magnitude.
Of course the basic issues are the same; we're talking about physics, after all. It's just that the complexity of coping with them varies enormously depending on what exactly you try to do.
As an example, your typical run-of-the-mill serial bus uses LVDS with embedded clocks. As such, you can by definition never have bits coming in early or late, and don't have to worry about piecing them back together in the proper way -- which just so happens to be a huge deal for high-speed parallel buses.
Yeah. But just because it's feasible to hand-tune every trace on a simple MMIC doesn't mean the same applies to the complex, multi-layer PCBs found in computers. It's a totally different order of complexity.
In any case, whether you believe me or not, the point that nobody in the industry uses GHz-speed off-chip parallel buses should kinda tip you off to the fact that it's a lot less trivial than you seem to think.
Well, if I'm not mistaken, it was you who brought up the "my phone runs at GHz speeds" point...
First of all, I never said it would easy, like the old "rolling off the log" saying. But it isn't THAT impossible either. If Apple wanted to implement it they could. Would the cost go up slightly; sure.
Very few engineers would equate a cell's radio frequency circuitry with that of a magnetron in a microwave.
8 layer boards were developed for two purposes. One was to allow traces to cross each other in complex circuits. Two was to allow for shielding planes for high frequency uses. This is not new technology.
What you are also forgetting is that the G5 was designed originally to scale past 3GHz. The problems that beset the chip industry, which have little in common in our discussion here, has put an end to that, for the time being.
IBM had stated a couple of years ago that the successors to the 970 would get to 6GHz and beyond. With no reason to believe that the memory bus would change its relationship to the chip clock speed, we could be seeing the very 3GHz memory bus you are denying the possibility of.
It was just four years ago that the Express but, which runs at vastly higher speeds than the PCI bus, was thought of as being much too expensive to implement. That was one of the reasons why PCI X was developed (the other reason was opposition to another Intel standard).
As Gateway released a series of machines recently using the Express bus starting at $649, we can see how that prediction held up.
Your lack of confidence in modern manufacturing and engineering knowhow reminds me of the definitiveness of some in the industry over PCI X and Express.
Unfortunately, with Apple leaving IBM, we will never find out which one of us would be right.
You mentioned different versions, which implied that there are non-Linux versions. There aren't.
Hardly, seeing as Crossover is just an implementation of WINE with a GUI.
Of course they are, as the benefit of getting thousands of small Windows apps to run would be minimal. People want IE to test their pages on, Word for that file OpenOffice can't open (rarely) and they want to run Photoshop and perhaps iTunes (in Linux). Nobody cares for the entire library of Windows-compatible apps because there are very good replacements for them.
And CrossOver = WINE with a GUI slapped on it. I don't see how that is "more useful" in running those 75% of other applications, unless you mean that people can use a mouse now instead of typing words for programs to start. Yep, useful as hell.
I'm amazed sometimes how people can parse words so finely that the concept of what is being said is lost.
There is a version for the PPC Mac. If you went to the web site, you would have seen that for yourself.
Is it ready for distribution? Not at all! I didn't say it was. But I've played around with it myself, and could see that it had promise.
They have been working on the x86 version for several months. This is even further behind. When it will it be usable, I have no idea. But it most likely will be.
I'm not advocating its use. Nor am I stating that it will be the best way to go. If you bothered to read my posts on this, you would have seen that I'm somewhat skeptical about its general purpose replacement for Windows, for those who want an easy, generally useful Windows experience.
But the version for Linux works well enough for those willing to go through the hassle of setting it up and installing the programs that do work.
Crossover makes this process very easy. Like it or not, for what it's intended, it works well. Read the article I posted, and you can see that for yourself. He is, by no means, the only writer to publish on this issue.
There are those who buy A Mac these days to use the Unix environment rather than the GUI that Apple prefers we use. does that make those users wrong because many Mac programs wont work? No, it doesn't.
The mechanical clock (for want of a better way of saying it) is 1/4 speed. The bus is then double pumped to 1/2 speed. Then each directions' traffic is 1/4 speed. Add them together and you again have 1/2 speed.
Could you explain that "each directions' traffic is 1/4 speed"?
Intel will never make a 64 bit Yonah. that is strictly a 32 bit design. It will be replaced mid to late 2006 with the Merom, which WILL be a 64 bit design.
Intel stated at the last idf that Yonah will coexist to Merom till Meroms transition to 45nm process in 2007.
First of all, I never said it would easy, like the old "rolling off the log" saying. But it isn't THAT impossible either. If Apple wanted to implement it they could. Would the cost go up slightly; sure.
Very few engineers would equate a cell's radio frequency circuitry with that of a magnetron in a microwave.
Which was exactly my point -- equally few engineers would equate a cell's (or cordless phone's) RF circuitry to a high-speed processor interconnect.
(Oh, and for the record, yeah, having tubes in your cell would kinda ruin the whole operating-on-battery experience. )
Quote:
8 layer boards were developed for two purposes. One was to allow traces to cross each other in complex circuits. Two was to allow for shielding planes for high frequency uses. This is not new technology.
