If the 970 costs less than (latest rumour 25% + less than current G4's), then it would be idiotic for Apple to put the more expensive chip in their low cost consumer computers wouldnt it? It would be better for Apple to differentiate between Pro and Consumer with single/dual configs than to use A more expensive chip in the iMac and iBook which are their lowest margin computers.
I can't say anything for sure, but I expect the 970 @ 130nm is cheaper than the 7455 @ 180nm + 1/2MB L3 cache, cache costs a lot. Apple could use a 7457/47 without L3 cache in their low cost consumer products, as they're doing now with the iMac, eMac and the current G4.
Die size for the 7457 is 98 square mm, and 121 for the 970, so it seems it'll be cheaper when produced on an equal process. Depends on MOTs ability to keep up the yields... someone else will probably be able to explain this better, or explain why the 970 will be cheaper anyways, if that's the case
What wafer size is Mot using? IBM is using 300mm wafers at East Fishkill which will also lower production costs as they get more CPU's per wafer despite a larger die size.
All of Mot's fabs are 200mm.
[edit: Except the Crolles2 joint venture, which is 300mm]
If the 970 costs less than (latest rumour 25% + less than current G4's), then it would be idiotic for Apple to put the more expensive chip in their low cost consumer computers wouldnt it? It would be better for Apple to differentiate between Pro and Consumer with single/dual configs than to use A more expensive chip in the iMac and iBook which are their lowest margin computers.
True, the 970 may cost less than a G4, but that doesn't mean that Apple would drop G4's. First of all, a 970 needs a new motherboard, and Apple needs to design new motherboards for all the machines that are currently using G4's. Portable and iMac motherboards are much harder than mini ATX motherboards due to their unique space considerations. Personally, I hope that Apple has been on the ball and that when the 970 is rolled out, that we'll see it across the tower lineup and that we'll se it in 17 and 15 inch laptops as well. This of course is wildly over optimistic. Anyhow, my main point is that there are many more considerations than the chip cost that will keep Apple using G4's for a while. It is a significant effort to revamp your entire hardware offerings to use a new chip. (of course, the TiBook is long overdue for a revamp!).
And also the word is that the 970 has rudimentary, if any, dedicated power saving logic on the processor, which makes it difficult if not precludes it for being used on a laptop just yet. This is not so much a consideration for the iMac as it is for a truly portable computer like the PowerBook or iBook.
However, Apple may integrate the power saving features on a companion chip somehow. I'm not sure how this would work, but it could be possible that the companion chip would be able to slow down the processor MHz wise while on battery power, turn certain parts of the processor off when not in use (maybe when a non-Altivec program is not in use it can be turned off for example), and other tricks to conserve battery power.
And also the word is that the 970 has rudimentary, if any, dedicated power saving logic on the processor, which makes it difficult if not precludes it for being used on a laptop just yet. This is not so much a consideration for the iMac as it is for a truly portable computer like the PowerBook or iBook.
The 970's power saving capabilities are hardly worse than the G4's. The "low power" G4 is just lower voltage and missing tags for L3 cache, and "low power" operation in the 'books consists of throttling the clock down.
If you make a chip efficient in the first place you don't need much in the way of power saving technology.
The 970's power saving capabilities are hardly worse than the G4's. The "low power" G4 is just lower voltage and missing tags for L3 cache, and "low power" operation in the 'books consists of throttling the clock down.
If you make a chip efficient in the first place you don't need much in the way of power saving technology.
True, if the chip is efficient, then a low MHz version can usually go in a notebook (esp when your chip speed is not dependent on high MHz).
Well, a 1.2 GHz 970 is thought to consume 19 W of power. This could certainly be put into a notebook, in light of the fact that the 7455 was consuming 21.5 W. After a die shrink, there is no reason why a 1.8 GHz 970 couldn't go in a notebook. (I'm secretly hoping for the 17 and 15 inch AlBooks to have 970s and the 12 inch AlBook having a faster G4).
