OK, I think a simple mathematics session is in order here.
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
1000 + (1000/100*30) = 1300
Yeah, that's 1300 Mhz for ya, 1.3 Ghz. With the most optimistic estimates. And you can be sure that the top-end dual in that scenario would probably come with 1.1 Ghz or something.
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.
I've read in a few places that IBM has taken over fabbing for Moto.
The one bad thing I read was this-
[quote] The manufacturing yields for the G5 via Motorola are not high enough at this time to support Apple's anticipated requirements. There are also some problems with both the g4 and g5 cache aritectures when clocked at over 900mhz. These are being worked.
Due to the low yields, Apple is in a position to negotiate with IBM to co-supply G5s.
So, in short .. you will not see an Apple product shipping with a G5 processor within the next eight to ten months. It is just not possible.<hr></blockquote>
So it looks like Moto may have screwed us again. They couldn't get high enough yield and now IBM may have to step in and take over. That would mean a delay in us seeing G5s.
The other one I read was something like "for the G5 don't look at Moto, look at IBM" or something to that effect.
<strong>OK, I think a simple mathematics session is in order here.
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
[1.3 Ghz.]
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.</strong><hr></blockquote>
Nothing prevents the G4 from migrating to .13 micron as well. In fact, it might just do that. That's another 30% boost up in the neighborhood of 1.7-1.8GHz.
In fact, that's probably what that Motorola exec meant when he said "there's a lot of life left in the G4."
[quote]<strong>We need a G5 , we really really do...</strong><hr></blockquote>
It certainly wouldn't hurt.
[quote]Originally posted by KidRed:
<strong>I've read in a few places that IBM has taken over fabbing for Moto.</strong><hr></blockquote>
It would be more accurate to say that Motorola is now subcontracting chip fabrication out to third parties. IBM is one, but not the only one - and in fact probably not the main one. It is the world's most expensive foundry, after all. There are companies in southeast Asia that do nothing but fab chips for other companies. I imagine Mot's leaning on those. There's no information I can find on what's being farmed out where. Given that Mot just reaped the fruits of a collaboration with AMD, I'd be surprised if they're giving up their new, advanced facilities. My guess is that they're contracting out embedded chips that run on mature processes so that they can shed some of their older plants, and only using IBM for the stuff that has to be produced on a brand-new process.
<strong>OK, I think a simple mathematics session is in order here.
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
1000 + (1000/100*30) = 1300
Yeah, that's 1300 Mhz for ya, 1.3 Ghz. With the most optimistic estimates. And you can be sure that the top-end dual in that scenario would probably come with 1.1 Ghz or something.
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.
We need a G5 , we really really do...</strong><hr></blockquote>
Silly. This is on the basis that there is no change to the basic design of the chip. What Motorola have said all along is the the Apollo is a redesign of the G4 family, so all bets on relative clock speed are off. The fact that a few chips are capable of operating at much higher speeds is indicative of concentrated choke point(s) rather than a major flaw in the G4 architecture limiting the speed.
Rumours are that the speed limiting in the G4 is due to flaws in the cache architectures, and it is possible that redesigning this could lead to current G4s on current processes considerably increasing in speed. That won't happen, of course, for cost reasons, but I'm certain that Motorola have worked extremely hard on identifying the flaws limiting scaling (and eliminating them from the Apollo design), as it is important to their sales of G4s for embedded use that they can scale the chip better.
Whilst I agree that the Athlon has a better FPU than the G4, I would'nt agree that it outperforms the G4 clock for clock on integer processing, quite the opposite, and the vector unit of the G4 is way ahead of even the SSE on the new Athlon XPs.
Silly. This is on the basis that there is no change to the basic design of the chip. What Motorola have said all along is the the Apollo is a redesign of the G4 family, so all bets on relative clock speed are off. </strong><hr></blockquote>
When did Motorola say that? Can you provide a link? I hope you're right. But everything I've read about the Apollo since 1999 always said it would be a G4+ on SOI. We already have the G4+ aka the MPC 7450.
Here's what Motorola said about the Apollo at the Microprocessor Forum in Fall 2000.