Of course it's not -- and yet nobody seems to be running their FSBs at several GHz. Doesn't that seem kindo odd to you, given that you seem to think it's an almost trivial thing to implement?
Quote:
What you are also forgetting is that the G5 was designed originally to scale past 3GHz. The problems that beset the chip industry, which have little in common in our discussion here, has put an end to that, for the time being.
I wholeheartedly agree that this has little to do with the discussion at hand.
Quote:
IBM had stated a couple of years ago that the successors to the 970 would get to 6GHz and beyond. With no reason to believe that the memory bus would change its relationship to the chip clock speed, we could be seeing the very 3GHz memory bus you are denying the possibility of.
What exactly makes you think the CPU to bus ratio would remain constant? Just take a look at how the clock frequencies of CPU cores and front side buses have been scaling for the past decade or so. Following that argument, we'd have been running our FSBs at a 1:1 ratio all along ever since the 8086... (Note that I'm talking about actual clock frequencies, not "it's DDR so we'll just go ahead and multiply the frequency by two in all our marketing brochures".)
Also, note that I'm not saying a 3 GHz FSB is somehow physically impossible to implement (that would be a rather stupid statement to make); it's just not feasible with current technology, and there would be little point for Apple to integrate such a capability in their northbridge when it's not likely to ever actually get used during the life cycle of said north bridge.
Quote:
It was just four years ago that the Express but, which runs at vastly higher speeds than the PCI bus, was thought of as being much too expensive to implement. That was one of the reasons why PCI X was developed (the other reason was opposition to another Intel standard).
To my knowledge, the major reason for PCI-X's coming into being was that, unlike PCIe, the former actually allowed for backwards compatibility with many existing PCI cards; I don't think anybody ever seriously expected it to become mainstream (and a quick glance at just what types of cards and motherboards are available as PCI-X versions kinda supports that notion...).
Also, as far as I'm aware, a single PCIe lane is actually cheaper to implement than PCI-X, while still offering much better performance than traditional PCI. No way to reuse your old PCI cards, tho.
Quote:
As Gateway released a series of machines recently using the Express bus starting at $649, we can see how that prediction held up.
Your lack of confidence in modern manufacturing and engineering knowhow reminds me of the definitiveness of some in the industry over PCI X and Express.
Well, see above for my take on that.
Quote:
Unfortunately, with Apple leaving IBM, we will never find out which one of us would be right.
True, I guess. Unless you somehow happen to run into that document again which you quoted earlier.
Could you explain that "each directions' traffic is 1/4 speed"?
The bus is split in half. One goes from the cpu out to memory, and the other half goes from memory back in to the cpu. Intel's chips have to negotiate every time memory access is being done.. Does the cpu have priority or does the memory? The advantage to that, though, it that there is a higher peak memory speed in either direction possible.
(Oh, and for the record, yeah, having tubes in your cell would kinda ruin the whole operating-on-battery experience. )
I don't even like them in my hi end audio equipment.
Quote:
Of course it's not -- and yet nobody seems to be running their FSBs at several GHz. Doesn't that seem kindo odd to you, given that you seem to think it's an almost trivial thing to implement?
I didn't say it's trivial. I said that it could be done. Lack of desire doesn't always mean lack of capability.
Apple just went to DDR2 533. Why? It offers little advantage. Even cheap PC mobo's are using 667 speeds. There are even 800 speed boards coming out.
Are you going to say that Apple didn't want to use that because it was so expensive and difficult to impliment? Because it isn't. They just chose not to.
Quote:
What exactly makes you think the CPU to bus ratio would remain constant? Just take a look at how the clock frequencies of CPU cores and front side buses have been scaling for the past decade or so. Following that argument, we'd have been running our FSBs at a 1:1 ratio all along ever since the 8086... (Note that I'm talking about actual clock frequencies, not "it's DDR so we'll just go ahead and multiply the frequency by two in all our marketing brochures".)
What makes you think that they wouldn't?
Quote:
Also, note that I'm not saying a 3 GHz FSB is somehow physically impossible to implement (that would be a rather stupid statement to make); it's just not feasible with current technology, and there would be little point for Apple to integrate such a capability in their northbridge when it's not likely to ever actually get used during the life cycle of said north bridge.
My statements here are just that it would have been put to a good use for dual core chips. Each core doesn't need the full speed, but dual cores running over the same line could use a higher bandwidth.
Quote:
To my knowledge, the major reason for PCI-X's coming into being was that, unlike PCIe, the former actually allowed for backwards compatibility with many existing PCI cards; I don't think anybody ever seriously expected it to become mainstream (and a quick glance at just what types of cards and motherboards are available as PCI-X versions kinda supports that notion...).
That was just another attempt to break Intel's stronghold over mobo standards. Give the customer another reason not to go Intel. Only a few boards have ever been made with the dual power ability, and with PCI X dying out quickly now, it's unlikely that too many new boards will continue to be available. It was being pushed for mainstream use. But with Intel making around 50% of all mobo's out there, it didn't catch on. Only workstation and server manufacturers adopted it.
