Seems as though IBM already has a companion chip in the wings... hmmm.</strong>
It probably does for its own $10K workstations, but I think Apple is a different story. The image doesn't say anything about a core logic ASIC, just that the PPC 970 needs an ASIC that supports its bus. Perhaps Apple will take the current U2 chip in the PowerMacs and modify it for the 970 bus, and everything else stays the same. This would seem to me the best solution until a better core logic ASIC can be designed.
Seems as though IBM already has a companion chip in the wings... hmmm.</strong>
It probably does for its own $10K workstations, but I think Apple is a different story. The image doesn't say anything about a core logic ASIC, just that the PPC 970 needs an ASIC that supports its bus. Perhaps Apple will take the current U2 chip in the PowerMacs and modify it for the 970 bus, and everything else stays the same. This would seem to me the best solution until a better core logic ASIC can be designed.[/QB]<hr></blockquote>
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).
<strong>I have to disagree here. If the PPC970 is twice as fast (which it should be) putting only one in a machine that used to be a dual processor machine would mean the machine would end up being no faster. At least in Apples "duals make up for the Mhz laps" marketing campagain. I tend to think that unless the GPUL is much more than twice as fast they will still need to put duals in.
</strong><hr></blockquote>
Note that my comment was about putting duals in across the board for the pro line. As I said, I believe Apple will always have at least one dual for the high end. Additionally, the new chip will be "twice as fast" at the same clock speed and it will also be availabe at higher clock speeds.
[quote]<strong>
We aren't trying to simply speed up the Macs in increments. We need to climb out of the speed hole we have fallen in. Electric Image (originally written on the Mac) is now cross platform and is rendering 2-3 times as fast on a PC as it is on the Mac. And this is from a VERY Mac friendly company.
</strong><hr></blockquote>
Also note that everytime people show an example of "how slow the G4 is", they use a 3D rendering program which is highly FP intensive. It's no secret that the G4, particularly the G4+ (7455 and above) is slow at FP. According to C't, the 1GHZ has a SpecFP score of only 187, while the new ppc970 will score 1050 or so at 1.8 ghz. Scaling for clock speed, that still gives the new chip a greater than 3x increase in FP performance at the same clock speed. Based on the RC5 numbers, it looks like the SIMD unit will still be powerful enough to give the PPC970 an advantage there as well. I'd expect the ppc970 to fall behind in INT performance just a bit to the equivalent P4 at the time of introduction. All said, the ppc970 won't crush the competition across the board, but it will compete nicely and put single processor Macs "back in the game".
[quote]<strong>
Then again, if the PPC970 is constained, I could see Apple shipping single processor machines until the line could be revved to duals again.
</strong><hr></blockquote>
Exactly.
[quote]Originally posted by krassy:
<strong>
so what have they all done from 1999 till now???? nothing? ... why is the g4 still produced at .18µ ?
i really don't get that...
</strong><hr></blockquote>
For starters, I don't want to put myself in a position of having to defend Motorola. However, over the past 3 years since the introduction of the G4, Motorola moved to a new production process (was .20u, now at .18u + SOI), and they changed the chip by deepening the instruction pipeline from 4 to 7 stages, added and int and altivec unit, etc. While this isn't great, it's not as if nothing has happend since 1999. I suspect the .13u + 512k L2 cache chips will be coming this January/February timeframe.
[quote]Originally posted by DaveGee:
<strong>
Well... if MOT knew of Apples move to IBM and they pretty much would have had to since it would been pretty strange for Apple not to 'sign up for' (or in some way officialy commit) MOTs next stage. Once MOT knew or suspected such a move happened how much effort do you think MOT would have given to the progress of the G4? I dunno if any of this is true to it sure is one way to explain why so very little has been done with the G4.
</strong><hr></blockquote>
Actually, I think Motorola does have incentive to continue pushing the G4. OTOH, they have no incentive to do any sort of desktop capable G5.
The G4 still has plenty of life left in it. I'd expect Apple to want to keep the G4 in the consumer machines for another year or two easily. If Motorola didn't keep the G4 attractive for the low end market, they could lose Apple's consumer line all together. They probably will eventually, but I'm sure they'll want to milk this design for as long as they can.