[quote]<strong> The Apollo G4 will use IBM's SOI (Silicon-on-Insulator) technology and be based on 0.18 micron process, Bearden said, this and a deeper pipeline will permit the higher clock speeds. He said that the SOI method added 22 per cent performance improvement over bulk CMOS alone. Apollo will feature a wider, 256-bit path between on-chip caches, on-die L2 cache of 256KB and up to 2MB of external L3 cache. Bearden cited power consumption of under 10W at 666MHz and under 23W at super-1GHz frequencies. </strong> <hr></blockquote>
The only difference I can see there is a wider internal bus to the cache. The longer/new pipeline they're talking about is the pipeline of the G4+, ie it's longer than the MPC 7400 which was the version of the G4 shipping at the time...
one potential positive side-effect to motorola's recent problems...if they can't produce apollo in mass, it may give apple the leverage to push for motorola to sell altivec and let IBM start fabrication. i have no idea how long it would take to start fabrication, but seeing as IBM already produces G4's with SOI....
When did Motorola say that? Can you provide a link? I hope you're right. But everything I've read about the Apollo since 1999 always said it would be a G4+ on SOI. We already have the G4+ aka the MPC 7450.
Here's what Motorola said about the Apollo at the Microprocessor Forum in Fall 2000.
The only difference I can see there is a wider internal bus to the cache. The longer/new pipeline they're talking about is the pipeline of the G4+, ie it's longer than the MPC 7400 which was the version of the G4 shipping at the time...</strong><hr></blockquote>
Probably the wrong word, they said they had relaid-out most of the circuits for the new SOI process, but this would certainly involve an analysis of time-critical paths.
The main difference is in the memory architecture, the 256 bit paths imply a fairly major redesign of the cache architecture, and as I said, I believe the yield limiting step is in the cache system.
<strong>When was the last time that Apple came out with a new machine where the 'high end' machine was all of a sudden completely out of the picture?</strong><hr></blockquote>
68040 -> PPC 601 pretty much put the Quadras to rest (Apple's marketing nomenclature aside).
What is more important to Apple, staying competitive with WIntel or not jumping the clock rate too much? I say they'll jump the clock rate as much as they can, and if they can jump processor generations they will. SOI should allow up to a 30% speed hike, and the rumours are talking about quite a bit more.
My point is that Apple has NEVER made a jump like this. Even when the Power Mac G4s came out in August of 1999, shipping delays were so bad that they were still selling the Power Mac G3 towers for a while, and even then, the clock speeds were still similar.
I'd like to believe that the Power Macs will clock high just like everyone else, but if we're being realistic, it's NEVER happened before. Why should we expect this time to be any different?
<strong>OK, I think a simple mathematics session is in order here.
(...)
1000 + (1000/100*30) = 1300
</strong><hr></blockquote>
I don't think this "add x% for SOI, y% for 130nm, ..." approach will really work. Just look at Intel and AMD and how far they have been able to scale the 180nm-Version of the Athlon without ever using SOI, longer pipes or similar gimmicks.
Other chip companies have their high-powered desktop processors running at double the clock rate -- this has never happened before either. Apple has been shackled by the scaling problems with the G4 and when those problems are removed there is no reason why it can't suddenly be competitive.
When the MacII arrived it was 4 times faster than the Mac128K. When the Quadra arrived, it was easily 50% faster than the IIfx. When the PPC601 arrived, it smoked the Quadra once you had native software. Generational changes frequently cause major performance leaps ... they introduce new technologies that make things go faster, frequently much faster.
Predicting that something isn't going to happen because it hasn't before, or because you haven't seen it happen yourself, is a fallacy. Reminds me of people saying we'd never need more than 640K of RAM, or that we'd never break the 100MHz "barrier", or the 1GHz "barrier".
When the MacII arrived it was 4 times faster than the Mac128K. When the Quadra arrived, it was easily 50% faster than the IIfx. When the PPC601 arrived, it smoked the Quadra once you had native software.
</strong><hr></blockquote>
I think he was specifically talking about the actual MHz number, which is undoubtedly a focal point of customer perception and marketing.
<strong>My point is that Apple has NEVER made a jump like this. Even when the Power Mac G4s came out in August of 1999, shipping delays were so bad that they were still selling the Power Mac G3 towers for a while, and even then, the clock speeds were still similar.
I'd like to believe that the Power Macs will clock high just like everyone else, but if we're being realistic, it's NEVER happened before. Why should we expect this time to be any different?</strong><hr></blockquote>
Well, just looking at the numbers on Apple's website, the last round of 68040s were in the Quadras and Performas. The Quadra 840AV had a 40 MHz 68040, while the Perfroma 640CD had a 68LC040 (no FPU) at 66 MHz. They were replaced by the PowerMacs 6100 - 8100, with PPC601s clocked from 60 -80 MHz, with system bus speeds about the same.
