That isn't true at all. People keep making this statement with the only basis in IBM's microprocessor forum literature. This type of literature often conflates reality, with best case scenario testing, with projections, and a healthy dose of optimism -- a sort of advertorial/infomercial session really. According to this type of report IBM should have had 2Ghz G3's, or moto's G4 would already be at 1.8Ghz with a DDR FSB. It's rather obvious that they don't represent reality.
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
Well said.
The G4 is a much simpler core than the G5 becuase it has shorter pipelines, far less out-of-order execution capability, is only 32-bit, and a high performance bus interface. As a result of the first two items the G4 has much smaller execution resources (registers, buffers, etc). These differences give the G5 the ability to scale to much higher clock rates and perform well despite the huge latencies on data from main memory, even in a multi-processor system. At G4 clock rates, however, the G5 will be slower than the G4 on many many tasks. Basically, if you don't run it at high speeds, it loses many of its advantages over the G4 because its architecture is designed from the ground up to run at high clock rates.
I'm not prepared to scoff at Freescale hitting 90nm this year because its not their fabs in question -- ST Microelectronics and Philips will very likely benefit them tremendously. If Freescale can focus on the processor / SoC desgin (which has always been their strength) then we may see a G4 that addresses the big weaknesses of past G4s. All it would take is an on-chip memory controller to eliminate the MPX bus, and an RIO or HT interface to the I/O controller. 90 nm would probably push the clock rate up to 1.8-2 GHz. Such a chip could blow the stuffing out of the Pentium-M, just like the G4 used to best the PentiumPro/II/III before Intel pushed its bus technology past what MPX could provide. And do this at a lower power dissapation point than the G5 would achieve with the same performance.
Apple does not need a single processor line -- they need 2 or 3. The iBook, iMac, PB, PowerMac, and Xserve cover a broad spectrum of processor requirements and the biggest problem with the G4 for all those years was that Apple only had the G4 to cover the whole gamut (there being not much to distinguish the G3 from the G4, aside from AltiVec the lack of which caused more problems). My hope is that now they can have 2 G5-class processors (970 & 975), and a heavily updated G4.
This isn't the old Motorola. It's not even just Motorola. You can bet against Philips Semiconductor and STM Micro (Europe's largest contract fab) if you want to.
Crolles is online now, making product at 90nm now. They're (wisely) starting out with simple stuff like memory before tackling something as complex as a CPU. So far, the whole Crolles partnership is on schedule.
Get used to it. Frankly, while I understand the impulse to be skeptical of Motorola, I have to wonder about the hostility toward the prospect of a valuable second supplier. Do you want Apple to have all their eggs in one basket again?
Oh, and check Motorola's last quarter results, will you? You might find them interesting.
That isn't true at all. People keep making this statement with the only basis in IBM's microprocessor forum literature. This type of literature often conflates reality, with best case scenario testing, with projections, and a healthy dose of optimism -- a sort of advertorial/infomercial session really. According to this type of report IBM should have had 2Ghz G3's, or moto's G4 would already be at 1.8Ghz with a DDR FSB. It's rather obvious that they don't represent reality.
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
That is all probably true, yet, if I remember correctly, the current G4 has more transistors on die than the G5. I believe the #'s are something like 52 million for the G5 and 58 million for the G4. So if my memory is good, , the G5 may in fact use less power than the G4.
That is all probably true, yet, if I remember correctly, the current G4 has more transistors on die than the G5. I believe the #'s are something like 52 million for the G5 and 58 million for the G4. So if my memory is good, , the G5 may in fact use less power than the G4.
I don't remember exactly, but I seem to remember that from somewhere.
I don't remember exactly, but I seem to remember that from somewhere.
The information is available in product information pdf's at both IBM and Motorola, but I'm too lazy to look it up. Other things affect power besided just transistor count, like clock speed, voltage, leakage, etc. I only mentioned it in passing to kind of re-inforce the distinct possibility that the G5 does in fact dissipate less heat.
The information is available in product information pdf's at both IBM and Motorola, but I'm too lazy to look it up. Other things affect power besided just transistor count, like clock speed, voltage, leakage, etc. I only mentioned it in passing to kind of re-inforce the distinct possibility that the G5 does in fact dissipate less heat.