Actually, I think Motorola does have incentive to continue pushing the G4. OTOH, they have no incentive to do any sort of desktop capable G5.
The G4 still has plenty of life left in it. I'd expect Apple to want to keep the G4 in the consumer machines for another year or two easily. If Motorola didn't keep the G4 attractive for the low end market, they could lose Apple's consumer line all together. They probably will eventually, but I'm sure they'll want to milk this design for as long as they can.
Steve</strong><hr></blockquote>
It's sounds right : pushing the G4 design in his lattest developpement seems logical. Afterall this chip is the best embedded chip of the year 2001, a chip that brings Money to Mot. If they stop increasing the performance of this chip, many others customers like Cisco will search an other supplier.
Cisco is most likely going to go with the Motorola 'G5' 85xx series in the near future...
So for the next two years, I would expect the Mac consumer line to run on Motorola G4s, and the iBook might 'move up' to a Moto G4, or maybe IBM will hook up a VMX unit to the Sahara G3, and unlock the bus speed...
All in all, Moto is on its way out of the Apple spotlight...
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).</strong>
Because they are or it is not suitable for Apple's needs. The current core logic ASIC for the Power4 machines does not support AGP, just PCI-X. It is not intregrated with Firewire, ethernet or ATA buses. It probably is too costly and consumes too much power.
It just seems to me that Apple's best option for now is to take the current "U2" ASIC, throw out the MPX bus support, add support for DDR 400 or DDR-2, add support for this new GX style bus, and integrate the various high bandwidth IO onto it.
A single 1.2 GHz IBM PPC 970 uses 19 watts typical.
A single 1 GHz Motorola MPC7455 uses 21.3 watts typical...30 watts maximum. My computer has a pair, so that's 42.6-60 watts.</strong><hr></blockquote>
Not confused, you didn't read the post I replied to. The single 1.8ghz uses-
[quote]At 1.8GHz, the PowerPC 970 will consume 1.3-volts and dissipate 42-Watts.<hr></blockquote>.
So a dual would be quite high around 60-80 if you double it. I wasn't talking about the 1.2ghz, I was talking about the 1.8ghz and it's 42 watts. I just think that's a lot of juice.
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).</strong>
Because they are or it is not suitable for Apple's needs. The current core logic ASIC for the Power4 machines does not support AGP, just PCI-X. It is not intregrated with Firewire, ethernet or ATA buses. It probably is too costly and consumes too much power.
It just seems to me that Apple's best option for now is to take the current "U2" ASIC, throw out the MPX bus support, add support for DDR 400 or DDR-2, add support for this new GX style bus, and integrate the various high bandwidth IO onto it.[/QB]<hr></blockquote>
I don't think the GX bus is compatible with the bus used on the 970. For one thing the Power4 does not use a DDR version of the bus. Doesn't need to. The higher bandwidth stuff like memory has it's own controller. Therefore a controller chip for the 970 would have to be built from the ground up. And why not include AGP and PCI-X support into it especially if it is to be used as a low cost solution for IBM as well as Apple? Well time will tell as soon as IBM releases info on this controller. And Apple can always put firewire and ethernet on the peripheral controller with out giving up too much performance than if it was directly located on the main controller.
I don't think the GX bus is compatible with the bus used on the 970. ... Therefore a controller chip for the 970 would have to be built from the ground up. And why not include AGP and PCI-X support into it especially if it is to be used as a low cost solution for IBM as well as Apple?</strong>
Perhaps. But you have to keep in mind that IBM will be using the PPC 970 in $10K workstations and servers. They'll have PCI-X, but AGP is questionable let alone integration of high bandwidth I/O stuff (like 1394, Ethernet or ATA). But these machines will have different requirements than Apple's $1.5K to $3.5K machines.