Quote:
Also, as far as I'm aware, a single PCIe lane is actually cheaper to implement than PCI-X, while still offering much better performance than traditional PCI. No way to reuse your old PCI cards, tho.
If X had become mainstream, then the costs would have taken a dive. But too many manufacturers were afraid to take the plunge.
And you are right about one thing here. PC users are ever the cheapest people around. If they could figure out a way to use their 33k ISA modems today, they would do it!
Lack of desire doesn't always mean lack of capability.
Apple just went to DDR2 533. Why? It offers little advantage. Even cheap PC mobo's are using 667 speeds. There are even 800 speed boards coming out.
Are you going to say that Apple didn't want to use that because it was so expensive and difficult to impliment? Because it isn't. They just chose not to.
I don't quite see how this really relates to our argument -- it appears Apple felt it didn't make business sense for them to go with the higher speed DDR2. That, however, doesn't in any way imply that all their decisions are made purely for political / business reasons.
Quote:
What makes you think that they wouldn't?
Do you even read what I write? Again, take a look at how the average core-to-bus frequency ratio has been developing in the past decade. At least until the great brick wall was hit, on-chip frequencies increased a lot faster than off-chip ones did.
Quote:
My statements here are just that it would have been put to a good use for dual core chips. Each core doesn't need the full speed, but dual cores running over the same line could use a higher bandwidth.
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
Quote:
That was just another attempt to break Intel's stronghold over mobo standards. Give the customer another reason not to go Intel. Only a few boards have ever been made with the dual power ability, and with PCI X dying out quickly now, it's unlikely that too many new boards will continue to be available.
What exactly are you referring to by "dual power ability"?
Quote:
It was being pushed for mainstream use. But with Intel making around 50% of all mobo's out there, it didn't catch on. Only workstation and server manufacturers adopted it.
If they we're really trying to push it for mainstream use, they were doing a rather lousy job. There's just not much point in PCI-X for Joe Average: For graphics, PCI-X didn't really provide any advantage over AGPx8. Most high-speed peripherals a consumer would be likely to encounter have a pretty good chance of already being integrated into the chipset, and for low-bandwidth stuff such as TV, Sound or WLAN cards, PCI-X doesn't provide any advantage at all. That leaves you with (10)GigE, FibreChannel and other workstation- and server-type hardware -- which is a) pretty far from mainstream at this point, and b) happens to be exactly what's available as PCI-X cards.
Quote:
If X had become mainstream, then the costs would have taken a dive. But too many manufacturers were afraid to take the plunge.
Sure, it would have gotten cheaper, but that equally applies to PCIe. And, as I said, PCIe x1 is cheaper to implement than PCI-X, and "good enough" for most uses.
Quote:
And you are right about one thing here. PC users are ever the cheapest people around. If they could figure out a way to use their 33k ISA modems today, they would do it!
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
That is a falsehood. Inter-core communication has to pass through the FSB control unit on the chip, but it does not have to actually travel out the FSB and back. This makes a huge difference to its performance.
That is a falsehood. Inter-core communication has to pass through the FSB control unit on the chip, but it does not have to actually travel out the FSB and back. This makes a huge difference to its performance.
The documentation I've seen explicitly states that this is not the case. (Which kinda makes sense, seeing as the cache coherency protocol relies on the north bridge to reflect coherency traffic as necessary...)
Exactly. Now that I am thinking again about this, it makes much more sense to start the transition with the entry level consumer machines (iBook, Mac mini), and not with machines like the Powerbook (pro) or the iMac (consumer but high end). It is a low-risk approach, and at the same time Apple shows it is determined and the transition is not a joke.
[B]I don't quite see how this really relates to our argument -- it appears Apple felt it didn't make business sense for them to go with the higher speed DDR2. That, however, doesn't in any way imply that all their decisions are made purely for political / business reasons.
DDR 2 speeds relate to the argument because they point up an area in which Apple could have easily made a necessary improvement, but didn't. There would have been practically zero cost differential. Neither of us knows why they didn't implement this one either.
Quote:
Do you even read what I write? Again, take a look at how the average core-to-bus frequency ratio has been developing in the past decade. At least until the great brick wall was hit, on-chip frequencies increased a lot faster than off-chip ones did.
Of course I read what you wrote. But you are wrong. If you look at APPLE's bus speeds on the G5 PM's over the years, (I'm not interested in what Intel has been doing because Apple has a completely different bus and controller design), you will see that they scale perfectly with the cpu speeds. You make the assumption, with no reason to, that at some point in time there will be some disconnect between bus and cpu speeds. There has been no evidence for that. If anything the evidence is quite the opposite.
Quote:
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
And that's why I'm suggesting a solution.
Quote:
What exactly are you referring to by "dual power ability"?
PCI is a 5 volt logic powered board, and PCI X is either a 3.3 or 3.5 (I forget which right now) volt logic powered board. 5 volt boards won't work on PCI X, and 3+ volt boards won't work on PCI. Some boards can use either .
