None at all. As I've said before, it may be feasible to put a 1.6GHZ 90nm G5 in a PowerBook (though the system controller likely makes it a bit stickier), but are their any real benefits to it, and will it be faster than the 1.5GHZ G4? I think that after the FSB is reduced greatly in speed the G5 begins to lose some of its advantage because bandwidth is reduced.
Shoot, cut it in half one more time and go all the way down to a 375MHz FSB on the G5 (hypothetical bus divisor feature, easy to do), and one of the two 32-bit pipes on the 970 would still be faster than the 167MHz G4 MaxBus.
I seriously doubt those heat disappation numbers for the PPC970FX at 2.5GHz are a good measure of how tenable it is for a laptop. Sure, there are laptops on the x86 with 100W chips, but are IBM's numbers reliable?
I seriously doubt those heat disappation numbers for the PPC970FX at 2.5GHz are a good measure of how tenable it is for a laptop. Sure, there are laptops on the x86 with 100W chips, but are IBM's numbers reliable?
Shoot, cut it in half one more time and go all the way down to a 375MHz FSB on the G5 (hypothetical bus divisor feature, easy to do), and one of the two 32-bit pipes on the 970 would still be faster than the 167MHz G4 MaxBus.
It'd have to be a slower bus because I can't see a laptop having a 750MHZ FSB for heat issues. We also aren't even getting into the differences between a bidirectional and unidirectional bus. I know several engineers who say the 32bit buses are a big problem for the 970 - or at least they see it as one in the future.
While there may be more bandwidth, that doesn't necessarily mean it is used. Also, as has been discussed elsewhere, the MaxBus is actually very efficient at doing what it does.
Even though the bandwidth would be greater, most numbers I've seen run still say that G4 would be pretty equal to the 970 at 1.5GHZ. Plus, the system controller makes the power drain far greater.
Besides, it may all be moot. If the e600 is delivered the FSB would likely be eliminated and the PowerBooks could have a dual core chip with direct access to the RAM. If this is in the offing, why would Apple move to the G5 in the PowerBooks? One, I say they can't do it because the required controller's requirements are too high.
Even though the bandwidth would be greater, most numbers I've seen run still say that G4 would be pretty equal to the 970 at 1.5GHZ. Plus, the system controller makes the power drain far greater.
I'm not convincde though (except about the power consumption part).
Quote:
PowerPC rocks on floating point thanks to a great SIMD unit
On single precision FP, Altivec is tough to beat. The Athlon has 3 scalar FPU units, more than the 2 the 970 has (but the more units you add, the harder it is to take advantage of them). The G4 actually has a comparatively weak scalar FPU.
The PIV can do "normal" FPU operations (2x 64bit FP numbers) through its SIMD unit (SSE2) to avoid the regular mess of x86/7 FPU.
Quote:
maybe it would be best to call the e600 that "M5" and begin to clearly delineate the mobile chips from the desktop chips.
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
I'm not convincde though (except about the power consumption part).
On single precision FP, Altivec is tough to beat. The Athlon has 3 scalar FPU units, more than the 2 the 970 has (but the more units you add, the harder it is to take advantage of them). The G4 actually has a comparatively weak scalar FPU.
The PIV can do "normal" FPU operations (2x 64bit FP numbers) through its SIMD unit (SSE2) to avoid the regular mess of x86/7 FPU.
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
Also, how does Centrino do on this. I think the 970 is the chip that they need to combat the AMD64. The e600 would be going up against a mobility Pentium. I believe Altivec stacks up very well there (and I think FreeScale is adding more FPUs to the e600). Add an on-chip memory controller and dual-core and we have a winner in the PowerBooks, especially if they can get it to 2GHZ dual-core with solid power consumption. I'll give Moto one thing, the G4 has always been a miser when it comes to power.
Quote:
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
M4 then. The e700 could be M5. I think it would be best to remove the G4 designation if the e600 hits.
Quote:
(about RAM:)
Cache. And more cache.
That's really the best thing, though memory is getting faster. An on-chip memory controller with fast memory plugged into it would provide a significant improvement for the G4. Add to that improved architecture and increased cache and an e600 would be much fast than the current 7447A at the same clockrate.
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
Altivec's number of units is largely fine as it is. Most Altivec code is bandwidth limited rather than execution rate limited.
In all these discussion I cannot understand what people are talking about! Why would I car about the kind of CPU in a laptop:
as long as it's fast like hell
friendly to battery life
as long as laptops do not accommodate more than 4GB of RAM
it does not fry my any part of my body
To be honest, the current 1.5GHz line is just plentyfast (tm)! If you need more power, will need a dual CPU desktop.