I thought clock for clock that, that was already known.
I thought clock for clock that, that was already known.
I thought so too, however, it seems that some people are contending that the typical and maximum usages as measured/reported by Motorola and IBM are not comparable, which may very well be the case, who knows. That, and people keep "banging the pot", saying that the G5 uses too much power to be used in a laptop, even though the numbers indicate otherwise.
Regarding the concern for a powerbook upgrade I see little need for it at the moment. Perhaps marketing concerns will push apple to develope a G5 powerbook quciker, but the idea that PC laptops are killing powerbooks is pretty silly. I can understand if you'd say pc's are punchier (quicker to exacute simpler commande) but in terms of number crunching the mac have it in spades. I own a G3 pismo 500 and a PIV 2.6 GHZ vaio, each has a gig of ram and I use both interchangeably depending on my need and the pismo at 1/4 the clock speed hold up very well to heavy 3d and video rendering in FCP, lightwave and after effects just as well if not better than the VAIO. Truthfully the pc is junk and I wouldn't own it if I didn't need to, people need to look past the specs and get themselves some real world "practice" on these machines, for thier purpose before they make vague blanket statements like pc laptops are better than apple's. Silly!Regarding the concern for a Powerbook upgrade I see little need for it at the moment. Perhaps marketing concerns will push apple to develop a G5 Powerbook quicker, but the idea that PC laptops are killing Powerbooks is pretty silly. I can understand if you'd say pc's are punchier (quicker to execute simpler commands) but in terms of number crunching the Mac have it in spades. I own a G3 Pismo 500 and a PIV 2.6 GHZ Vaio, each has a gig of ram and I use both interchangeably depending on my need and the Pismo at 1/4 the clock speed hold up very well to heavy 3d and video rendering in FCP, Lightwave and after effects just as well if not better than the VAIO. Truthfully the pc is junk and I wouldn't own it if I didn't need to, people need to look past the specs and get themselves some real world "practice" on these machines, for their purpose before they make vague blanket statements like pc laptops are better than apple's. Silly!
Regarding the concern for a powerbook upgrade I see little need for it at the moment. Perhaps marketing concerns will push apple to develope a G5 powerbook quciker, but the idea that PC laptops are killing powerbooks is pretty silly. I can understand if you'd say pc's are punchier (quicker to exacute simpler commande) but in terms of number crunching the mac have it in spades. I own a G3 pismo 500 and a PIV 2.6 GHZ vaio, each has a gig of ram and I use both interchangeably depending on my need and the pismo at 1/4 the clock speed hold up very well to heavy 3d and video rendering in FCP, lightwave and after effects just as well if not better than the VAIO. Truthfully the pc is junk and I wouldn't own it if I didn't need to, people need to look past the specs and get themselves some real world "practice" on these machines, for thier purpose before they make vague blanket statements like pc laptops are better than apple's. Silly!
The current PowerBooks may or may not hold up to the new 1.7Ghz centrinos for no benchmarks have been released yet. If I remember correctly the last revision fell behind in many tasks, although not by all that much. The G4 in its current form is not going to be able to keep up with the centrino. It needs a faster Bus, a smaller die shrink, and a 1MB L2 cache. If Freescale/motorola can pull this off the G4 should be able to surpass the centrino. The problem currently stems from the slow bus which substantially affects the G4s ability to process large amounts of data.
The G3 is a good chip, but I don't believe for a second that it comes anywhere close to the 2.6Ghz P4M. The P4M may be a lot slower than its desktop sibling but it is not that slow.
The PowerBooks do have one large advantage over most PC laptops and thats the high end graphic cards apple puts into them. They should really help any video benchmarks out.
The current PowerBooks may or may not hold up to the new 1.7Ghz centrinos for no benchmarks have been released yet. If I remember correctly the last revision fell behind in many tasks, although not by all that much. The G4 in its current form is not going to be able to keep up with the centrino. It needs a faster Bus, a smaller die shrink, and a 1MB L2 cache. If Freescale/motorola can pull this off the G4 should be able to surpass the centrino. The problem currently stems from the slow bus which substantially affects the G4s ability to process large amounts of data.