Quote:
If they we're really trying to push it for mainstream use, they were doing a rather lousy job. There's just not much point in PCI-X for Joe Average: For graphics, PCI-X didn't really provide any advantage over AGPx8. Most high-speed peripherals a consumer would be likely to encounter have a pretty good chance of already being integrated into the chipset, and for low-bandwidth stuff such as TV, Sound or WLAN cards, PCI-X doesn't provide any advantage at all. That leaves you with (10)GigE, FibreChannel and other workstation- and server-type hardware -- which is a) pretty far from mainstream at this point, and b) happens to be exactly what's available as PCI-X cards.
I was very disappointed when Apple went for PCI X, but they had no choice. PCI isn't up to running the boards that would take full advantage of the G5 and its fast memory bus. Since Express wasn't ready when Apple came out with the G5, they went with that.
PCI X isn't just about AGP. Express serves no purpose for the majority of users either. The 16x slot is needed only by those buying the highest performing boards, and the rest of the slots are at least twice as fast as the slots in PCI X, which in turn are twice as fast as those in PCI.
Like it or not, technology moves on, but yes, they did a lousy job of pushing it.
Quote:
Sure, it would have gotten cheaper, but that equally applies to PCIe. And, as I said, PCIe x1 is cheaper to implement than PCI-X, and "good enough" for most uses.
It's odd that you would say that, because the opinion stated many times over the years leading up to the introduction, was that Express would be more expensive than PCI X. That was one of the reasons PCI X was developed, and one of the reasons given in public statements.
Everything electronic gets cheaper over the years, and sometimes by the time something is introduced, the price has dropped past expectations. Even Intel admitted that Express would be more expensive than PCI X. And I'll state again that the reason PCI X wasn't adopted wasn't because of its expense, but because Intel makes at least 50% of the mobo's used by manufacturers, and they were afraid of getting stuck with something that Intel's boards would swamp. It was politics, and the fact that the PCI X manufacturers had no clear upgrade path from version one.
Exactly. Now that I am thinking again about this, it makes much more sense to start the transition with the entry level consumer machines (iBook, Mac mini), and not with machines like the Powerbook (pro) or the iMac (consumer but high end). It is a low-risk approach, and at the same time Apple shows it is determined and the transition is not a joke.
Now the interesting thing is that these machines will have to compete price wise (ballpark; Macs should always be a little more expensive) and they may well 'rock' in terms of performance AND looks. If Apple get this right they could have a hit on their hands. It doesn't matter if they out-perform PowerBooks in the short term (as they probably will). Its a transition phase and hit product would be more important for Apple.
Just out of curiosity, what is the general impression about the relative speed of the two major OS's that will be competing on the same chips?
e.g. Say Apple and some hack-job PC maker *cough*Dell*cough* both make laptops using the same Intel CPU.
Will X, assuming it is fully optimized for the chip, run at a par at the OS level (e.g. copying, finder-tasks, etc.)? Third-party software will, it seems to me, be a grab-bag; but the OS's are what I am thinking about here....
I am just an ideologue; I have no idea what I am talking about--- but it seems to me that Windows is awfully bloated to run at a par with X on the same hardware, but I could be totally wrong.
Geniuses... chime in?
Hope Springs Eternal,
Mandricard
AppleOutsider
Post Scriptum: ( OR, conversely, will the chips that Intel gives Apple not be available to Dell?)
Just out of curiosity, what is the general impression about the relative speed of the two major OS's that will be competing on the same chips?
e.g. Say Apple and some hack-job PC maker *cough*Dell*cough* both make laptops using the same Intel CPU.
Will X, assuming it is fully optimized for the chip, run at a par at the OS level (e.g. copying, finder-tasks, etc.)? Third-party software will, it seems to me, be a grab-bag; but the OS's are what I am thinking about here....
I am just an ideologue; I have no idea what I am talking about--- but it seems to me that Windows is awfully bloated to run at a par with X on the same hardware, but I could be totally wrong.
Geniuses... chime in?
Hope Springs Eternal,
Mandricard
AppleOutsider
Post Scriptum: ( OR, conversely, will the chips that Intel gives Apple not be available to Dell?)
Well, I would NEVER perform anything of questionable legality, but if I happened to have played around with OS X on an Intel chipset already, I guess I would say that the two are surprisingly similar. Now, if I happened to have tried the original 10.4.1 dev version of OSx86 on a Celeron M 1.3GHz laptop with 512 megs of RAM (fairly similar to the base iBook), I'd say that neither XP nor OSx86 had a clear advantage. But note that this is hacked beta software on unoptimized hardware, nowhere near what it will be like on the actual Intel Macs. I've got my money on it running slightly faster on equivalent CPUs, but not very much.
Comments
Originally posted by RazzFazz
This is just the writer's being sloppy (which does seem to be rather common these days). As is clearly stated in the PPC 970MP user manual, the processor bus is double-pumped.
OK I believe you on word.
It's a shame that Apple's writers are so sloppy.
Originally posted by RazzFazz
Well, no offense, but after the whole "my cordless phone works at several GHz" argument, I tend to think that that statement actually applies to you...
Similar maybe, but the complexity differs by orders of magnitude.
Of course the basic issues are the same; we're talking about physics, after all. It's just that the complexity of coping with them varies enormously depending on what exactly you try to do.