Personally I give sh*t whether it's Motorola G4 or IBM G5, I just want power and battery life in a reasonably small package. Instead of having a G5 eat my battery I'd rather gave a 2x1GHz 7447A!
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
The AltiVec units are fine, they just need more bandwidth -- which can be delivered by an on-chip memory controller and ditching the FSB. The AltiVec unit is a vector unit, and only a vector unit. Motorola could specifically work on the G4's FPU which has seen essentially no changes since it was first put into the 604 back in '95 (just a pipeline lengthening with the 7450), and that would improve the scalar FPU performance. There are some internal resource limitations that could be addressed that would improve the G4's FPU performance noticably without a big investment, plus they could add an additional unit if they determined it was important enough.
I've maintained for years that a 2 GHz G4 w/ on-chip memory controller would be an impressive chip -- better in some ways than the 970. Moving to a 90nm process, moving a few other devices on-chip (Ethernet, DMA, USB, FireWire) and adding RIO/HT for the rest would be an excellent solution for Apple's portable and compact machine needs. It would also free up the IBM design team(s) to work on more desktop chips for the towers.
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
It doesn't have to - if they release a whole new processor brand (ie M5 or i5), the "5" merely indicates a performance level (more than 4), and the "G" could then indicate 64-bit.
One apparent fear is that putting a G5 in the iMac will cannibalize PowerMac sales. It's still true that consumers don't "need" 64-bit. The only way to solve this is to make "64-bit" synonymous with "Power User" and "Power{mac,book}" while keeping the consumer line at 32 bit. When {Photoshop,Shake,Autocad?,etc} is 64-bit optimized, what "pro" will settle for a 32-bit offering?
A 32-bit, low power, 2+GHz "{M,i}5" processor gives an upgrade path for the entire "i" line while providing real differentiation between the "i" and "Power" lines.
This is all just marketing. The only problem is what to put in the Powerbooks. Fast 32-bit processors are fine for consumers at least for 3-4 more years.
How about this: The next time someone uses the word "canabalize" with respect to Apple product line configurations gets 30 lashes.
Lets face it guys; Apple is fighting for its life, at this point any and all sales are good. The only thing impacting sales of the PowerMac is Apples marketing of the device. If that doesn't change soon then you might as well eat the PowerMac yourself.
speculate for me when the e600 could be ready to go
It's really hard to say. If it were announced by Motorola's semiconductor division under old management then most of us would think that we'd never see it. FreeScale gives us hope that this is different. Also the new Crolles 2 fab does that same. I personally think we may see the first ones filter through in the fall.
As others who know more about internal CPU design than I have indicated, the e600 looks to be a very nice, low power chip. If FreeScale can eliminate the FSB and place the memory controller, USB, Firewire, and ethernet on the chip while upgrading the architecture slightly then this chip would be a strong addition for the Mac. Add to that dual-core availability and I believe the PowerBook becomes one of the fastest laptops around, especially in respects to power consumption and heat.
....I've maintained for years that a 2 GHz G4 w/ on-chip memory controller would be an impressive chip -- ...., moving a few other devices on-chip (Ethernet, DMA, USB, FireWire) and adding RIO/HT for the rest would be an excellent solution for Apple's portable and compact machine needs. ......
So sad, isn't it.
And we still haven't seen the mythical MPC8540/8560, granted not desktop cpu's, but still a severe indictment of Motorola. Remember all the excitement and speculation surrounding these chips and what people believed would be in Apple computers. Time flies, I can't believe it's been soooo verrrryyyy long.
My info was that the 7448, a 90-nm G4 w/1 MB L2 cache (up to 2 GHz), wouldn't be ready until early 2005. That would suggest, optimistically, a MWSF introduction, or perhaps a release in Feb-March. Where that would leave the PowerBook line until then, I don't know. Going beyond MWSF would be another looooong update cycle.
The 746x (argh, did I remember that right?), with integrated memory controllers and up to 2 cores, is scheduled for late 2005. *That* should be a nice chip, if it ever comes out.
Whether the projected schedules come anywhere close to reality, who knows.
Comments
Originally posted by Mr. MacPhisto
None at all. As I've said before, it may be feasible to put a 1.6GHZ 90nm G5 in a PowerBook (though the system controller likely makes it a bit stickier), but are their any real benefits to it, and will it be faster than the 1.5GHZ G4? I think that after the FSB is reduced greatly in speed the G5 begins to lose some of its advantage because bandwidth is reduced.