The G3 is a good chip, but I don't believe for a second that it comes anywhere close to the 2.6Ghz P4M. The P4M may be a lot slower than its desktop sibling but it is not that slow.
And its getting WORSE!, 2GHz Centrino , and its shipping in May!.
So hopefully, the next bump in Jan is a bit more than 100MHz? Oh no let me guess, we dont need faster CPU's because its all about the OS???
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
Not only that, but some recent real data **confirm** that this G5 would not go in a Powerbook. We will see perhaps one day a G5 Powerbook, but I seriously doubt it will have the G5 we all know today.
And I explain. xlr8yourmac posted yesterday some comments of a XServe G5 user. Let's see:
Quote:
Xserve G5 owner comments on CPU Power Usage/Heat
- A reader with a new Xserve G5 wrote regarding the lower-power/wattage version (compared to the original G5 towers) of the IBM PowerPC970 it uses
"Re: observations G5 970FX in new XServe
hi mike,
we received our xserve G5 a few days ago. it is the model with one 2.0GHz processor.
theserver-monitor application is pretty much verbose: Vcore of this G5 is 1.32 Volts. when idling it draws about 15 watts (or 11.8amps) (15W/1.32V = appx 11.36A), under 100% cpu-load it draws 44.5 watts or 33.5 amps. (remember this is at 1.32V and using 44.5W/1.32V = appx 33.7A)
the maximum core-temp from the cpu under 100% load was 75 degrees celsius. ambient temp. was at about 25 degrees celsius.
best regards,
michael b.
P4 northwood draws about 18 amps when idle. "
Now, let someone explain to me how do you put a 44.5 W chip (processor alone--granted, at 100% CPU load) in a thin notebook like the Powerbook.
And not only that, as I posted in another thread, it might not be worth slowing it down to get the G5 into an acceptable power consumption range:
The G5 relies on high Mhz and a huge FSB for it's speed. Slowing it down to 1.3-1.6Ghz and limiting it to a DDR333/400 FSB may render it little better than a G4 in it's current state.
Basically, cooling it down enough for laptop use, may render it too slow. It isn't looking like a laptop chip, even in 970FX form.
The G5 relies on high Mhz and a huge FSB for it's speed. Slowing it down to 1.3-1.6Ghz and limiting it to a DDR333/400 FSB may render it little better than a G4 in it's current state.
Of course, and even if it is worth to slow it down to, say, 1.6 GHz, then what will be the next upgrade step, when this chip is hot as hell (in the slim Powerbook case) once it passes beyond 1.8 GHz? I don't think Apple would try such a solution without having an upgrade path for 3-4 generations of the same chip (with light alterations eventually). Unless it is 100% compatible with something more efficient in its late development stages.
The 970fx in the Xserve is running at 1.32 v. IBM has posted variants of the fx that run at 1 volt, so that seems to be possible.
Once yields are better, IBM can take high-clocking 970fx's, and run them at 1 volt, for a 1/3 reduction in power and (approximately) clock rate. That won't be a cheap line, though, because it's taking the cream of the crop and clocking it low.
A low voltage 970 would, at least, hardly be worse than the 7400 G4s that went into the first TiBooks. But it leaves Freescale a lot of room to repackage a G4-like core into a low-wattage, high-voltage (i.e., cheap) notebook CPU. Move the memory controller on die (no MaxBus), and you're pretty much there. Add an FPU, and you have a sweet little chip. At 90nm, it could either run really cool, or Mot could take it dual core and it would still run cool enough.
Add an FPU, and you have a sweet little chip. At 90nm, it could either run really cool, or Mot could take it dual core and it would still run cool enough.
You mean to have two FPUs? What impact would have this on performance and heat?
You mean to have two FPUs? What impact would have this on performance and heat?
Right now, the G4 gets slaughtered on scalar FP because it only has one floating point unit. So all floating point work queues up to go through this bottleneck one at a time. Contrast this to the G4's four integer units. A second floating point unit and a little extra support for instruction scheduling and reordering would nearly double the G4's FP performance. This is an extremely common tactic. I believe it's the only "high-end" CPU left that doesn't have two FP units, although I could be missing a "mobile" CPU somewhere.