As an example, your typical run-of-the-mill serial bus uses LVDS with embedded clocks. As such, you can by definition never have bits coming in early or late, and don't have to worry about piecing them back together in the proper way -- which just so happens to be a huge deal for high-speed parallel buses.
Yeah. But just because it's feasible to hand-tune every trace on a simple MMIC doesn't mean the same applies to the complex, multi-layer PCBs found in computers. It's a totally different order of complexity.
In any case, whether you believe me or not, the point that nobody in the industry uses GHz-speed off-chip parallel buses should kinda tip you off to the fact that it's a lot less trivial than you seem to think.
Well, if I'm not mistaken, it was you who brought up the "my phone runs at GHz speeds" point...
First of all, I never said it would easy, like the old "rolling off the log" saying. But it isn't THAT impossible either. If Apple wanted to implement it they could. Would the cost go up slightly; sure.
Very few engineers would equate a cell's radio frequency circuitry with that of a magnetron in a microwave.
8 layer boards were developed for two purposes. One was to allow traces to cross each other in complex circuits. Two was to allow for shielding planes for high frequency uses. This is not new technology.
What you are also forgetting is that the G5 was designed originally to scale past 3GHz. The problems that beset the chip industry, which have little in common in our discussion here, has put an end to that, for the time being.
IBM had stated a couple of years ago that the successors to the 970 would get to 6GHz and beyond. With no reason to believe that the memory bus would change its relationship to the chip clock speed, we could be seeing the very 3GHz memory bus you are denying the possibility of.
It was just four years ago that the Express but, which runs at vastly higher speeds than the PCI bus, was thought of as being much too expensive to implement. That was one of the reasons why PCI X was developed (the other reason was opposition to another Intel standard).
As Gateway released a series of machines recently using the Express bus starting at $649, we can see how that prediction held up.
Your lack of confidence in modern manufacturing and engineering knowhow reminds me of the definitiveness of some in the industry over PCI X and Express.
Unfortunately, with Apple leaving IBM, we will never find out which one of us would be right.
Originally posted by Gene Clean
You mentioned different versions, which implied that there are non-Linux versions. There aren't.
Hardly, seeing as Crossover is just an implementation of WINE with a GUI.
Of course they are, as the benefit of getting thousands of small Windows apps to run would be minimal. People want IE to test their pages on, Word for that file OpenOffice can't open (rarely) and they want to run Photoshop and perhaps iTunes (in Linux). Nobody cares for the entire library of Windows-compatible apps because there are very good replacements for them.
And CrossOver = WINE with a GUI slapped on it. I don't see how that is "more useful" in running those 75% of other applications, unless you mean that people can use a mouse now instead of typing words for programs to start. Yep, useful as hell.
I'm amazed sometimes how people can parse words so finely that the concept of what is being said is lost.
There is a version for the PPC Mac. If you went to the web site, you would have seen that for yourself.
Is it ready for distribution? Not at all! I didn't say it was. But I've played around with it myself, and could see that it had promise.
They have been working on the x86 version for several months. This is even further behind. When it will it be usable, I have no idea. But it most likely will be.
I'm not advocating its use. Nor am I stating that it will be the best way to go. If you bothered to read my posts on this, you would have seen that I'm somewhat skeptical about its general purpose replacement for Windows, for those who want an easy, generally useful Windows experience.
But the version for Linux works well enough for those willing to go through the hassle of setting it up and installing the programs that do work.
Crossover makes this process very easy. Like it or not, for what it's intended, it works well. Read the article I posted, and you can see that for yourself. He is, by no means, the only writer to publish on this issue.
There are those who buy A Mac these days to use the Unix environment rather than the GUI that Apple prefers we use. does that make those users wrong because many Mac programs wont work? No, it doesn't.
Originally posted by melgross
The mechanical clock (for want of a better way of saying it) is 1/4 speed. The bus is then double pumped to 1/2 speed. Then each directions' traffic is 1/4 speed. Add them together and you again have 1/2 speed.
Could you explain that "each directions' traffic is 1/4 speed"?
Originally posted by melgross
Intel will never make a 64 bit Yonah. that is strictly a 32 bit design. It will be replaced mid to late 2006 with the Merom, which WILL be a 64 bit design.
Intel stated at the last idf that Yonah will coexist to Merom till Meroms transition to 45nm process in 2007.
Originally posted by melgross
First of all, I never said it would easy, like the old "rolling off the log" saying. But it isn't THAT impossible either. If Apple wanted to implement it they could. Would the cost go up slightly; sure.
Very few engineers would equate a cell's radio frequency circuitry with that of a magnetron in a microwave.
Which was exactly my point -- equally few engineers would equate a cell's (or cordless phone's) RF circuitry to a high-speed processor interconnect.
(Oh, and for the record, yeah, having tubes in your cell would kinda ruin the whole operating-on-battery experience.
8 layer boards were developed for two purposes. One was to allow traces to cross each other in complex circuits. Two was to allow for shielding planes for high frequency uses. This is not new technology.
Of course it's not -- and yet nobody seems to be running their FSBs at several GHz. Doesn't that seem kindo odd to you, given that you seem to think it's an almost trivial thing to implement?