Given the cost associated with designing a new powerbook motherboard, I would say that Apple is going to design one for the 970 instead of for some future G4, and that at worst, they will stick a 1.2 GHz 970 in it. So will any future Moto G4's have a life in the iBook? This is about the only place left for them (and Apple doesn't mind the iBooks running outdated CPUs).
Given the cost associated with designing a new powerbook motherboard, I would say that Apple is going to design one for the 970 instead of for some future G4, and that at worst, they will stick a 1.2 GHz 970 in it. So will any future Moto G4's have a life in the iBook? This is about the only place left for them (and Apple doesn't mind the iBooks running outdated CPUs).
The 7457 is billed as a drop-in replacement for the 7455, so I don't see that any additional engineering would be necessary to accomodate it. It'll run a lot cooler than a 7455, as well.
On the other hand, Apple's had a lot of time to work on RapidIO-based boards, and a goodly amount of time to work on 970-powered, RapidIO-based boards. But a 970 will be significantly hotter than a 7457, and a real challenge for the engineers working on the PowerBook's cooling system.
If the PowerBook does go with the 970, the 7457 would make a nice little chip for the iBook - although AFAIK it'll be hotter than the G3 that's in there now.
So were Apple's as far as Powermacs are concerned; imagine if they didn't have the intuition to push their laptop design to the edge. They'd be hurting for sure.
Screw anything Moto has to say or offer...let them just go die!
If the PowerBook does go with the 970, the 7457 would make a nice little chip for the iBook - although AFAIK it'll be hotter than the G3 that's in there now.
Most anything is hotter than the G3 in the iBook right now...
I hadn't considered the possibility that Apple would already have a RapidIO motherboard. If they do come out with such a motherboard, then no PPC 970's in powerbooks anytime soon (unless Apple is going to pull another Mac IIvx scandal).
The G4 isn't a horrible laptop chip, it just needs a better bus.
If by some miracle Motorola is able to get this dual core chip working and ready in a timely manner, will it have two complete altivec units or one that both cores will share?
This makes me wonder if the dual core PPC 9XX chip will have two or one complete altivec units on it. And if there were two complete altivec systems on a dual core chip, could they be uses as one 256 bit wide altivec unit?
Apple have bought 75% of ALL G4s produced by Motorola.
Seems to me that Apple's not a customer for Motorola to mess with.
Are my numbers exaggerated?
But Moto has been messing with Apple. The fact that a G4 still uses the MPX bus means that Moto isn't paying attention to the needs of their biggest customer. Moto's cancelling the G5 would be nothing but messing with their biggest customer.
I don't know if Moto's messing with Apple is intentional or unintentional. I think that it is mostly unintentional in that they are just plain incompetent, but it is also a bit intentional in that their management doesn't want to spend the money to build the state of the art fabs that are required to keep up to speed with Intel, AMD, and IBM. Moto's self promotion as a provider of embedded CPU's is messing with Apple, plain and simple.
If by some miracle Motorola is able to get this dual core chip working and ready in a timely manner, will it have two complete altivec units or one that both cores will share?
This makes me wonder if the dual core PPC 9XX chip will have two or one complete altivec units on it. And if there were two complete altivec systems on a dual core chip, could they be uses as one 256 bit wide altivec unit?
I presume that if Moto came out with this chip that it would have dual altivec units.
I would say a dual core 970 would have two altivec units. That part is pretty straight forward.
No, two 128 bit altivec units can not be combined to work together as a 256 bit altivec unit. This would be like making a big SUV by trying to glue two cars together. (sure they hold as much as the SUV, but they are still two cars, and moving smething from one car to another is a pain). Dual Altivec units would really work to starve the bandwidth of a machine. Further, any 256 bit altivec unit would have to have custom assembly code written for it (either in individual code, or have altivec aware libraries rewritten), which simply isn't going to happen. Altivec uses 128 bit data chunks and I don't think that is going to change anytime soon.