Basically, look at the difference between the 970's FP performance and the G4's: The G4 has a very nice FP unit, so that's not the problem. The remaining 970 advantages are bandwidth, dual FP units, and clockrate. An on-die memory controller and a second FP unit knock down two out of three, and a process shrink will at least keep the third under control.
As for heat and die size, the entire AltiVec unit on the 7455 takes up less than 10% of the die, so I wouldn't expect a single FP unit to have a more than negligible impact, even accounting for the additional circuitry that will almost certainly be needed to take advantage of the additional parallelism. Die size might not change at all: There's a lot of space on the 745x die, because a lot of the circuitry is spaced out generously to avoid hot spots on the die.
Basically, look at the difference between the 970's FP performance and the G4's: The G4 has a very nice FP unit, so that's not the problem. The remaining 970 advantages are bandwidth, dual FP units, and clockrate.
Which CPUs, other than the 970, have two FPU units?
Quote:
As for heat and die size, the entire AltiVec unit on the 7455 takes up less than 10% of the die, so I wouldn't expect a single FP unit to have a more than negligible impact, even accounting for the additional circuitry that will almost certainly be needed to take advantage of the additional parallelism. Die size might not change at all: There's a lot of space on the 745x die, because a lot of the circuitry is spaced out generously to avoid hot spots on the die.
From what you are saying, I understand that adding a second FPU in the G4 depends only on a decision, since there are no significant technical difficulties. Right? So, why the G4s don't have this second FPU? Is it again all about the well known Motorola specialisation in screwing up things?
Comments
Originally posted by Matsu
That isn't true at all. People keep making this statement with the only basis in IBM's microprocessor forum literature. This type of literature often conflates reality, with best case scenario testing, with projections, and a healthy dose of optimism -- a sort of advertorial/infomercial session really. According to this type of report IBM should have had 2Ghz G3's, or moto's G4 would already be at 1.8Ghz with a DDR FSB. It's rather obvious that they don't represent reality.
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
Well said.
The G4 is a much simpler core than the G5 becuase it has shorter pipelines, far less out-of-order execution capability, is only 32-bit, and a high performance bus interface. As a result of the first two items the G4 has much smaller execution resources (registers, buffers, etc). These differences give the G5 the ability to scale to much higher clock rates and perform well despite the huge latencies on data from main memory, even in a multi-processor system. At G4 clock rates, however, the G5 will be slower than the G4 on many many tasks. Basically, if you don't run it at high speeds, it loses many of its advantages over the G4 because its architecture is designed from the ground up to run at high clock rates.
I'm not prepared to scoff at Freescale hitting 90nm this year because its not their fabs in question -- ST Microelectronics and Philips will very likely benefit them tremendously. If Freescale can focus on the processor / SoC desgin (which has always been their strength) then we may see a G4 that addresses the big weaknesses of past G4s. All it would take is an on-chip memory controller to eliminate the MPX bus, and an RIO or HT interface to the I/O controller. 90 nm would probably push the clock rate up to 1.8-2 GHz. Such a chip could blow the stuffing out of the Pentium-M, just like the G4 used to best the PentiumPro/II/III before Intel pushed its bus technology past what MPX could provide. And do this at a lower power dissapation point than the G5 would achieve with the same performance.
Apple does not need a single processor line -- they need 2 or 3. The iBook, iMac, PB, PowerMac, and Xserve cover a broad spectrum of processor requirements and the biggest problem with the G4 for all those years was that Apple only had the G4 to cover the whole gamut (there being not much to distinguish the G3 from the G4, aside from AltiVec the lack of which caused more problems). My hope is that now they can have 2 G5-class processors (970 & 975), and a heavily updated G4.
Originally posted by Fat Freddy
Unbelievable!
This isn't the old Motorola. It's not even just Motorola. You can bet against Philips Semiconductor and STM Micro (Europe's largest contract fab) if you want to.