<strong>Well time will tell as soon as IBM releases info on this controller. And Apple can always put firewire and ethernet on the peripheral controller with out giving up too much performance than if it was directly located on the main controller.</strong>
What kind of memory would actually support a FSB of 450MHz? </strong>
Rambus A nice little dual channel 450 MHz DRDRAM solution can pump 3.6 GB/s. Quad channel will go 7.2 GB/s.
Otherwise if they can have a 9:4 processor bus to memory bus ratio, then DDR 400 or DDR-2 will work, dual channel or otherwise. Or the easier solution is to only use processor clock rates that will work at lower integer multiples.
Rambus A nice little dual channel 450 MHz DRDRAM solution can pump 3.6 GB/s. Quad channel will go 7.2 GB/s.
Otherwise if they can have a 9:4 processor bus to memory bus ratio, then DDR 400 or DDR-2 will work, dual channel or otherwise. Or the easier solution is to only use processor clock rates that will work at lower integer multiples.
Not certain whether anyone has posted this link as yet and I don't think it adds anything new other than a specific ApplePI mention, but it's worth reading just for more colour.
The 970 should available at the end of 2003, right?
At speeds up to 1.8GHz, right?
900MHz FSB, right?
2.9MIPS/ MHz, right?
The G4 should reach 1.8GHz at the end of 2003, right?
133MHz FSB today, at the end of the year maybe 800MHz (RIO), right?
2.3MIPS/ MHz, right?
Thats an advantage of 2.9/ 2.3 = 1.26 points , right?
It doesn't impress me.
I hope i am wrong!
The only advantage of the 970 is 64bit and the SPEC benchmark!</strong><hr></blockquote>
Well, consider this then,
1. The PPC970 has not one, but TWO Altivec units. Altivec performance will be off the charts. Furthermore, the PPC970's altivec performance won't be choked by a slow bus....according to those who know about this sort of thing, the current G4 altivec quickly chokes because the MPX bus cannot feed it fast enough. So not only will a major bottleneck be removed, but an additional SIMD unit will be added. For altivec-optimized applications, the 970 will shame Pentium 4s, no doubt about it.
2. The typical trade-off for MHz is to lower the MIPS/MHz, partially by adding extra pipelines. Yet, the 970 clocks significantly faster than the G4, and STILL turns out more MIPS. Also, the 970 is going to debut at 1.8 GHz, probably on a 130 nm process. This leaves lots of room for scaling...IBM could migrate the 970 to 90 nm fairly quickly, and probably scale to 3 GHz within a year or so of the 970's introduction. In contrast, the G4 debuted at 400 MHz and is near the end of its life.
I think the PPC 970 is far, far superior to the G4, and it's going to be faster than x86 for some time after it's introduction. Maybe on par with x86 for FP caculations, but for any Altivec enabled tasks, the CPU is going to blow x86 away.
Apple was right all along about their focus on altivec....it's a secret weapon that has allowed Apple to keep up despite the disastrous G4, and it will enable Apple to smoke the competition when IBM finally begins fabbing the PPC.
The general concencus around here suggests that Apple will adopt the 970 for their high-end (or possibly a new ultra high-end) Mac and either use a scaled down version of the chip (or a G4) for consumer machines.
After reading all the reports, press releases and following these threads, I keep coming back to a theory i had which suggests that the 970 is *not* the high (or ultra high-end) processor, but instead, the chip intended for use within *consumer* (and possibly low-high-end) machines. The differences simply being the GHz or number of processors used.
That of course leads us to the question about Apple's high-end machines... It wouldn't seem unreasonable that Apple would decide to make its high-end machines (ultra high-end machines) equipped with a genuine Power4 chip.
This would make Apple's consumer models running in parallel speed (maybe slightly higher) with that of x86's highest-end systems while putting Apple's high-end leaps and bounds ahead.
That of course leads us to the question about Apple's high-end machines... It wouldn't seem unreasonable that Apple would decide to make its high-end machines (ultra high-end machines) equipped with a genuine Power4 chip.
This would make Apple's consumer models running in parallel speed (maybe slightly higher) with that of x86's highest-end systems while putting Apple's high-end leaps and bounds ahead.
