I kinda sorta see David's point, but faulting Hannibal for not taking a more holistic view in an article aimed fairly narrowly at the changes wrought by x86-64 seems to me to be tilting at windmills. It's a narrowly aimed technical article written by an engineer and published on a site that publishes narrowly aimed technical articles. If there's a bit of a slant toward PC technology, well, Ars Technica is well known for that. Everyone has their own spin on things, not least David K. Every.
It seemed clear to me that Hannibal took a cautionary tone in the one paragraph where he addresses Mac users because he's just got done describing a bunch of architectural changes that are unique to the x86-64, not the result of a generic 32-bit to 64-bit transition. He doesn't say that the 970 won't be faster, and he does use the phrase "no cleaned up addressing scheme," which acknowledges (however implicitly) that there's no addressing scheme to clean up in the first place. It's clearly not spun to David's liking, but I don't see that that's worth a whole article - unless David would like to take Hannibal's paragraph as a starting point and write a supplementary article explaining where the 970's speed improvements will come from, if they're not coming from the transition to 64 bit.
Hannibal has been preparing a followup article on the 970 for some time, and it seems that it has only failed to appear because he hasn't been able to research it as well as he'd like to. Given that, I see no reason to presume bad faith on his part, or a pro-x86 slant steeper than can be explained by simple familiarity. The proof will come, of course, with the arrival of his examination of the 970.
If a hypothetical next generation PPC system added only 64bit execution capability but held everything else constant (FSB speed, pipeline depth, functional units, clock rate, cache sizes)... then to claim that there would be no speed boost on 32-bit code would be very reasonable.
But: we already know that the 32/64 bit switch is not the only trick under the 970's hood.
An alternative statement: "64-bit mode will not directly contribute to the performance gains expected on 970 systems in comparison with G4 systems."
Macwhispers says that there is bidding going on for two new Apple motherboards which are completely new and that one is a dual config and the other is a single config. Could it be the fact that we can see single 970s and dual G4's because of cost? Almost like a two processor line for the PMs or could it be the fact the we could see a 1.8 dual 970 on the fastest and single 1.4 ans 1.6 970s on the fast and faster?
<strong>Macwhispers says that there is bidding going on for two new Apple motherboards which are completely new and that one is a dual config and the other is a single config. Could it be the fact that we can see single 970s and dual G4's because of cost? Almost like a two processor line for the PMs or could it be the fact the we could see a 1.8 dual 970 on the fastest and single 1.4 ans 1.6 970s on the fast and faster? </strong><hr></blockquote>
I think it will be more the latter. The PM line will be all 970 with a low cost entry level single processor 1.4, -dual processor mid 1.4 or 1.6 and -upper level dual 1.8. Even the single processor 1.4 would be an awesome replacement for my single 800 G4.
I have started saving for whatever they bring out.
think it will be more the latter. The PM line will be all 970 with a low cost entry level single processor 1.4, -dual processor mid 1.4 or 1.6 and -upper level dual 1.8. Even the single processor 1.4 would be an awesome replacement for my single 800 G4.
I have started saving for whatever they bring out. <hr></blockquote>
I have been saving as well. I got a old B&W PM that needs to get upgraded and the 970 would be a real nice thing. I was just saying that the Apple may have two processors for the PM because of the cost of the 970. I assume that the 970 is going to be costly and it may be to expensive for some people, but I am sure that they will have plenty of orders for the 970s as I will be eagerly awaiting to get one of those puppys even if I have to sell my right arm
Regarding Apple pushing the limits of the 970 or not.
With the variable bus mulitplier of the G4 the same motherboard, buss and memory could be used for G4 runining 350 MHz as well as a 3.2 GHz G4 (3.5x and 32 x a 100 Mhz bus. So with that design a faster CPu appear to be easy to put in, just increase the bus multiplier and off you are!
With the fixed 2x bus ratio of the 970, some questions arise. With the bus speed rising along with the CPu speed of the 970, what does that mean for Apple in terms of technical problem and cost of memory chips etc???
Or in more practial tems , how much more expensive would a 900 Mhz or 1.25 GHz bus be than say a 700 MHz bus?
And how about dual implementation? Can a dual MB with slower bus become cheaper than a single really fast?
The general jist of what I have heard, is that the 970 won't be significantly more expensive if any. Motorola has always had trouble getting decent yields which has kept the cost of their chips up. Rumor is IBM is getting much better yields then expected, and that generally brings the cost down. I really suspect the cost of the new motherboards will be more significant then the cost of the new CPU.
maybe i just do not get it..but can somebody confirm ..the 970s architecture is 2times faster than a g4 and therefore almost 4 times faster than a pentium4?