The top-of-the line 68040 was really the Quadra - the higher clock speed on the Performa was negated and then some by lack of an FPU. According to Macspeedzone, the jump from the Quadra 840AV to the PM 8100/80 was a tripling of overall speed (relative score of 52 to 154). The 60 MHz 6100 was twice as fast. So a doubling or tripling of performance from one generation to the next is not unprecedented. I would be very happy if Apple were to pull that again.
I'm wondering if this guy is someone we have heard from before under a different name? I'm also wondering if he will be around anymore if a G5 does get released soon. I mean, why the campaign?
Ok, lets get one thing straight, the iMacs are now using G4s, HELLO?
The apollos will be used in the iMacs because introducing them at 700Mhz was needed. But Apple won't gain many sales if they kepp at the smae clock speed.
The G5s have been worked on for 2 years. Don't you think Apple would be more secretive about G5s than n iMac which just uses the same old tech? I think so. Maya was not built to run on G4s, but G5s.
My father works for McCann Erikson advertising agency, and macs are used all the time. Everyone who uses a computer uses a mac. Thats right, even the secretaries Windows boxes were thrown out and replaced by iMacs.
The G5 will e a lot faster than a G4, I hink it will have higher performance then you doubters expect. The PPC is getting far more backing now than it has before, and it has renewed motorolas plans. They were even thinking of ditching the G5, at one point, now they're reseaching G7s.
I expect 1.4Ghz G5s low-end, and 2Ghz G5s high end for $3499
Why is the Apollo taking so long if it's simply a SOI G4? Not even a process change. Bear with me for a moment. If Motorola is indeed moving the G4 floor plan around for SOI perhaps they are being more ambitious than we think. They could be in fact lengthening the pipeline to 10 stages like the Athlon is now, using a double or quad pumped FSB, and adding more L2 cache getting rid of L3 tags all together. Keep it a 32bit processor and save the G5 for future use in MWSF-2003. I have no evidence to support this argument but it is a likely senario.
Comments
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
1000 + (1000/100*30) = 1300
Yeah, that's 1300 Mhz for ya, 1.3 Ghz. With the most optimistic estimates. And you can be sure that the top-end dual in that scenario would probably come with 1.1 Ghz or something.
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.
We need a G5 , we really really do...
The one bad thing I read was this-
[quote] The manufacturing yields for the G5 via Motorola are not high enough at this time to support Apple's anticipated requirements. There are also some problems with both the g4 and g5 cache aritectures when clocked at over 900mhz. These are being worked.
Due to the low yields, Apple is in a position to negotiate with IBM to co-supply G5s.
So, in short .. you will not see an Apple product shipping with a G5 processor within the next eight to ten months. It is just not possible.<hr></blockquote>
So it looks like Moto may have screwed us again. They couldn't get high enough yield and now IBM may have to step in and take over. That would mean a delay in us seeing G5s.
The other one I read was something like "for the G5 don't look at Moto, look at IBM" or something to that effect.
<strong>OK, I think a simple mathematics session is in order here.
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
[1.3 Ghz.]
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.</strong><hr></blockquote>
Nothing prevents the G4 from migrating to .13 micron as well. In fact, it might just do that. That's another 30% boost up in the neighborhood of 1.7-1.8GHz.
In fact, that's probably what that Motorola exec meant when he said "there's a lot of life left in the G4."
[quote]<strong>We need a G5 , we really really do...</strong><hr></blockquote>
It certainly wouldn't hurt.
[quote]Originally posted by KidRed:
<strong>I've read in a few places that IBM has taken over fabbing for Moto.</strong><hr></blockquote>
It would be more accurate to say that Motorola is now subcontracting chip fabrication out to third parties. IBM is one, but not the only one - and in fact probably not the main one. It is the world's most expensive foundry, after all. There are companies in southeast Asia that do nothing but fab chips for other companies. I imagine Mot's leaning on those. There's no information I can find on what's being farmed out where. Given that Mot just reaped the fruits of a collaboration with AMD, I'd be surprised if they're giving up their new, advanced facilities. My guess is that they're contracting out embedded chips that run on mature processes so that they can shed some of their older plants, and only using IBM for the stuff that has to be produced on a brand-new process.
<strong>OK, I think a simple mathematics session is in order here.
Latest word from Mot was that the Apollo would not be produced on a 13 micron SOI process as we'd all been hoping but on 18 micron instead.
Currently the G4 tops out at 867. It has been overclocked to 1000 Mhz by some. Let's assume that Motorola has gotten better at it too, and they're able to churn out 1000 Mhz 7450s routinely by now.
According to IBM's best marketing estimates SOI enables a 20 to 30 % increase in clock rate on the same process.