What you are also forgetting is that the G5 was designed originally to scale past 3GHz. The problems that beset the chip industry, which have little in common in our discussion here, has put an end to that, for the time being.
I wholeheartedly agree that this has little to do with the discussion at hand.
IBM had stated a couple of years ago that the successors to the 970 would get to 6GHz and beyond. With no reason to believe that the memory bus would change its relationship to the chip clock speed, we could be seeing the very 3GHz memory bus you are denying the possibility of.
What exactly makes you think the CPU to bus ratio would remain constant? Just take a look at how the clock frequencies of CPU cores and front side buses have been scaling for the past decade or so. Following that argument, we'd have been running our FSBs at a 1:1 ratio all along ever since the 8086... (Note that I'm talking about actual clock frequencies, not "it's DDR so we'll just go ahead and multiply the frequency by two in all our marketing brochures".)
Also, note that I'm not saying a 3 GHz FSB is somehow physically impossible to implement (that would be a rather stupid statement to make); it's just not feasible with current technology, and there would be little point for Apple to integrate such a capability in their northbridge when it's not likely to ever actually get used during the life cycle of said north bridge.
It was just four years ago that the Express but, which runs at vastly higher speeds than the PCI bus, was thought of as being much too expensive to implement. That was one of the reasons why PCI X was developed (the other reason was opposition to another Intel standard).
To my knowledge, the major reason for PCI-X's coming into being was that, unlike PCIe, the former actually allowed for backwards compatibility with many existing PCI cards; I don't think anybody ever seriously expected it to become mainstream (and a quick glance at just what types of cards and motherboards are available as PCI-X versions kinda supports that notion...).
Also, as far as I'm aware, a single PCIe lane is actually cheaper to implement than PCI-X, while still offering much better performance than traditional PCI. No way to reuse your old PCI cards, tho.
As Gateway released a series of machines recently using the Express bus starting at $649, we can see how that prediction held up.
Your lack of confidence in modern manufacturing and engineering knowhow reminds me of the definitiveness of some in the industry over PCI X and Express.
Well, see above for my take on that.
Unfortunately, with Apple leaving IBM, we will never find out which one of us would be right.
True, I guess. Unless you somehow happen to run into that document again which you quoted earlier.
Originally posted by smalM
Could you explain that "each directions' traffic is 1/4 speed"?
The bus is split in half. One goes from the cpu out to memory, and the other half goes from memory back in to the cpu. Intel's chips have to negotiate every time memory access is being done.. Does the cpu have priority or does the memory? The advantage to that, though, it that there is a higher peak memory speed in either direction possible.
Originally posted by RazzFazz
[B
(Oh, and for the record, yeah, having tubes in your cell would kinda ruin the whole operating-on-battery experience.
I don't even like them in my hi end audio equipment.
Of course it's not -- and yet nobody seems to be running their FSBs at several GHz. Doesn't that seem kindo odd to you, given that you seem to think it's an almost trivial thing to implement?
I didn't say it's trivial. I said that it could be done. Lack of desire doesn't always mean lack of capability.
Apple just went to DDR2 533. Why? It offers little advantage. Even cheap PC mobo's are using 667 speeds. There are even 800 speed boards coming out.
Are you going to say that Apple didn't want to use that because it was so expensive and difficult to impliment? Because it isn't. They just chose not to.
What exactly makes you think the CPU to bus ratio would remain constant? Just take a look at how the clock frequencies of CPU cores and front side buses have been scaling for the past decade or so. Following that argument, we'd have been running our FSBs at a 1:1 ratio all along ever since the 8086... (Note that I'm talking about actual clock frequencies, not "it's DDR so we'll just go ahead and multiply the frequency by two in all our marketing brochures".)
What makes you think that they wouldn't?
Also, note that I'm not saying a 3 GHz FSB is somehow physically impossible to implement (that would be a rather stupid statement to make); it's just not feasible with current technology, and there would be little point for Apple to integrate such a capability in their northbridge when it's not likely to ever actually get used during the life cycle of said north bridge.
My statements here are just that it would have been put to a good use for dual core chips. Each core doesn't need the full speed, but dual cores running over the same line could use a higher bandwidth.
To my knowledge, the major reason for PCI-X's coming into being was that, unlike PCIe, the former actually allowed for backwards compatibility with many existing PCI cards; I don't think anybody ever seriously expected it to become mainstream (and a quick glance at just what types of cards and motherboards are available as PCI-X versions kinda supports that notion...).
That was just another attempt to break Intel's stronghold over mobo standards. Give the customer another reason not to go Intel. Only a few boards have ever been made with the dual power ability, and with PCI X dying out quickly now, it's unlikely that too many new boards will continue to be available. It was being pushed for mainstream use. But with Intel making around 50% of all mobo's out there, it didn't catch on. Only workstation and server manufacturers adopted it.
Also, as far as I'm aware, a single PCIe lane is actually cheaper to implement than PCI-X, while still offering much better performance than traditional PCI. No way to reuse your old PCI cards, tho.