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
<strong>maybe i just do not get it..but can somebody confirm ..the 970s architecture is 2times faster than a g4 and therefore almost 4 times faster than a pentium4?
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
err..am i nuts?</strong><hr></blockquote>
about your last question: i don't know
about your first question: i don't know eighter <img src="embarrassed.gif" border="0">
i thought it's more about twice as powerfull than a G4 instead of twice as fast.
ummm..whats the difference?</strong><hr></blockquote>
the difference between a weightlifter and a atlete:
the atlete is twice as fast as the weightlifter and the weightlifter is twice as strong as the atlete. but they are both human and able to talk, something like that but than completly different.
there is also a car analogy about small european cars with big diesel or gasoline engines.
the difference between a weightlifter and a atlete:
the atlete is twice as fast as the weightlifter and the weightlifter is twice as strong as the atlete. but they are both human and able to talk, something like that but than completly different.
there is also a car analogy about small european cars with big diesel or gasoline engines.</strong><hr></blockquote>
well...whatever...so powerful doesnt mean faster? it means it can drain more power...how cool!
With the variable bus mulitplier of the G4 the same motherboard, buss and memory could be used for G4 runining 350 MHz as well as a 3.2 GHz G4 (3.5x and 32 x a 100 Mhz bus. So with that design a faster CPu appear to be easy to put in, just increase the bus multiplier and off you are!
With the fixed 2x bus ratio of the 970, some questions arise. With the bus speed rising along with the CPu speed of the 970, what does that mean for Apple in terms of technical problem and cost of memory chips etc???
...</strong><hr></blockquote>
Remeber, the 970 has a companion chip which would be between the CPU and memory. Thus you don't need to worry about memory speed if you want to increase the CPU clock. (well, except for the obvious concern of feeding data) Apple would, however, have to have a companion chip able to run the CPU bus(es) at the higher speed. But since Apple is involved with its design and such, they should know very well what the availablity of those faster-bussed chips would be.
Did that make any sense?? <img src="confused.gif" border="0">
[quote]Originally posted by Transcendental Octothorpe:
<strong>
Remeber, the 970 has a companion chip which would be between the CPU and memory. Thus you don't need to worry about memory speed if you want to increase the CPU clock. (well, except for the obvious concern of feeding data) Apple would, however, have to have a companion chip able to run the CPU bus(es) at the higher speed. But since Apple is involved with its design and such, they should know very well what the availablity of those faster-bussed chips would be.
Did that make any sense?? </strong><hr></blockquote>
It did and Apple has already done it. It's called the "System Controller" which has been part of the motherboard architecture since the Xserve was introduced. For the PPC970 it will be updated, but did anyone think that this architecture was developed just to give us the current lame DDR implementation? <img src="graemlins/oyvey.gif" border="0" alt="[oyvey]" />
BTW - It also shows how long ago Apple was planning for the PPC970 - it's coming sooner than later.
well...whatever...so powerful doesnt mean faster? it means it can drain more power...how cool!</strong><hr></blockquote>
no einstein... it means that it can do more complicated tasks per clockcycle than a P4, so clock for clock it could indeed be faster like in photoshop or premiere. but in simple tasks like email or office etc. the p4 is definitly faster.
take yourself for instance: you can think very fast, your a smartass, but obviously not very deep: so your a pentium 4.
no einstein... it means that it can do more complicated tasks per clockcycle than a P4, so clock for clock it could indeed be faster like in photoshop or premiere. but in simple tasks like email or office etc. the p4 is definitly faster.
take yourself for instance: you can think very fast, your a smartass, but obviously not very deep: so your a pentium 4. </strong><hr></blockquote>
[quote]It did and Apple has already done it. It's called the "System Controller" which has been part of the motherboard architecture since the Xserve was introduced. For the PPC970 it will be updated, but did anyone think that this architecture was developed just to give us the current lame DDR implementation?<hr></blockquote>
Except that the 970 is such a different beast from the G4+ in terms of memory interface. It would be like trying to take the drive train from an Corolla and put it in a Ferrari. The principles may be similar, but if you don't redesign from scratch, you'll be wasting 3/4 of the potential...
That, and from what I understand, the 970 memory controller is a beast to design, although admittedly any rumours I've heard about Apple's difficulties with it are now quite old.
<strong>maybe i just do not get it..but can somebody confirm ..the 970s architecture is 2times faster than a g4 and therefore almost 4 times faster than a pentium4?