I think it's silly for Motorola to market the G4 primarily as an embedded chip when 75% of G4 processors clearly are CPUs in desktop computers.
Can history be like this?
In 2000 hot steam went out of Steves ears as the matter know as the 500 MHz debacle unfolded.
Steve: No more Motorola, give me IBM on the phone!
IBM: OK, we have this idea.. if we took our Power4 and stripped it a bit..
Steve: Great!
Steve says to Motorola: We have this awsome CPU from IBM in a couple of years. Screw you!
Motorola: OK, we'll try and pitch the G4 to other customers. They might appriciate our effort. You'll have to take the scraps of their table in the future.
Some months later:
Motorola: Ooops.. the Telecom marked just went belly up! We must cancel our grand CPU plans. Kill the G5, kill HiP7, fire 10 000 people. Kill kill kill! Just like Apple killed the clones!
We still have Apple though.. they just have no one else to turn to. Ha haaa! Losers!
Some years later:
Steve: Well.. We have these furnace like CPUs overclocked as hell (pun intended), draws power like nothin we've ever seen, and have an ancient bus.. Well.. we do have some cool (pun not intended) cases to put them in.. And where the hell is IBM?
Motorola has shown that roadmaps, plans and intentions does not power a computer, a delivered CPU does.
They can have a roadmap for quad core G4 at 6 GHz consuming 2W of power for all I care
While it might be technically possible to put the 970 in all product lines at the same time I do not think it is logistically feasable. My guess is first the towers then the servers after that the iMac and PB.
The towers and the iMac are the weakest lines and they need the 970 the most. I hope that we soon see the end of crippled CPUs in the iMacs and that the difference iMac and tower will be single and dual CPUs.
One VERY significant difference between the 970 and IA-64 is that the 970 will look at the instructions it is given and attempt to reschedule them so that it can keep itself busy. The IA 64 does not do this. The IA 64 forces such scheduling off on the compiler, making for a much more complex and difficult compiler. If I remember correctly, the IA-64 has four way parallelism in its integer units, so it can in theory do quite a bit of work per clock cycle, so long as the compiler was able to keep all the pipes full. No, the magic number isn't 2 integer units and 2 floating point units, the magic number actually seems to be 4 of both because it is very very difficult to figure out a way to keep more than four units busy on most code. You could make a quad core die with only one execution unit, but keeping two busy isn't too difficult, so you might as well do two. The possible performance benefit of such a system would be that if you could get the transistor count to be very low, that you run the CPU faster, but such a CPU would definitely be the poster child for the MHz myth. It is generally understood that parallelism is a good thing in CPU design.
well, the ideal number of execution units depends on the instruction set.
For exemple, on x86, the AMD's Athlon has 3 FPU units.
But on typical FP codes, the average active units is around 2. So in this case, there are no reason to go for 4 units.
In comparaison, the P4, with its unique FP unit, it lagging behind on FP-intensive code. But it is saved by SSE2 and its high clock-rate (what it was really designed for)
There is a huge difference between a 4-units core and a 1-unit quad core. The first one will always gain performance in comparaison to a 1-unit core because the sheduler is able to extract instruction-level parallelism for all code.
The quad core is only fonctionnal on a multi-threaded software, wich is not always possible.
And the OS has the responsability to assign a thread to each core and be sure they all keep busy. This is more easier for the sheduler of the 4-units CPU because he make his decisions on an instruction by instruction basis.
But as the exemple of the x86 show, a 4-units core is perhaps too much. It will probably better to have a 2-core with 2-units each in this case. It is alway the problem of having a well balanced design.
Comments
Originally posted by @homenow
If the 970 costs less than (latest rumour 25% + less than current G4's), then it would be idiotic for Apple to put the more expensive chip in their low cost consumer computers wouldnt it? It would be better for Apple to differentiate between Pro and Consumer with single/dual configs than to use A more expensive chip in the iMac and iBook which are their lowest margin computers.