G4 @ 1.33 GHz, 64-bit-wide bidirectional 167MHz FSB:
-> 1.3GB/s or so max reading- or writing-bandwidth
hypothetical G5 @ 1.5GHz, 750MHz FSB (32-bit read, 32-bit write)
-> 3GB/s max reading-bandwidth
-> 3GB/s max writing-bandwidth
-> can do both at the same time.
Shoot, cut it in half one more time and go all the way down to a 375MHz FSB on the G5 (hypothetical bus divisor feature, easy to do), and one of the two 32-bit pipes on the 970 would still be faster than the 167MHz G4 MaxBus.
Just look at the radiator-like liquid cooling system images from the new G5. This thing is going NO WHERE NEAR a laptop.
Originally posted by Existence
I seriously doubt those heat disappation numbers for the PPC970FX at 2.5GHz are a good measure of how tenable it is for a laptop. Sure, there are laptops on the x86 with 100W chips, but are IBM's numbers reliable?
Just look at the radiator-like liquid cooling system images from the new G5. This thing is going NO WHERE NEAR a laptop.
Not at 2.5 GHz it isn't..
Originally posted by boots
G4 @ 1.33 GHz, 64-bit-wide bidirectional 167MHz FSB:
-> 1.3GB/s or so max reading- or writing-bandwidth
hypothetical G5 @ 1.5GHz, 750MHz FSB (32-bit read, 32-bit write)
-> 3GB/s max reading-bandwidth
-> 3GB/s max writing-bandwidth
-> can do both at the same time.
Shoot, cut it in half one more time and go all the way down to a 375MHz FSB on the G5 (hypothetical bus divisor feature, easy to do), and one of the two 32-bit pipes on the 970 would still be faster than the 167MHz G4 MaxBus.
It'd have to be a slower bus because I can't see a laptop having a 750MHZ FSB for heat issues. We also aren't even getting into the differences between a bidirectional and unidirectional bus. I know several engineers who say the 32bit buses are a big problem for the 970 - or at least they see it as one in the future.
While there may be more bandwidth, that doesn't necessarily mean it is used. Also, as has been discussed elsewhere, the MaxBus is actually very efficient at doing what it does.
Even though the bandwidth would be greater, most numbers I've seen run still say that G4 would be pretty equal to the 970 at 1.5GHZ. Plus, the system controller makes the power drain far greater.
Besides, it may all be moot. If the e600 is delivered the FSB would likely be eliminated and the PowerBooks could have a dual core chip with direct access to the RAM. If this is in the offing, why would Apple move to the G5 in the PowerBooks? One, I say they can't do it because the required controller's requirements are too high.
Even though the bandwidth would be greater, most numbers I've seen run still say that G4 would be pretty equal to the 970 at 1.5GHZ. Plus, the system controller makes the power drain far greater.
I'm not convincde though (except about the power consumption part).
PowerPC rocks on floating point thanks to a great SIMD unit
On single precision FP, Altivec is tough to beat. The Athlon has 3 scalar FPU units, more than the 2 the 970 has (but the more units you add, the harder it is to take advantage of them). The G4 actually has a comparatively weak scalar FPU.
The PIV can do "normal" FPU operations (2x 64bit FP numbers) through its SIMD unit (SSE2) to avoid the regular mess of x86/7 FPU.
maybe it would be best to call the e600 that "M5" and begin to clearly delineate the mobile chips from the desktop chips.
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
(about RAM:)
any solutions or remedies?
Cache. And more cache.
Originally posted by Stoo
I'm not convincde though (except about the power consumption part).
On single precision FP, Altivec is tough to beat. The Athlon has 3 scalar FPU units, more than the 2 the 970 has (but the more units you add, the harder it is to take advantage of them). The G4 actually has a comparatively weak scalar FPU.
The PIV can do "normal" FPU operations (2x 64bit FP numbers) through its SIMD unit (SSE2) to avoid the regular mess of x86/7 FPU.
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
Also, how does Centrino do on this. I think the 970 is the chip that they need to combat the AMD64. The e600 would be going up against a mobility Pentium. I believe Altivec stacks up very well there (and I think FreeScale is adding more FPUs to the e600). Add an on-chip memory controller and dual-core and we have a winner in the PowerBooks, especially if they can get it to 2GHZ dual-core with solid power consumption. I'll give Moto one thing, the G4 has always been a miser when it comes to power.
Quote:
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
M4 then. The e700 could be M5. I think it would be best to remove the G4 designation if the e600 hits.
Quote:
(about RAM:)
Cache. And more cache.
That's really the best thing, though memory is getting faster. An on-chip memory controller with fast memory plugged into it would provide a significant improvement for the G4. Add to that improved architecture and increased cache and an e600 would be much fast than the current 7447A at the same clockrate.