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
err..am i nuts?
</strong><hr></blockquote>
No, you're just oversimplifying things. Most of these measurements are non-linear, and performance metrics like "faster" are highly dependent on what software & data you are with. Doubling a machine's clock rate will not double its performance. Doubling the number of processors will typically not double its performance. Doubling the number of execution units will typically not double its performance. So be careful tossing around notions like "this is twice as fast as that, therefore..." -- they just don't hold water.
I expect in general, for most tasks and given roughly the same system & memory, a 1.8 GHz 970 will be in roughly the same ballpark as a 2.5-3 GHz Pentium4. For some things it'll be a lot faster, and I'm sure people will find things where it is slower. One of the great things about the 970 compared to the G4 is that there will be many few things where it is slower at all, and its much more likely that any given piece of code will run much better on the 970 than it does on the G4. With the G4 the software developers typically have to work harder to eek out the top-notch performance, compared to the Pentium4, Athlon and 970.
It did and Apple has already done it. It's called the "System Controller" which has been part of the motherboard architecture since the Xserve was introduced. For the PPC970 it will be updated, but did anyone think that this architecture was developed just to give us the current lame DDR implementation? <img src="graemlins/oyvey.gif" border="0" alt="[oyvey]" />
BTW - It also shows how long ago Apple was planning for the PPC970 - it's coming sooner than later. </strong><hr></blockquote>
The "extra" bandwidth of the DDR333 memory exceeds the processor's 167 MHz bandwidth, but that doesn't mean it goes unused. Its interesting to sit and play with the system because you can see that even when the processors are both have the shared, bandwidth limited MPX bus going at full tilt the GPU can still be streaming vertex & texture data from main memory, and the ATA bus can still be read/writing data to the disk(s). Since Quartz Extreme is constantly using the GPU to stream textures, this is very relevent to current MacOS X performance. The system is not a "hack", it is the fastest system Apple could build given the limitations of the processor they had to use.
It will be interesting to see what the architecture of Apple's new machines looks like -- they might start from their existing system controller, instead just canabilize some of the design elements from it, or throw it away and completely start from scratch using lessons learned. It could stay as one chip, or it could become a north/south bridge pair. There might be one companion per 970, or a ring of 970s with on companion. I have no doubt that it'll use at least DDR333, but I'm not holding my breath for anything more than that (at least in version 1).
Comments
<strong>Hmmm...but which of the 52 Tuesdays will it be?
Second Tuesday next week!
<strong>
I kinda sorta see David's point, but faulting Hannibal for not taking a more holistic view in an article aimed fairly narrowly at the changes wrought by x86-64 seems to me to be tilting at windmills. It's a narrowly aimed technical article written by an engineer and published on a site that publishes narrowly aimed technical articles. If there's a bit of a slant toward PC technology, well, Ars Technica is well known for that. Everyone has their own spin on things, not least David K. Every.
It seemed clear to me that Hannibal took a cautionary tone in the one paragraph where he addresses Mac users because he's just got done describing a bunch of architectural changes that are unique to the x86-64, not the result of a generic 32-bit to 64-bit transition. He doesn't say that the 970 won't be faster, and he does use the phrase "no cleaned up addressing scheme," which acknowledges (however implicitly) that there's no addressing scheme to clean up in the first place. It's clearly not spun to David's liking, but I don't see that that's worth a whole article - unless David would like to take Hannibal's paragraph as a starting point and write a supplementary article explaining where the 970's speed improvements will come from, if they're not coming from the transition to 64 bit.
Hannibal has been preparing a followup article on the 970 for some time, and it seems that it has only failed to appear because he hasn't been able to research it as well as he'd like to. Given that, I see no reason to presume bad faith on his part, or a pro-x86 slant steeper than can be explained by simple familiarity. The proof will come, of course, with the arrival of his examination of the 970.
[ 03-13-2003: Message edited by: Amorph ]</strong><hr></blockquote>
If a hypothetical next generation PPC system added only 64bit execution capability but held everything else constant (FSB speed, pipeline depth, functional units, clock rate, cache sizes)... then to claim that there would be no speed boost on 32-bit code would be very reasonable.
But: we already know that the 32/64 bit switch is not the only trick under the 970's hood.
An alternative statement: "64-bit mode will not directly contribute to the performance gains expected on 970 systems in comparison with G4 systems."