What does all this mean?
1000 + (1000/100*30) = 1300
Yeah, that's 1300 Mhz for ya, 1.3 Ghz. With the most optimistic estimates. And you can be sure that the top-end dual in that scenario would probably come with 1.1 Ghz or something.
The Athlon XP performs better than the G4, clock for clock especially with FP operations. It runs at 1.67 Ghz now, and will soon go SOI at 13 micron.
We need a G5 , we really really do...</strong><hr></blockquote>
Silly. This is on the basis that there is no change to the basic design of the chip. What Motorola have said all along is the the Apollo is a redesign of the G4 family, so all bets on relative clock speed are off. The fact that a few chips are capable of operating at much higher speeds is indicative of concentrated choke point(s) rather than a major flaw in the G4 architecture limiting the speed.
Rumours are that the speed limiting in the G4 is due to flaws in the cache architectures, and it is possible that redesigning this could lead to current G4s on current processes considerably increasing in speed. That won't happen, of course, for cost reasons, but I'm certain that Motorola have worked extremely hard on identifying the flaws limiting scaling (and eliminating them from the Apollo design), as it is important to their sales of G4s for embedded use that they can scale the chip better.
Whilst I agree that the Athlon has a better FPU than the G4, I would'nt agree that it outperforms the G4 clock for clock on integer processing, quite the opposite, and the vector unit of the G4 is way ahead of even the SSE on the new Athlon XPs.
Michael
<strong>
Silly. This is on the basis that there is no change to the basic design of the chip. What Motorola have said all along is the the Apollo is a redesign of the G4 family, so all bets on relative clock speed are off. </strong><hr></blockquote>
When did Motorola say that? Can you provide a link? I hope you're right. But everything I've read about the Apollo since 1999 always said it would be a G4+ on SOI. We already have the G4+ aka the MPC 7450.
Here's what Motorola said about the Apollo at the Microprocessor Forum in Fall 2000.
[quote]<strong> The Apollo G4 will use IBM's SOI (Silicon-on-Insulator) technology and be based on 0.18 micron process, Bearden said, this and a deeper pipeline will permit the higher clock speeds. He said that the SOI method added 22 per cent performance improvement over bulk CMOS alone. Apollo will feature a wider, 256-bit path between on-chip caches, on-die L2 cache of 256KB and up to 2MB of external L3 cache. Bearden cited power consumption of under 10W at 666MHz and under 23W at super-1GHz frequencies. </strong> <hr></blockquote>
The only difference I can see there is a wider internal bus to the cache. The longer/new pipeline they're talking about is the pipeline of the G4+, ie it's longer than the MPC 7400 which was the version of the G4 shipping at the time...
You're talking about almost doubling the clock speed of the high end G4.
According to the theories on this board, the configurations of the new G4s will be:
Dual Processor G4/1 GHz
Dual Processor G4/1.2 GHz
Dual Processor G4/1.4 GHz
That's quite a substantial increase from:
G4/733 MHz
G4/867 MHz
Dual Processor G4/800 MHz
When was the last time that Apple came out with a new machine where the 'high end' machine was all of a sudden completely out of the picture?
This just seems a little optomistic to me, that's all. Like I said, I'd love to see the new G4s at 1.4 GHz, but I just can't see it happening.
<strong>
When did Motorola say that? Can you provide a link? I hope you're right. But everything I've read about the Apollo since 1999 always said it would be a G4+ on SOI. We already have the G4+ aka the MPC 7450.
Here's what Motorola said about the Apollo at the Microprocessor Forum in Fall 2000.
The only difference I can see there is a wider internal bus to the cache. The longer/new pipeline they're talking about is the pipeline of the G4+, ie it's longer than the MPC 7400 which was the version of the G4 shipping at the time...</strong><hr></blockquote>
Probably the wrong word, they said they had relaid-out most of the circuits for the new SOI process, but this would certainly involve an analysis of time-critical paths.
The main difference is in the memory architecture, the 256 bit paths imply a fairly major redesign of the cache architecture, and as I said, I believe the yield limiting step is in the cache system.
Michael
<strong>When was the last time that Apple came out with a new machine where the 'high end' machine was all of a sudden completely out of the picture?</strong><hr></blockquote>
68040 -> PPC 601 pretty much put the Quadras to rest (Apple's marketing nomenclature aside).
What is more important to Apple, staying competitive with WIntel or not jumping the clock rate too much? I say they'll jump the clock rate as much as they can, and if they can jump processor generations they will. SOI should allow up to a 30% speed hike, and the rumours are talking about quite a bit more.
