While I don't doubt that the 980 is a little way off, I think it is well beyond the "gleam in IBM's eye" stage. I would suspect that the gen _after_ the 980 is at that stage, but has IBM not made a few pronouncments regarding the 980 already?
Even if it hasn't, since Jobs said "a few years", when talking about the 970, it seems likely that 980 work would also have started, at the very least.
While I don't doubt that the 980 is a little way off, I think it is well beyond the "gleam in IBM's eye" stage. I would suspect that the gen _after_ the 980 is at that stage, but has IBM not made a few pronouncments regarding the 980 already?
Even if it hasn't, since Jobs said "a few years", when talking about the 970, it seems likely that 980 work would also have started, at the very least.
IBM has said a few things about the POWER5, but I don't think they've ever said "PowerPC 980" in public. The existance of such a thing is all supposition and rumour. I do agree that the 980 will follow the POWER5 much more quickly than the 970 followed the POWER4. Whether it sees the light of day in 2004, we'll just have to wait and see.
1. I would speculate that a mature 130 nm PPC 970 could ship at 2.5 GHz.
2. I would speculate a 1st iteration 90 nm PPC 970 could ship at 3 GHz.
3. I would speculate that a 90 nm PPC 970 could eventually ship at 3.6 GHz.
Why?
1a. The 10 stage execution pipeline Athlon XP 3200+ (Barton) is a 130 nm processor that ships at 2.2 GHz.
1b. The PPC 970 is a 14 stage execution pipeline processor. It should clock higher than the Athlon on the same process and do it by a fair amount. A 2.5 GHz PPC 970 would have about 12% more clock rate than an Athlon 3200+, well within reason.
1c. It's probably fair to say that IBM's 300 mm 130 nm wafer fab is better than AMD's 200 mm 130 nm fab (which is Motorola influenced btw), and they can do it if they and Apple want to do it. It means 20% more power consumption per processor, but very doable.
2a. 45% increase in clock rate due to a die shrink is conservative for deeply pipelined processors. Intel Willamette went from 1.3 to 2 GHz while Northwood went from 2.2 GHz to 3.2 GHz, about 60% increase. AMD's Athlon K7 went from 500 to 850 MHz at 250 nm to 0.9 to 1.6 GHz at 180 nm, an amazing 100% increase from the top-end of process to the top end of another. It's only their 130 nm fab that's sort of gone awry.
2b. Given a 2.5 GHz 130 nm PPC 970, or a 2.4 or 2.2 GHz, 3 GHz is only an increase of 20%, 25%, or 36% above the 130 nm top-end. These percentages should be doable.
3a. Given a 2.5 GHz 130 nm PPC 970 top-end, a 3.6 GHz 90 nm top-end is only 45% higher.
3b. Given 2.0, 2.2, or 2.4 GHz 130 nm PPC 970 top-end, a 3.6 GHz 90 nm top-end is 80%, 64% or a 50% increase in clock rate. All doable given process improvements.
So, if Apple ships something like a 2.4 GHz 130 nm PPC 970 machine, a very doable 400 MHz I think, in Q1 04, I think a 90 nm 3 GHz machine is in the bag.
While I think a 2.4 GHz 130 nm 970 is possible, I want to comment that IBM has a different approach in processor design. The design is much more automated. AMD and Intel squeeze every little bit of Hertz out of their design. The changes in the designs from the first Athlon or P4 to the current ones are not really big. The Athlon 64 is changing a bit more the Prescott not quite that much. If both processor designs develop as slowly in the future as they did in the past, then IBM and therefore Apple will gain an advantage. Look at Power 5 and Power 6 in the works there are substantial changes in th design. So while the IBM chips may be clocked lower than Intel and AMD chips the integration will become much greater and widen the performance gap. (If Intel has nothing up their sleeves, but with Itanium I guess not)
While I think a 2.4 GHz 130 nm 970 is possible, I want to comment that IBM has a different approach in processor design. The design is much more automated. AMD and Intel squeeze every little bit of Hertz out of their design.
I'm talking more about tweaking the fab process to increase yields on higher MHz parts. Btw, AMD's 130 nm fab is probably underperforming. Maybe it is because of Motorola's influence, or AMD is the problem, but AMD should have been shipping a 2.4 GHz Barton by now.
Quote:
The changes in the designs from the first Athlon or P4 to the current ones are not really big.
In design yes. But in process no. The K7 microarchitecture went from 0.5 to 2.25 GHz across 3 process sizes. The P4 microarchitecture went from 1.4 to 3.2 GHz in one process change will likely go to 4.5 GHz with 90 nm. It's not about design so much, but probably 75% process tech and 25% circuit tweaks to increase clock rates on CPUs.
Quote:
If both processor designs develop as slowly in the future as they did in the past, then IBM and therefore Apple will gain an advantage. Look at Power 5 and Power 6 in the works there are substantial changes in th design. So while the IBM chips may be clocked lower than Intel and AMD chips the integration will become much greater and widen the performance gap. (If Intel has nothing up their sleeves, but with Itanium I guess not)
The design needs to be tweaked and improved yes, but without the process tech, the ability to move to 90 nm, 65 nm, and 45 nm in 18 to 24 month increments, the design won't help much. That's why Alpha and all the other RISC cpus couldn't keep up or compete against Intel.
If IBM can't ship 2.4 GHz out of 300 mm 130 nm fab, using a 14 stage execution pipeline processor, something is seriously lacking in their fab. It'll be a hot processor, probably around 60 Watts, but power consumption such as that is a new problem.