Just a thought
<strong>Macwhispers says that there is bidding going on for two new Apple motherboards which are completely new and that one is a dual config and the other is a single config. Could it be the fact that we can see single 970s and dual G4's because of cost? Almost like a two processor line for the PMs or could it be the fact the we could see a 1.8 dual 970 on the fastest and single 1.4 ans 1.6 970s on the fast and faster? </strong><hr></blockquote>
I think it will be more the latter. The PM line will be all 970 with a low cost entry level single processor 1.4, -dual processor mid 1.4 or 1.6 and -upper level dual 1.8. Even the single processor 1.4 would be an awesome replacement for my single 800 G4.
I have started saving for whatever they bring out.
think it will be more the latter. The PM line will be all 970 with a low cost entry level single processor 1.4, -dual processor mid 1.4 or 1.6 and -upper level dual 1.8. Even the single processor 1.4 would be an awesome replacement for my single 800 G4.
I have started saving for whatever they bring out.
I have been saving as well. I got a old B&W PM that needs to get upgraded and the 970 would be a real nice thing. I was just saying that the Apple may have two processors for the PM because of the cost of the 970. I assume that the 970 is going to be costly and it may be to expensive for some people, but I am sure that they will have plenty of orders for the 970s as I will be eagerly awaiting to get one of those puppys even if I have to sell my right arm
With the variable bus mulitplier of the G4 the same motherboard, buss and memory could be used for G4 runining 350 MHz as well as a 3.2 GHz G4 (3.5x and 32 x a 100 Mhz bus. So with that design a faster CPu appear to be easy to put in, just increase the bus multiplier and off you are!
With the fixed 2x bus ratio of the 970, some questions arise. With the bus speed rising along with the CPu speed of the 970, what does that mean for Apple in terms of technical problem and cost of memory chips etc???
Or in more practial tems , how much more expensive would a 900 Mhz or 1.25 GHz bus be than say a 700 MHz bus?
And how about dual implementation? Can a dual MB with slower bus become cheaper than a single really fast?
[ 03-14-2003: Message edited by: DrBoar ]</p>
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
err..am i nuts?
[ 03-14-2003: Message edited by: cycle ]</p>
<strong>maybe i just do not get it..but can somebody confirm ..the 970s architecture is 2times faster than a g4 and therefore almost 4 times faster than a pentium4?
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
err..am i nuts?</strong><hr></blockquote>
about your last question: i don't know
about your first question: i don't know eighter <img src="embarrassed.gif" border="0">
i thought it's more about twice as powerfull than a G4 instead of twice as fast.
<strong>
about your last question: i don't know
about your first question: i don't know eighter <img src="embarrassed.gif" border="0">
i thought it's more about twice as powerfull than a G4 instead of twice as fast.</strong><hr></blockquote>
ummm..whats the difference?
<strong>
ummm..whats the difference?</strong><hr></blockquote>
the difference between a weightlifter and a atlete:
the atlete is twice as fast as the weightlifter and the weightlifter is twice as strong as the atlete. but they are both human and able to talk, something like that but than completly different.
there is also a car analogy about small european cars with big diesel or gasoline engines.
<strong>
the difference between a weightlifter and a atlete:
the atlete is twice as fast as the weightlifter and the weightlifter is twice as strong as the atlete. but they are both human and able to talk, something like that but than completly different.
there is also a car analogy about small european cars with big diesel or gasoline engines.</strong><hr></blockquote>
well...whatever...so powerful doesnt mean faster? it means it can drain more power...how cool!
<strong>...
With the variable bus mulitplier of the G4 the same motherboard, buss and memory could be used for G4 runining 350 MHz as well as a 3.2 GHz G4 (3.5x and 32 x a 100 Mhz bus. So with that design a faster CPu appear to be easy to put in, just increase the bus multiplier and off you are!
With the fixed 2x bus ratio of the 970, some questions arise. With the bus speed rising along with the CPu speed of the 970, what does that mean for Apple in terms of technical problem and cost of memory chips etc???
...</strong><hr></blockquote>
Remeber, the 970 has a companion chip which would be between the CPU and memory. Thus you don't need to worry about memory speed if you want to increase the CPU clock. (well, except for the obvious concern of feeding data) Apple would, however, have to have a companion chip able to run the CPU bus(es) at the higher speed. But since Apple is involved with its design and such, they should know very well what the availablity of those faster-bussed chips would be.
Did that make any sense?? <img src="confused.gif" border="0">
<strong>
Remeber, the 970 has a companion chip which would be between the CPU and memory. Thus you don't need to worry about memory speed if you want to increase the CPU clock. (well, except for the obvious concern of feeding data) Apple would, however, have to have a companion chip able to run the CPU bus(es) at the higher speed. But since Apple is involved with its design and such, they should know very well what the availablity of those faster-bussed chips would be.