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
Altivec's number of units is largely fine as it is. Most Altivec code is bandwidth limited rather than execution rate limited.
- as long as it's fast like hell
- friendly to battery life
- as long as laptops do not accommodate more than 4GB of RAM
- it does not fry my any part of my body
To be honest, the current 1.5GHz line is just plentyfast (tm)! If you need more power, will need a dual CPU desktop.Personally I give sh*t whether it's Motorola G4 or IBM G5, I just want power and battery life in a reasonably small package. Instead of having a G5 eat my battery I'd rather gave a 2x1GHz 7447A!
Originally posted by Mr. MacPhisto
How would increased Altivec units hold up, and can it be worked on to increase its ability in regards to scalar FPU (I honestly know very little in the area)?
The AltiVec units are fine, they just need more bandwidth -- which can be delivered by an on-chip memory controller and ditching the FSB. The AltiVec unit is a vector unit, and only a vector unit. Motorola could specifically work on the G4's FPU which has seen essentially no changes since it was first put into the 604 back in '95 (just a pipeline lengthening with the 7450), and that would improve the scalar FPU performance. There are some internal resource limitations that could be addressed that would improve the G4's FPU performance noticably without a big investment, plus they could add an additional unit if they determined it was important enough.
I've maintained for years that a 2 GHz G4 w/ on-chip memory controller would be an impressive chip -- better in some ways than the 970. Moving to a 90nm process, moving a few other devices on-chip (Ethernet, DMA, USB, FireWire) and adding RIO/HT for the rest would be an excellent solution for Apple's portable and compact machine needs. It would also free up the IBM design team(s) to work on more desktop chips for the towers.
...
Originally posted by Stoo
The "5" part of the name still suggests that it's the generation after G4, which implies 64 bit.
It doesn't have to - if they release a whole new processor brand (ie M5 or i5), the "5" merely indicates a performance level (more than 4), and the "G" could then indicate 64-bit.
One apparent fear is that putting a G5 in the iMac will cannibalize PowerMac sales. It's still true that consumers don't "need" 64-bit. The only way to solve this is to make "64-bit" synonymous with "Power User" and "Power{mac,book}" while keeping the consumer line at 32 bit. When {Photoshop,Shake,Autocad?,etc} is 64-bit optimized, what "pro" will settle for a 32-bit offering?
A 32-bit, low power, 2+GHz "{M,i}5" processor gives an upgrade path for the entire "i" line while providing real differentiation between the "i" and "Power" lines.
This is all just marketing. The only problem is what to put in the Powerbooks. Fast 32-bit processors are fine for consumers at least for 3-4 more years.
How about this: The next time someone uses the word "canabalize" with respect to Apple product line configurations gets 30 lashes.
Lets face it guys; Apple is fighting for its life, at this point any and all sales are good. The only thing impacting sales of the PowerMac is Apples marketing of the device. If that doesn't change soon then you might as well eat the PowerMac yourself.
Dave
Originally posted by Peter North
speculate for me when the e600 could be ready to go
It's really hard to say. If it were announced by Motorola's semiconductor division under old management then most of us would think that we'd never see it. FreeScale gives us hope that this is different. Also the new Crolles 2 fab does that same. I personally think we may see the first ones filter through in the fall.
As others who know more about internal CPU design than I have indicated, the e600 looks to be a very nice, low power chip. If FreeScale can eliminate the FSB and place the memory controller, USB, Firewire, and ethernet on the chip while upgrading the architecture slightly then this chip would be a strong addition for the Mac. Add to that dual-core availability and I believe the PowerBook becomes one of the fastest laptops around, especially in respects to power consumption and heat.
didnt you say that apple canceled the "vx" project from ibm because of problems with migrating to 90nm?
did you not also state that apple either sent back or canceled an order of 130nm g5 chips?
indulge me........i have a reason for the clarifications.
Originally posted by Programmer
....I've maintained for years that a 2 GHz G4 w/ on-chip memory controller would be an impressive chip -- ...., moving a few other devices on-chip (Ethernet, DMA, USB, FireWire) and adding RIO/HT for the rest would be an excellent solution for Apple's portable and compact machine needs. ......
So sad, isn't it.
And we still haven't seen the mythical MPC8540/8560, granted not desktop cpu's, but still a severe indictment of Motorola. Remember all the excitement and speculation surrounding these chips and what people believed would be in Apple computers. Time flies, I can't believe it's been soooo verrrryyyy long.
I'm rooting for Freescale.
The 746x (argh, did I remember that right?), with integrated memory controllers and up to 2 cores, is scheduled for late 2005. *That* should be a nice chip, if it ever comes out.
Whether the projected schedules come anywhere close to reality, who knows.