It seems like a natural fit if you ask me.</strong><hr></blockquote>
Or maybe there's something like a 990 that we don't know about yet.... after all, those in the know have made it fairly clear that the 970 is not an only child.
1. The PPC970 has not one, but TWO Altivec units. Altivec performance will be off the charts. Furthermore, the PPC970's altivec performance won't be choked by a slow bus....according to those who know about this sort of thing, the current G4 altivec quickly chokes because the MPX bus cannot feed it fast enough. So not only will a major bottleneck be removed, but an additional SIMD unit will be added. For altivec-optimized applications, the 970 will shame Pentium 4s, no doubt about it.
2. The typical trade-off for MHz is to lower the MIPS/MHz, partially by adding extra pipelines. Yet, the 970 clocks significantly faster than the G4, and STILL turns out more MIPS. Also, the 970 is going to debut at 1.8 GHz, probably on a 130 nm process. This leaves lots of room for scaling...IBM could migrate the 970 to 90 nm fairly quickly, and probably scale to 3 GHz within a year or so of the 970's introduction. In contrast, the G4 debuted at 400 MHz and is near the end of its life.
</strong><hr></blockquote>
1. i thought the VMX unit on the 970 is divided into two seperate working units which work on the same instructions but split them up before processing (data handling, instruction processing or s.th.) however - i'm no expert on this but i think this is the way the G4/7400 managed altivec-processing too.... so these TWO altivec units are nothing more than ONE altivec unit...
2. the G4 will be available at speeds up to 1.8Ghz on a .13µ process in the first half of 2003...
Comments
Seems as though IBM already has a companion chip in the wings... hmmm.</strong>
It probably does for its own $10K workstations, but I think Apple is a different story. The image doesn't say anything about a core logic ASIC, just that the PPC 970 needs an ASIC that supports its bus. Perhaps Apple will take the current U2 chip in the PowerMacs and modify it for the 970 bus, and everything else stays the same. This would seem to me the best solution until a better core logic ASIC can be designed.
<strong>[qb]Originally posted by Outsider:
Seems as though IBM already has a companion chip in the wings... hmmm.</strong>
It probably does for its own $10K workstations, but I think Apple is a different story. The image doesn't say anything about a core logic ASIC, just that the PPC 970 needs an ASIC that supports its bus. Perhaps Apple will take the current U2 chip in the PowerMacs and modify it for the 970 bus, and everything else stays the same. This would seem to me the best solution until a better core logic ASIC can be designed.[/QB]<hr></blockquote>
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).
<strong>I have to disagree here. If the PPC970 is twice as fast (which it should be) putting only one in a machine that used to be a dual processor machine would mean the machine would end up being no faster. At least in Apples "duals make up for the Mhz laps" marketing campagain. I tend to think that unless the GPUL is much more than twice as fast they will still need to put duals in.
</strong><hr></blockquote>
Note that my comment was about putting duals in across the board for the pro line. As I said, I believe Apple will always have at least one dual for the high end. Additionally, the new chip will be "twice as fast" at the same clock speed and it will also be availabe at higher clock speeds.
[quote]<strong>
We aren't trying to simply speed up the Macs in increments. We need to climb out of the speed hole we have fallen in. Electric Image (originally written on the Mac) is now cross platform and is rendering 2-3 times as fast on a PC as it is on the Mac. And this is from a VERY Mac friendly company.
</strong><hr></blockquote>
Also note that everytime people show an example of "how slow the G4 is", they use a 3D rendering program which is highly FP intensive. It's no secret that the G4, particularly the G4+ (7455 and above) is slow at FP. According to C't, the 1GHZ has a SpecFP score of only 187, while the new ppc970 will score 1050 or so at 1.8 ghz. Scaling for clock speed, that still gives the new chip a greater than 3x increase in FP performance at the same clock speed. Based on the RC5 numbers, it looks like the SIMD unit will still be powerful enough to give the PPC970 an advantage there as well. I'd expect the ppc970 to fall behind in INT performance just a bit to the equivalent P4 at the time of introduction. All said, the ppc970 won't crush the competition across the board, but it will compete nicely and put single processor Macs "back in the game".