If X had become mainstream, then the costs would have taken a dive. But too many manufacturers were afraid to take the plunge.
And you are right about one thing here. PC users are ever the cheapest people around. If they could figure out a way to use their 33k ISA modems today, they would do it!
EXCLUSIVE: Apple Planning Intel-Ready iBook Debut for January
Originally posted by melgross
Lack of desire doesn't always mean lack of capability.
Apple just went to DDR2 533. Why? It offers little advantage. Even cheap PC mobo's are using 667 speeds. There are even 800 speed boards coming out.
Are you going to say that Apple didn't want to use that because it was so expensive and difficult to impliment? Because it isn't. They just chose not to.
I don't quite see how this really relates to our argument -- it appears Apple felt it didn't make business sense for them to go with the higher speed DDR2. That, however, doesn't in any way imply that all their decisions are made purely for political / business reasons.
What makes you think that they wouldn't?
Do you even read what I write? Again, take a look at how the average core-to-bus frequency ratio has been developing in the past decade. At least until the great brick wall was hit, on-chip frequencies increased a lot faster than off-chip ones did.
My statements here are just that it would have been put to a good use for dual core chips. Each core doesn't need the full speed, but dual cores running over the same line could use a higher bandwidth.
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
That was just another attempt to break Intel's stronghold over mobo standards. Give the customer another reason not to go Intel. Only a few boards have ever been made with the dual power ability, and with PCI X dying out quickly now, it's unlikely that too many new boards will continue to be available.
What exactly are you referring to by "dual power ability"?
It was being pushed for mainstream use. But with Intel making around 50% of all mobo's out there, it didn't catch on. Only workstation and server manufacturers adopted it.
If they we're really trying to push it for mainstream use, they were doing a rather lousy job. There's just not much point in PCI-X for Joe Average: For graphics, PCI-X didn't really provide any advantage over AGPx8. Most high-speed peripherals a consumer would be likely to encounter have a pretty good chance of already being integrated into the chipset, and for low-bandwidth stuff such as TV, Sound or WLAN cards, PCI-X doesn't provide any advantage at all. That leaves you with (10)GigE, FibreChannel and other workstation- and server-type hardware -- which is a) pretty far from mainstream at this point, and b) happens to be exactly what's available as PCI-X cards.
If X had become mainstream, then the costs would have taken a dive. But too many manufacturers were afraid to take the plunge.
Sure, it would have gotten cheaper, but that equally applies to PCIe. And, as I said, PCIe x1 is cheaper to implement than PCI-X, and "good enough" for most uses.
And you are right about one thing here. PC users are ever the cheapest people around. If they could figure out a way to use their 33k ISA modems today, they would do it!
Glad we agree on something after all.
Originally posted by RazzFazz
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
That is a falsehood. Inter-core communication has to pass through the FSB control unit on the chip, but it does not have to actually travel out the FSB and back. This makes a huge difference to its performance.
Originally posted by Programmer
That is a falsehood. Inter-core communication has to pass through the FSB control unit on the chip, but it does not have to actually travel out the FSB and back. This makes a huge difference to its performance.
The documentation I've seen explicitly states that this is not the case. (Which kinda makes sense, seeing as the cache coherency protocol relies on the north bridge to reflect coherency traffic as necessary...)
Originally posted by PB
Excuse me to interrupt, but there are some news (perhaps a new thread is needed):
EXCLUSIVE: Apple Planning Intel-Ready iBook Debut for January
No kidding. Someone start a thread. If it has a dual core, I'm there.
Originally posted by PB
Excuse me to interrupt, but there are some news (perhaps a new thread is needed):
EXCLUSIVE: Apple Planning Intel-Ready iBook Debut for January
That's more like it. That makes sense.
Originally posted by vinney57
That's more like it. That makes sense.
Exactly. Now that I am thinking again about this, it makes much more sense to start the transition with the entry level consumer machines (iBook, Mac mini), and not with machines like the Powerbook (pro) or the iMac (consumer but high end). It is a low-risk approach, and at the same time Apple shows it is determined and the transition is not a joke.
Originally posted by RazzFazz
[B]I don't quite see how this really relates to our argument -- it appears Apple felt it didn't make business sense for them to go with the higher speed DDR2. That, however, doesn't in any way imply that all their decisions are made purely for political / business reasons.
DDR 2 speeds relate to the argument because they point up an area in which Apple could have easily made a necessary improvement, but didn't. There would have been practically zero cost differential. Neither of us knows why they didn't implement this one either.
Do you even read what I write? Again, take a look at how the average core-to-bus frequency ratio has been developing in the past decade. At least until the great brick wall was hit, on-chip frequencies increased a lot faster than off-chip ones did.
Of course I read what you wrote. But you are wrong. If you look at APPLE's bus speeds on the G5 PM's over the years, (I'm not interested in what Intel has been doing because Apple has a completely different bus and controller design), you will see that they scale perfectly with the cpu speeds. You make the assumption, with no reason to, that at some point in time there will be some disconnect between bus and cpu speeds. There has been no evidence for that. If anything the evidence is quite the opposite.