I can't say anything for sure, but I expect the 970 @ 130nm is cheaper than the 7455 @ 180nm + 1/2MB L3 cache, cache costs a lot. Apple could use a 7457/47 without L3 cache in their low cost consumer products, as they're doing now with the iMac, eMac and the current G4.
Die size for the 7457 is 98 square mm, and 121 for the 970, so it seems it'll be cheaper when produced on an equal process. Depends on MOTs ability to keep up the yields... someone else will probably be able to explain this better, or explain why the 970 will be cheaper anyways, if that's the case
Originally posted by AirSluf
What wafer size is Mot using? IBM is using 300mm wafers at East Fishkill which will also lower production costs as they get more CPU's per wafer despite a larger die size.
All of Mot's fabs are 200mm.
[edit: Except the Crolles2 joint venture, which is 300mm]
Originally posted by @homenow
If the 970 costs less than (latest rumour 25% + less than current G4's), then it would be idiotic for Apple to put the more expensive chip in their low cost consumer computers wouldnt it? It would be better for Apple to differentiate between Pro and Consumer with single/dual configs than to use A more expensive chip in the iMac and iBook which are their lowest margin computers.
True, the 970 may cost less than a G4, but that doesn't mean that Apple would drop G4's. First of all, a 970 needs a new motherboard, and Apple needs to design new motherboards for all the machines that are currently using G4's. Portable and iMac motherboards are much harder than mini ATX motherboards due to their unique space considerations. Personally, I hope that Apple has been on the ball and that when the 970 is rolled out, that we'll see it across the tower lineup and that we'll se it in 17 and 15 inch laptops as well. This of course is wildly over optimistic. Anyhow, my main point is that there are many more considerations than the chip cost that will keep Apple using G4's for a while. It is a significant effort to revamp your entire hardware offerings to use a new chip. (of course, the TiBook is long overdue for a revamp!).
However, Apple may integrate the power saving features on a companion chip somehow. I'm not sure how this would work, but it could be possible that the companion chip would be able to slow down the processor MHz wise while on battery power, turn certain parts of the processor off when not in use (maybe when a non-Altivec program is not in use it can be turned off for example), and other tricks to conserve battery power.
Originally posted by Outsider
And also the word is that the 970 has rudimentary, if any, dedicated power saving logic on the processor, which makes it difficult if not precludes it for being used on a laptop just yet. This is not so much a consideration for the iMac as it is for a truly portable computer like the PowerBook or iBook.
The 970's power saving capabilities are hardly worse than the G4's. The "low power" G4 is just lower voltage and missing tags for L3 cache, and "low power" operation in the 'books consists of throttling the clock down.
If you make a chip efficient in the first place you don't need much in the way of power saving technology.
By then we'll be looking forward to the 980+, or possibly the 990.
I'm gushing lemon juice with excitement.
Can't wait!!!
Lemon Bon Bon
Originally posted by Amorph
The 970's power saving capabilities are hardly worse than the G4's. The "low power" G4 is just lower voltage and missing tags for L3 cache, and "low power" operation in the 'books consists of throttling the clock down.
If you make a chip efficient in the first place you don't need much in the way of power saving technology.
True, if the chip is efficient, then a low MHz version can usually go in a notebook (esp when your chip speed is not dependent on high MHz).
Well, a 1.2 GHz 970 is thought to consume 19 W of power. This could certainly be put into a notebook, in light of the fact that the 7455 was consuming 21.5 W. After a die shrink, there is no reason why a 1.8 GHz 970 couldn't go in a notebook. (I'm secretly hoping for the 17 and 15 inch AlBooks to have 970s and the 12 inch AlBook having a faster G4).
Given the cost associated with designing a new powerbook motherboard, I would say that Apple is going to design one for the 970 instead of for some future G4, and that at worst, they will stick a 1.2 GHz 970 in it. So will any future Moto G4's have a life in the iBook? This is about the only place left for them (and Apple doesn't mind the iBooks running outdated CPUs).