[quote]<strong>
Then again, if the PPC970 is constained, I could see Apple shipping single processor machines until the line could be revved to duals again.
</strong><hr></blockquote>
Exactly.
[quote]Originally posted by krassy:
<strong>
so what have they all done from 1999 till now???? nothing? ... why is the g4 still produced at .18µ ?
i really don't get that...
</strong><hr></blockquote>
For starters, I don't want to put myself in a position of having to defend Motorola. However, over the past 3 years since the introduction of the G4, Motorola moved to a new production process (was .20u, now at .18u + SOI), and they changed the chip by deepening the instruction pipeline from 4 to 7 stages, added and int and altivec unit, etc. While this isn't great, it's not as if nothing has happend since 1999. I suspect the .13u + 512k L2 cache chips will be coming this January/February timeframe.
[quote]Originally posted by DaveGee:
<strong>
Well... if MOT knew of Apples move to IBM and they pretty much would have had to since it would been pretty strange for Apple not to 'sign up for' (or in some way officialy commit) MOTs next stage. Once MOT knew or suspected such a move happened how much effort do you think MOT would have given to the progress of the G4? I dunno if any of this is true to it sure is one way to explain why so very little has been done with the G4.
</strong><hr></blockquote>
Actually, I think Motorola does have incentive to continue pushing the G4. OTOH, they have no incentive to do any sort of desktop capable G5.
The G4 still has plenty of life left in it. I'd expect Apple to want to keep the G4 in the consumer machines for another year or two easily. If Motorola didn't keep the G4 attractive for the low end market, they could lose Apple's consumer line all together. They probably will eventually, but I'm sure they'll want to milk this design for as long as they can.
Steve
<strong>
Actually, I think Motorola does have incentive to continue pushing the G4. OTOH, they have no incentive to do any sort of desktop capable G5.
The G4 still has plenty of life left in it. I'd expect Apple to want to keep the G4 in the consumer machines for another year or two easily. If Motorola didn't keep the G4 attractive for the low end market, they could lose Apple's consumer line all together. They probably will eventually, but I'm sure they'll want to milk this design for as long as they can.
Steve</strong><hr></blockquote>
It's sounds right : pushing the G4 design in his lattest developpement seems logical. Afterall this chip is the best embedded chip of the year 2001, a chip that brings Money to Mot. If they stop increasing the performance of this chip, many others customers like Cisco will search an other supplier.
So for the next two years, I would expect the Mac consumer line to run on Motorola G4s, and the iBook might 'move up' to a Moto G4, or maybe IBM will hook up a VMX unit to the Sahara G3, and unlock the bus speed...
All in all, Moto is on its way out of the Apple spotlight...
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).</strong>
Because they are or it is not suitable for Apple's needs. The current core logic ASIC for the Power4 machines does not support AGP, just PCI-X. It is not intregrated with Firewire, ethernet or ATA buses. It probably is too costly and consumes too much power.
It just seems to me that Apple's best option for now is to take the current "U2" ASIC, throw out the MPX bus support, add support for DDR 400 or DDR-2, add support for this new GX style bus, and integrate the various high bandwidth IO onto it.
<strong>
I think you've confused something here.
A single 1.2 GHz IBM PPC 970 uses 19 watts typical.
A single 1 GHz Motorola MPC7455 uses 21.3 watts typical...30 watts maximum. My computer has a pair, so that's 42.6-60 watts.</strong><hr></blockquote>
Not confused, you didn't read the post I replied to. The single 1.8ghz uses-
[quote]At 1.8GHz, the PowerPC 970 will consume 1.3-volts and dissipate 42-Watts.<hr></blockquote>.
So a dual would be quite high around 60-80 if you double it. I wasn't talking about the 1.2ghz, I was talking about the 1.8ghz and it's 42 watts. I just think that's a lot of juice.
<strong>[qb]Originally posted by Outsider:
Why not use the one IBM has already? Apple would only need to design a moderately not-to-complex peripheral controller (basic IO).</strong>
Because they are or it is not suitable for Apple's needs. The current core logic ASIC for the Power4 machines does not support AGP, just PCI-X. It is not intregrated with Firewire, ethernet or ATA buses. It probably is too costly and consumes too much power.