Absolutely. I fully agree the FSB is a weak point of the 970MP's design -- especially considering that all communication, even between cores on the same chip, has to go through the north bridge.
And that's why I'm suggesting a solution.
What exactly are you referring to by "dual power ability"?
PCI is a 5 volt logic powered board, and PCI X is either a 3.3 or 3.5 (I forget which right now) volt logic powered board. 5 volt boards won't work on PCI X, and 3+ volt boards won't work on PCI. Some boards can use either .
If they we're really trying to push it for mainstream use, they were doing a rather lousy job. There's just not much point in PCI-X for Joe Average: For graphics, PCI-X didn't really provide any advantage over AGPx8. Most high-speed peripherals a consumer would be likely to encounter have a pretty good chance of already being integrated into the chipset, and for low-bandwidth stuff such as TV, Sound or WLAN cards, PCI-X doesn't provide any advantage at all. That leaves you with (10)GigE, FibreChannel and other workstation- and server-type hardware -- which is a) pretty far from mainstream at this point, and b) happens to be exactly what's available as PCI-X cards.
I was very disappointed when Apple went for PCI X, but they had no choice. PCI isn't up to running the boards that would take full advantage of the G5 and its fast memory bus. Since Express wasn't ready when Apple came out with the G5, they went with that.
PCI X isn't just about AGP. Express serves no purpose for the majority of users either. The 16x slot is needed only by those buying the highest performing boards, and the rest of the slots are at least twice as fast as the slots in PCI X, which in turn are twice as fast as those in PCI.
Like it or not, technology moves on, but yes, they did a lousy job of pushing it.
Sure, it would have gotten cheaper, but that equally applies to PCIe. And, as I said, PCIe x1 is cheaper to implement than PCI-X, and "good enough" for most uses.
It's odd that you would say that, because the opinion stated many times over the years leading up to the introduction, was that Express would be more expensive than PCI X. That was one of the reasons PCI X was developed, and one of the reasons given in public statements.
Everything electronic gets cheaper over the years, and sometimes by the time something is introduced, the price has dropped past expectations. Even Intel admitted that Express would be more expensive than PCI X. And I'll state again that the reason PCI X wasn't adopted wasn't because of its expense, but because Intel makes at least 50% of the mobo's used by manufacturers, and they were afraid of getting stuck with something that Intel's boards would swamp. It was politics, and the fact that the PCI X manufacturers had no clear upgrade path from version one.
Originally posted by PB
Exactly. Now that I am thinking again about this, it makes much more sense to start the transition with the entry level consumer machines (iBook, Mac mini), and not with machines like the Powerbook (pro) or the iMac (consumer but high end). It is a low-risk approach, and at the same time Apple shows it is determined and the transition is not a joke.
Now the interesting thing is that these machines will have to compete price wise (ballpark; Macs should always be a little more expensive) and they may well 'rock' in terms of performance AND looks. If Apple get this right they could have a hit on their hands. It doesn't matter if they out-perform PowerBooks in the short term (as they probably will). Its a transition phase and hit product would be more important for Apple.
e.g. Say Apple and some hack-job PC maker *cough*Dell*cough* both make laptops using the same Intel CPU.
Will X, assuming it is fully optimized for the chip, run at a par at the OS level (e.g. copying, finder-tasks, etc.)? Third-party software will, it seems to me, be a grab-bag; but the OS's are what I am thinking about here....
I am just an ideologue; I have no idea what I am talking about--- but it seems to me that Windows is awfully bloated to run at a par with X on the same hardware, but I could be totally wrong.
Geniuses... chime in?
Hope Springs Eternal,
Mandricard
AppleOutsider
Post Scriptum: ( OR, conversely, will the chips that Intel gives Apple not be available to Dell?)
Originally posted by Mandricard
Just out of curiosity, what is the general impression about the relative speed of the two major OS's that will be competing on the same chips?
e.g. Say Apple and some hack-job PC maker *cough*Dell*cough* both make laptops using the same Intel CPU.
Will X, assuming it is fully optimized for the chip, run at a par at the OS level (e.g. copying, finder-tasks, etc.)? Third-party software will, it seems to me, be a grab-bag; but the OS's are what I am thinking about here....
I am just an ideologue; I have no idea what I am talking about--- but it seems to me that Windows is awfully bloated to run at a par with X on the same hardware, but I could be totally wrong.
Geniuses... chime in?
Hope Springs Eternal,
Mandricard
AppleOutsider
Post Scriptum: ( OR, conversely, will the chips that Intel gives Apple not be available to Dell?)
Well, I would NEVER perform anything of questionable legality, but if I happened to have played around with OS X on an Intel chipset already, I guess I would say that the two are surprisingly similar. Now, if I happened to have tried the original 10.4.1 dev version of OSx86 on a Celeron M 1.3GHz laptop with 512 megs of RAM (fairly similar to the base iBook), I'd say that neither XP nor OSx86 had a clear advantage. But note that this is hacked beta software on unoptimized hardware, nowhere near what it will be like on the actual Intel Macs. I've got my money on it running slightly faster on equivalent CPUs, but not very much.
All speculative, of course... *cough cough*