Originally posted by Lemon Bon Bon
I'm gushing lemon juice with excitement.
Originally posted by Yevgeny
Given the cost associated with designing a new powerbook motherboard, I would say that Apple is going to design one for the 970 instead of for some future G4, and that at worst, they will stick a 1.2 GHz 970 in it. So will any future Moto G4's have a life in the iBook? This is about the only place left for them (and Apple doesn't mind the iBooks running outdated CPUs).
The 7457 is billed as a drop-in replacement for the 7455, so I don't see that any additional engineering would be necessary to accomodate it. It'll run a lot cooler than a 7455, as well.
On the other hand, Apple's had a lot of time to work on RapidIO-based boards, and a goodly amount of time to work on 970-powered, RapidIO-based boards. But a 970 will be significantly hotter than a 7457, and a real challenge for the engineers working on the PowerBook's cooling system.
If the PowerBook does go with the 970, the 7457 would make a nice little chip for the iBook - although AFAIK it'll be hotter than the G3 that's in there now.
So were Apple's as far as Powermacs are concerned; imagine if they didn't have the intuition to push their laptop design to the edge. They'd be hurting for sure.
Screw anything Moto has to say or offer...let them just go die!
Originally posted by Amorph
If the PowerBook does go with the 970, the 7457 would make a nice little chip for the iBook - although AFAIK it'll be hotter than the G3 that's in there now.
Most anything is hotter than the G3 in the iBook right now...
I hadn't considered the possibility that Apple would already have a RapidIO motherboard. If they do come out with such a motherboard, then no PPC 970's in powerbooks anytime soon (unless Apple is going to pull another Mac IIvx scandal).
The G4 isn't a horrible laptop chip, it just needs a better bus.
Let's count G4s used by Apple:
PowerMacs: approx 150 000 units / Q the last 16 Qs, 1/3 of those MPs = 3.2 million
PowerBooks: approx 100 000 units / Q the last 8 Qs = .8 million
iMac & eMacs: approx 200 000 units / Q the last 6 Qs = 1.2 million
Xserves: approx 5 000 units / Q the last 4 Qs= .02 million
Sum: 5.2 million.
Apple have bought 75% of ALL G4s produced by Motorola.
Seems to me that Apple's not a customer for Motorola to mess with.
Are my numbers exaggerated?
This makes me wonder if the dual core PPC 9XX chip will have two or one complete altivec units on it. And if there were two complete altivec systems on a dual core chip, could they be uses as one 256 bit wide altivec unit?
Originally posted by Henriok
Motorola say they've shipped 7 million G4s..
Apple have bought 75% of ALL G4s produced by Motorola.
Seems to me that Apple's not a customer for Motorola to mess with.
Are my numbers exaggerated?
But Moto has been messing with Apple. The fact that a G4 still uses the MPX bus means that Moto isn't paying attention to the needs of their biggest customer. Moto's cancelling the G5 would be nothing but messing with their biggest customer.
I don't know if Moto's messing with Apple is intentional or unintentional. I think that it is mostly unintentional in that they are just plain incompetent, but it is also a bit intentional in that their management doesn't want to spend the money to build the state of the art fabs that are required to keep up to speed with Intel, AMD, and IBM. Moto's self promotion as a provider of embedded CPU's is messing with Apple, plain and simple.
Originally posted by sc_markt
If by some miracle Motorola is able to get this dual core chip working and ready in a timely manner, will it have two complete altivec units or one that both cores will share?
This makes me wonder if the dual core PPC 9XX chip will have two or one complete altivec units on it. And if there were two complete altivec systems on a dual core chip, could they be uses as one 256 bit wide altivec unit?
I presume that if Moto came out with this chip that it would have dual altivec units.
I would say a dual core 970 would have two altivec units. That part is pretty straight forward.