It just seems to me that Apple's best option for now is to take the current "U2" ASIC, throw out the MPX bus support, add support for DDR 400 or DDR-2, add support for this new GX style bus, and integrate the various high bandwidth IO onto it.[/QB]<hr></blockquote>
I don't think the GX bus is compatible with the bus used on the 970. For one thing the Power4 does not use a DDR version of the bus. Doesn't need to. The higher bandwidth stuff like memory has it's own controller. Therefore a controller chip for the 970 would have to be built from the ground up. And why not include AGP and PCI-X support into it especially if it is to be used as a low cost solution for IBM as well as Apple? Well time will tell as soon as IBM releases info on this controller. And Apple can always put firewire and ethernet on the peripheral controller with out giving up too much performance than if it was directly located on the main controller.
I don't think the GX bus is compatible with the bus used on the 970. ... Therefore a controller chip for the 970 would have to be built from the ground up. And why not include AGP and PCI-X support into it especially if it is to be used as a low cost solution for IBM as well as Apple?</strong>
Perhaps. But you have to keep in mind that IBM will be using the PPC 970 in $10K workstations and servers. They'll have PCI-X, but AGP is questionable let alone integration of high bandwidth I/O stuff (like 1394, Ethernet or ATA). But these machines will have different requirements than Apple's $1.5K to $3.5K machines.
<strong>Well time will tell as soon as IBM releases info on this controller. And Apple can always put firewire and ethernet on the peripheral controller with out giving up too much performance than if it was directly located on the main controller.</strong>
Yes, time will tell. It'll be a very long time.
<strong>What kind of memory would actually support a FSB of 450MHz?
Dual channel memory, Intel and nvidia are already make chipset that support this feature.
What kind of memory would actually support a FSB of 450MHz?
Rambus
Otherwise if they can have a 9:4 processor bus to memory bus ratio, then DDR 400 or DDR-2 will work, dual channel or otherwise. Or the easier solution is to only use processor clock rates that will work at lower integer multiples.
[ 10-19-2002: Message edited by: THT ]</p>
Rambus
Otherwise if they can have a 9:4 processor bus to memory bus ratio, then DDR 400 or DDR-2 will work, dual channel or otherwise. Or the easier solution is to only use processor clock rates that will work at lower integer multiples.
[ 10-19-2002: Message edited by: THT ][/QB]<hr></blockquote>
Rambus, of course, but Rambus is game over, i think
<img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
and
DDR-RAM (PC3200) is supporting a 400MHz FSB
What's about DDR-II
thx
<strong>
Not confused, you didn't read the post I replied to.</strong><hr></blockquote>
You said:
"Wow, following this i don't think we'll see duals acrosse the board then. A single 1.2ghjz GPUL is almos the same as a dual gig."
That is quite wrong, or a typo.
[ 10-19-2002: Message edited by: Eugene ]</p>
<a href="http://www.eweek.com/article2/0,3959,638499,00.asp"; target="_blank">www.eweek.com/article2/0,3959,638499,00.asp</a>
"The new 300 millimeter plant is in test production and on schedule. We have
orders that would fully load the facility well into midyear of 2003. From what
we can tell about our competition, we may be the only manufacturer capable of
130 and 90 nanometer technologies."
Is this the 970 plant? If so, anyone care to put this blurb in context of when the 970 is expected to show-up in a Mac?
<strong>I don't hold my breath on the new 970.
The 970 should available at the end of 2003, right?
At speeds up to 1.8GHz, right?
900MHz FSB, right?
2.9MIPS/ MHz, right?
The G4 should reach 1.8GHz at the end of 2003, right?
133MHz FSB today, at the end of the year maybe 800MHz (RIO), right?
2.3MIPS/ MHz, right?
Thats an advantage of 2.9/ 2.3 = 1.26 points
It doesn't impress me.
I hope i am wrong!