Did that make any sense??
It did and Apple has already done it. It's called the "System Controller" which has been part of the motherboard architecture since the Xserve was introduced. For the PPC970 it will be updated, but did anyone think that this architecture was developed just to give us the current lame DDR implementation? <img src="graemlins/oyvey.gif" border="0" alt="[oyvey]" />
BTW - It also shows how long ago Apple was planning for the PPC970 - it's coming sooner than later.
<strong>
well...whatever...so powerful doesnt mean faster? it means it can drain more power...how cool!</strong><hr></blockquote>
no einstein... it means that it can do more complicated tasks per clockcycle than a P4, so clock for clock it could indeed be faster like in photoshop or premiere. but in simple tasks like email or office etc. the p4 is definitly faster.
take yourself for instance: you can think very fast, your a smartass, but obviously not very deep: so your a pentium 4.
<strong>
no einstein... it means that it can do more complicated tasks per clockcycle than a P4, so clock for clock it could indeed be faster like in photoshop or premiere. but in simple tasks like email or office etc. the p4 is definitly faster.
take yourself for instance: you can think very fast, your a smartass, but obviously not very deep: so your a pentium 4.
man..that wasnt nice
<strong>
man..that wasnt nice
indeed it wasn't, i apologise
Except that the 970 is such a different beast from the G4+ in terms of memory interface. It would be like trying to take the drive train from an Corolla and put it in a Ferrari. The principles may be similar, but if you don't redesign from scratch, you'll be wasting 3/4 of the potential...
That, and from what I understand, the 970 memory controller is a beast to design, although admittedly any rumours I've heard about Apple's difficulties with it are now quite old.
<strong>maybe i just do not get it..but can somebody confirm ..the 970s architecture is 2times faster than a g4 and therefore almost 4 times faster than a pentium4?
i mean worst case...1.4ghz single (in imac too?!)
would be equal to a 2.8ghz g4 or a 5.6ghz pentium4?
and then..maybe a dual 1.8ghz equal to 14.4ghz pentium4?
which is about 4 times faster than the best pentium money can buy?
err..am i nuts?
</strong><hr></blockquote>
No, you're just oversimplifying things. Most of these measurements are non-linear, and performance metrics like "faster" are highly dependent on what software & data you are with. Doubling a machine's clock rate will not double its performance. Doubling the number of processors will typically not double its performance. Doubling the number of execution units will typically not double its performance. So be careful tossing around notions like "this is twice as fast as that, therefore..." -- they just don't hold water.
I expect in general, for most tasks and given roughly the same system & memory, a 1.8 GHz 970 will be in roughly the same ballpark as a 2.5-3 GHz Pentium4. For some things it'll be a lot faster, and I'm sure people will find things where it is slower. One of the great things about the 970 compared to the G4 is that there will be many few things where it is slower at all, and its much more likely that any given piece of code will run much better on the 970 than it does on the G4. With the G4 the software developers typically have to work harder to eek out the top-notch performance, compared to the Pentium4, Athlon and 970.
<strong>
It did and Apple has already done it. It's called the "System Controller" which has been part of the motherboard architecture since the Xserve was introduced. For the PPC970 it will be updated, but did anyone think that this architecture was developed just to give us the current lame DDR implementation? <img src="graemlins/oyvey.gif" border="0" alt="[oyvey]" />
BTW - It also shows how long ago Apple was planning for the PPC970 - it's coming sooner than later.
The "extra" bandwidth of the DDR333 memory exceeds the processor's 167 MHz bandwidth, but that doesn't mean it goes unused. Its interesting to sit and play with the system because you can see that even when the processors are both have the shared, bandwidth limited MPX bus going at full tilt the GPU can still be streaming vertex & texture data from main memory, and the ATA bus can still be read/writing data to the disk(s). Since Quartz Extreme is constantly using the GPU to stream textures, this is very relevent to current MacOS X performance. The system is not a "hack", it is the fastest system Apple could build given the limitations of the processor they had to use.
It will be interesting to see what the architecture of Apple's new machines looks like -- they might start from their existing system controller, instead just canabilize some of the design elements from it, or throw it away and completely start from scratch using lessons learned. It could stay as one chip, or it could become a north/south bridge pair. There might be one companion per 970, or a ring of 970s with on companion. I have no doubt that it'll use at least DDR333, but I'm not holding my breath for anything more than that (at least in version 1).