No, two 128 bit altivec units can not be combined to work together as a 256 bit altivec unit. This would be like making a big SUV by trying to glue two cars together. (sure they hold as much as the SUV, but they are still two cars, and moving smething from one car to another is a pain). Dual Altivec units would really work to starve the bandwidth of a machine. Further, any 256 bit altivec unit would have to have custom assembly code written for it (either in individual code, or have altivec aware libraries rewritten), which simply isn't going to happen. Altivec uses 128 bit data chunks and I don't think that is going to change anytime soon.
Can history be like this?
In 2000 hot steam went out of Steves ears as the matter know as the 500 MHz debacle unfolded.
Steve: No more Motorola, give me IBM on the phone!
IBM: OK, we have this idea.. if we took our Power4 and stripped it a bit..
Steve: Great!
Steve says to Motorola: We have this awsome CPU from IBM in a couple of years. Screw you!
Motorola: OK, we'll try and pitch the G4 to other customers. They might appriciate our effort. You'll have to take the scraps of their table in the future.
Some months later:
Motorola: Ooops.. the Telecom marked just went belly up! We must cancel our grand CPU plans. Kill the G5, kill HiP7, fire 10 000 people. Kill kill kill! Just like Apple killed the clones!
We still have Apple though.. they just have no one else to turn to. Ha haaa! Losers!
Some years later:
Steve: Well.. We have these furnace like CPUs overclocked as hell (pun intended), draws power like nothin we've ever seen, and have an ancient bus.. Well.. we do have some cool (pun not intended) cases to put them in.. And where the hell is IBM?
Originally posted by Henriok
Kill kill kill!
LOL
D@mn, that was funny!
They can have a roadmap for quad core G4 at 6 GHz consuming 2W of power for all I care
While it might be technically possible to put the 970 in all product lines at the same time I do not think it is logistically feasable. My guess is first the towers then the servers after that the iMac and PB.
The towers and the iMac are the weakest lines and they need the 970 the most. I hope that we soon see the end of crippled CPUs in the iMacs and that the difference iMac and tower will be single and dual CPUs.
Originally posted by Yevgeny
One VERY significant difference between the 970 and IA-64 is that the 970 will look at the instructions it is given and attempt to reschedule them so that it can keep itself busy. The IA 64 does not do this. The IA 64 forces such scheduling off on the compiler, making for a much more complex and difficult compiler. If I remember correctly, the IA-64 has four way parallelism in its integer units, so it can in theory do quite a bit of work per clock cycle, so long as the compiler was able to keep all the pipes full. No, the magic number isn't 2 integer units and 2 floating point units, the magic number actually seems to be 4 of both because it is very very difficult to figure out a way to keep more than four units busy on most code. You could make a quad core die with only one execution unit, but keeping two busy isn't too difficult, so you might as well do two. The possible performance benefit of such a system would be that if you could get the transistor count to be very low, that you run the CPU faster, but such a CPU would definitely be the poster child for the MHz myth. It is generally understood that parallelism is a good thing in CPU design.
well, the ideal number of execution units depends on the instruction set.
For exemple, on x86, the AMD's Athlon has 3 FPU units.
But on typical FP codes, the average active units is around 2. So in this case, there are no reason to go for 4 units.
In comparaison, the P4, with its unique FP unit, it lagging behind on FP-intensive code. But it is saved by SSE2 and its high clock-rate (what it was really designed for)
There is a huge difference between a 4-units core and a 1-unit quad core. The first one will always gain performance in comparaison to a 1-unit core because the sheduler is able to extract instruction-level parallelism for all code.
The quad core is only fonctionnal on a multi-threaded software, wich is not always possible.
And the OS has the responsability to assign a thread to each core and be sure they all keep busy. This is more easier for the sheduler of the 4-units CPU because he make his decisions on an instruction by instruction basis.
But as the exemple of the x86 show, a 4-units core is perhaps too much. It will probably better to have a 2-core with 2-units each in this case. It is alway the problem of having a well balanced design.