The only advantage of the 970 is 64bit and the SPEC benchmark!</strong><hr></blockquote>
Well, consider this then,
1. The PPC970 has not one, but TWO Altivec units. Altivec performance will be off the charts. Furthermore, the PPC970's altivec performance won't be choked by a slow bus....according to those who know about this sort of thing, the current G4 altivec quickly chokes because the MPX bus cannot feed it fast enough. So not only will a major bottleneck be removed, but an additional SIMD unit will be added. For altivec-optimized applications, the 970 will shame Pentium 4s, no doubt about it.
2. The typical trade-off for MHz is to lower the MIPS/MHz, partially by adding extra pipelines. Yet, the 970 clocks significantly faster than the G4, and STILL turns out more MIPS. Also, the 970 is going to debut at 1.8 GHz, probably on a 130 nm process. This leaves lots of room for scaling...IBM could migrate the 970 to 90 nm fairly quickly, and probably scale to 3 GHz within a year or so of the 970's introduction. In contrast, the G4 debuted at 400 MHz and is near the end of its life.
I think the PPC 970 is far, far superior to the G4, and it's going to be faster than x86 for some time after it's introduction. Maybe on par with x86 for FP caculations, but for any Altivec enabled tasks, the CPU is going to blow x86 away.
Apple was right all along about their focus on altivec....it's a secret weapon that has allowed Apple to keep up despite the disastrous G4, and it will enable Apple to smoke the competition when IBM finally begins fabbing the PPC.
After reading all the reports, press releases and following these threads, I keep coming back to a theory i had which suggests that the 970 is *not* the high (or ultra high-end) processor, but instead, the chip intended for use within *consumer* (and possibly low-high-end) machines. The differences simply being the GHz or number of processors used.
That of course leads us to the question about Apple's high-end machines... It wouldn't seem unreasonable that Apple would decide to make its high-end machines (ultra high-end machines) equipped with a genuine Power4 chip.
This would make Apple's consumer models running in parallel speed (maybe slightly higher) with that of x86's highest-end systems while putting Apple's high-end leaps and bounds ahead.
It seems like a natural fit if you ask me.
<strong>
That of course leads us to the question about Apple's high-end machines... It wouldn't seem unreasonable that Apple would decide to make its high-end machines (ultra high-end machines) equipped with a genuine Power4 chip.
This would make Apple's consumer models running in parallel speed (maybe slightly higher) with that of x86's highest-end systems while putting Apple's high-end leaps and bounds ahead.
It seems like a natural fit if you ask me.</strong><hr></blockquote>
Or maybe there's something like a 990 that we don't know about yet.... after all, those in the know have made it fairly clear that the 970 is not an only child.
<strong>
Well, consider this then,
1. The PPC970 has not one, but TWO Altivec units. Altivec performance will be off the charts. Furthermore, the PPC970's altivec performance won't be choked by a slow bus....according to those who know about this sort of thing, the current G4 altivec quickly chokes because the MPX bus cannot feed it fast enough. So not only will a major bottleneck be removed, but an additional SIMD unit will be added. For altivec-optimized applications, the 970 will shame Pentium 4s, no doubt about it.
2. The typical trade-off for MHz is to lower the MIPS/MHz, partially by adding extra pipelines. Yet, the 970 clocks significantly faster than the G4, and STILL turns out more MIPS. Also, the 970 is going to debut at 1.8 GHz, probably on a 130 nm process. This leaves lots of room for scaling...IBM could migrate the 970 to 90 nm fairly quickly, and probably scale to 3 GHz within a year or so of the 970's introduction. In contrast, the G4 debuted at 400 MHz and is near the end of its life.
</strong><hr></blockquote>
1. i thought the VMX unit on the 970 is divided into two seperate working units which work on the same instructions but split them up before processing (data handling, instruction processing or s.th.) however - i'm no expert on this but i think this is the way the G4/7400 managed altivec-processing too.... so these TWO altivec units are nothing more than ONE altivec unit...
2. the G4 will be available at speeds up to 1.8Ghz on a .13µ process in the first half of 2003...