I'd like to believe that the Power Macs will clock high just like everyone else, but if we're being realistic, it's NEVER happened before. Why should we expect this time to be any different?
<strong>OK, I think a simple mathematics session is in order here.
(...)
1000 + (1000/100*30) = 1300
</strong><hr></blockquote>
I don't think this "add x% for SOI, y% for 130nm, ..." approach will really work. Just look at Intel and AMD and how far they have been able to scale the 180nm-Version of the Athlon without ever using SOI, longer pipes or similar gimmicks.
Bye,
RazzFazz
<strong>
The only difference I can see there is a wider internal bus to the cache.
</strong><hr></blockquote>
In fact, according to some datasheets on Motos website, the 744x / 745x already have that 256 bit connection to the L2 cache.
(EDIT) Actually, the reference is <a href="http://e-www.motorola.com/webapp/sps/site/prod_summary.jsp?code=MPC7451&nodeId=03M9430304504 67M98653" target="_blank">here</a>, in the "Cache and MMU support" section.(/EDIT)
Bye,
RazzFazz
[ 01-15-2002: Message edited by: RazzFazz ]</p>
When the MacII arrived it was 4 times faster than the Mac128K. When the Quadra arrived, it was easily 50% faster than the IIfx. When the PPC601 arrived, it smoked the Quadra once you had native software. Generational changes frequently cause major performance leaps ... they introduce new technologies that make things go faster, frequently much faster.
Predicting that something isn't going to happen because it hasn't before, or because you haven't seen it happen yourself, is a fallacy. Reminds me of people saying we'd never need more than 640K of RAM, or that we'd never break the 100MHz "barrier", or the 1GHz "barrier".
<strong>
When the MacII arrived it was 4 times faster than the Mac128K. When the Quadra arrived, it was easily 50% faster than the IIfx. When the PPC601 arrived, it smoked the Quadra once you had native software.
</strong><hr></blockquote>
I think he was specifically talking about the actual MHz number, which is undoubtedly a focal point of customer perception and marketing.
Bye,
RazzFazz
<strong>My point is that Apple has NEVER made a jump like this. Even when the Power Mac G4s came out in August of 1999, shipping delays were so bad that they were still selling the Power Mac G3 towers for a while, and even then, the clock speeds were still similar.
I'd like to believe that the Power Macs will clock high just like everyone else, but if we're being realistic, it's NEVER happened before. Why should we expect this time to be any different?</strong><hr></blockquote>
Well, just looking at the numbers on Apple's website, the last round of 68040s were in the Quadras and Performas. The Quadra 840AV had a 40 MHz 68040, while the Perfroma 640CD had a 68LC040 (no FPU) at 66 MHz. They were replaced by the PowerMacs 6100 - 8100, with PPC601s clocked from 60 -80 MHz, with system bus speeds about the same.
The top-of-the line 68040 was really the Quadra - the higher clock speed on the Performa was negated and then some by lack of an FPU. According to Macspeedzone, the jump from the Quadra 840AV to the PM 8100/80 was a tripling of overall speed (relative score of 52 to 154). The 60 MHz 6100 was twice as fast. So a doubling or tripling of performance from one generation to the next is not unprecedented. I would be very happy if Apple were to pull that again.
The apollos will be used in the iMacs because introducing them at 700Mhz was needed. But Apple won't gain many sales if they kepp at the smae clock speed.
The G5s have been worked on for 2 years. Don't you think Apple would be more secretive about G5s than n iMac which just uses the same old tech? I think so. Maya was not built to run on G4s, but G5s.
My father works for McCann Erikson advertising agency, and macs are used all the time. Everyone who uses a computer uses a mac. Thats right, even the secretaries Windows boxes were thrown out and replaced by iMacs.
The G5 will e a lot faster than a G4, I hink it will have higher performance then you doubters expect. The PPC is getting far more backing now than it has before, and it has renewed motorolas plans. They were even thinking of ditching the G5, at one point, now they're reseaching G7s.
I expect 1.4Ghz G5s low-end, and 2Ghz G5s high end for $3499
<strong>Well SCSI drives will usually always outperform an ATA drive whether its ATA/66 or 100 or whatever.
anyone know why seek times on ATA drives are so slow compared to SCSI? </strong><hr></blockquote>
Because they are cheap drives?
[quote]
<strong>and is it possible to make a 10,000 RPM ATA drive?</strong><hr></blockquote>
It's possible, but the ATA market won't bear those prices, whereas the SCSI market will.