Comments
Even if it hasn't, since Jobs said "a few years", when talking about the 970, it seems likely that 980 work would also have started, at the very least.
Originally posted by jouster
While I don't doubt that the 980 is a little way off, I think it is well beyond the "gleam in IBM's eye" stage. I would suspect that the gen _after_ the 980 is at that stage, but has IBM not made a few pronouncments regarding the 980 already?
Even if it hasn't, since Jobs said "a few years", when talking about the 970, it seems likely that 980 work would also have started, at the very least.
IBM has said a few things about the POWER5, but I don't think they've ever said "PowerPC 980" in public. The existance of such a thing is all supposition and rumour. I do agree that the 980 will follow the POWER5 much more quickly than the 970 followed the POWER4. Whether it sees the light of day in 2004, we'll just have to wait and see.
Originally posted by THT
1. I would speculate that a mature 130 nm PPC 970 could ship at 2.5 GHz.
2. I would speculate a 1st iteration 90 nm PPC 970 could ship at 3 GHz.
3. I would speculate that a 90 nm PPC 970 could eventually ship at 3.6 GHz.
Why?
1a. The 10 stage execution pipeline Athlon XP 3200+ (Barton) is a 130 nm processor that ships at 2.2 GHz.
1b. The PPC 970 is a 14 stage execution pipeline processor. It should clock higher than the Athlon on the same process and do it by a fair amount. A 2.5 GHz PPC 970 would have about 12% more clock rate than an Athlon 3200+, well within reason.
1c. It's probably fair to say that IBM's 300 mm 130 nm wafer fab is better than AMD's 200 mm 130 nm fab (which is Motorola influenced btw), and they can do it if they and Apple want to do it. It means 20% more power consumption per processor, but very doable.
2a. 45% increase in clock rate due to a die shrink is conservative for deeply pipelined processors. Intel Willamette went from 1.3 to 2 GHz while Northwood went from 2.2 GHz to 3.2 GHz, about 60% increase. AMD's Athlon K7 went from 500 to 850 MHz at 250 nm to 0.9 to 1.6 GHz at 180 nm, an amazing 100% increase from the top-end of process to the top end of another. It's only their 130 nm fab that's sort of gone awry.
2b. Given a 2.5 GHz 130 nm PPC 970, or a 2.4 or 2.2 GHz, 3 GHz is only an increase of 20%, 25%, or 36% above the 130 nm top-end. These percentages should be doable.
3a. Given a 2.5 GHz 130 nm PPC 970 top-end, a 3.6 GHz 90 nm top-end is only 45% higher.
3b. Given 2.0, 2.2, or 2.4 GHz 130 nm PPC 970 top-end, a 3.6 GHz 90 nm top-end is 80%, 64% or a 50% increase in clock rate. All doable given process improvements.
So, if Apple ships something like a 2.4 GHz 130 nm PPC 970 machine, a very doable 400 MHz I think, in Q1 04, I think a 90 nm 3 GHz machine is in the bag.
While I think a 2.4 GHz 130 nm 970 is possible, I want to comment that IBM has a different approach in processor design. The design is much more automated. AMD and Intel squeeze every little bit of Hertz out of their design. The changes in the designs from the first Athlon or P4 to the current ones are not really big. The Athlon 64 is changing a bit more the Prescott not quite that much. If both processor designs develop as slowly in the future as they did in the past, then IBM and therefore Apple will gain an advantage. Look at Power 5 and Power 6 in the works there are substantial changes in th design. So while the IBM chips may be clocked lower than Intel and AMD chips the integration will become much greater and widen the performance gap. (If Intel has nothing up their sleeves, but with Itanium I guess not)
Originally posted by stjg
While I think a 2.4 GHz 130 nm 970 is possible, I want to comment that IBM has a different approach in processor design. The design is much more automated. AMD and Intel squeeze every little bit of Hertz out of their design.
I'm talking more about tweaking the fab process to increase yields on higher MHz parts. Btw, AMD's 130 nm fab is probably underperforming. Maybe it is because of Motorola's influence, or AMD is the problem, but AMD should have been shipping a 2.4 GHz Barton by now.
The changes in the designs from the first Athlon or P4 to the current ones are not really big.
In design yes. But in process no. The K7 microarchitecture went from 0.5 to 2.25 GHz across 3 process sizes. The P4 microarchitecture went from 1.4 to 3.2 GHz in one process change will likely go to 4.5 GHz with 90 nm. It's not about design so much, but probably 75% process tech and 25% circuit tweaks to increase clock rates on CPUs.
If both processor designs develop as slowly in the future as they did in the past, then IBM and therefore Apple will gain an advantage. Look at Power 5 and Power 6 in the works there are substantial changes in th design. So while the IBM chips may be clocked lower than Intel and AMD chips the integration will become much greater and widen the performance gap. (If Intel has nothing up their sleeves, but with Itanium I guess not)
The design needs to be tweaked and improved yes, but without the process tech, the ability to move to 90 nm, 65 nm, and 45 nm in 18 to 24 month increments, the design won't help much. That's why Alpha and all the other RISC cpus couldn't keep up or compete against Intel.
If IBM can't ship 2.4 GHz out of 300 mm 130 nm fab, using a 14 stage execution pipeline processor, something is seriously lacking in their fab. It'll be a hot processor, probably around 60 Watts, but power consumption such as that is a new problem.