Crolles is online now, making product at 90nm now. They're (wisely) starting out with simple stuff like memory before tackling something as complex as a CPU. So far, the whole Crolles partnership is on schedule.
Get used to it. Frankly, while I understand the impulse to be skeptical of Motorola, I have to wonder about the hostility toward the prospect of a valuable second supplier.
Oh, and check Motorola's last quarter results, will you? You might find them interesting.
Originally posted by Matsu
That isn't true at all. People keep making this statement with the only basis in IBM's microprocessor forum literature. This type of literature often conflates reality, with best case scenario testing, with projections, and a healthy dose of optimism -- a sort of advertorial/infomercial session really. According to this type of report IBM should have had 2Ghz G3's, or moto's G4 would already be at 1.8Ghz with a DDR FSB. It's rather obvious that they don't represent reality.
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
That is all probably true, yet, if I remember correctly, the current G4 has more transistors on die than the G5. I believe the #'s are something like 52 million for the G5 and 58 million for the G4. So if my memory is good,
Originally posted by rickag
That is all probably true, yet, if I remember correctly, the current G4 has more transistors on die than the G5. I believe the #'s are something like 52 million for the G5 and 58 million for the G4. So if my memory is good,
I don't remember exactly, but I seem to remember that from somewhere.
Originally posted by oldmacfan
I don't remember exactly, but I seem to remember that from somewhere.
The information is available in product information pdf's at both IBM and Motorola, but I'm too lazy to look it up. Other things affect power besided just transistor count, like clock speed, voltage, leakage, etc. I only mentioned it in passing to kind of re-inforce the distinct possibility that the G5 does in fact dissipate less heat.
Originally posted by rickag
The information is available in product information pdf's at both IBM and Motorola, but I'm too lazy to look it up. Other things affect power besided just transistor count, like clock speed, voltage, leakage, etc. I only mentioned it in passing to kind of re-inforce the distinct possibility that the G5 does in fact dissipate less heat.
I thought clock for clock that, that was already known.
Originally posted by oldmacfan
I thought clock for clock that, that was already known.
I thought so too, however, it seems that some people are contending that the typical and maximum usages as measured/reported by Motorola and IBM are not comparable, which may very well be the case, who knows. That, and people keep "banging the pot", saying that the G5 uses too much power to be used in a laptop, even though the numbers indicate otherwise.
Originally posted by spliff monkey
Regarding the concern for a powerbook upgrade I see little need for it at the moment. Perhaps marketing concerns will push apple to develope a G5 powerbook quciker, but the idea that PC laptops are killing powerbooks is pretty silly. I can understand if you'd say pc's are punchier (quicker to exacute simpler commande) but in terms of number crunching the mac have it in spades. I own a G3 pismo 500 and a PIV 2.6 GHZ vaio, each has a gig of ram and I use both interchangeably depending on my need and the pismo at 1/4 the clock speed hold up very well to heavy 3d and video rendering in FCP, lightwave and after effects just as well if not better than the VAIO. Truthfully the pc is junk and I wouldn't own it if I didn't need to, people need to look past the specs and get themselves some real world "practice" on these machines, for thier purpose before they make vague blanket statements like pc laptops are better than apple's. Silly!
The current PowerBooks may or may not hold up to the new 1.7Ghz centrinos for no benchmarks have been released yet. If I remember correctly the last revision fell behind in many tasks, although not by all that much. The G4 in its current form is not going to be able to keep up with the centrino. It needs a faster Bus, a smaller die shrink, and a 1MB L2 cache. If Freescale/motorola can pull this off the G4 should be able to surpass the centrino. The problem currently stems from the slow bus which substantially affects the G4s ability to process large amounts of data.
The G3 is a good chip, but I don't believe for a second that it comes anywhere close to the 2.6Ghz P4M. The P4M may be a lot slower than its desktop sibling but it is not that slow.
The PowerBooks do have one large advantage over most PC laptops and thats the high end graphic cards apple puts into them. They should really help any video benchmarks out.
Originally posted by Algol
The P4M may be a lot slower than its desktop sibling but it is not that slow.
I thought it is about the same thing. In what they do differ?
Originally posted by Algol
The current PowerBooks may or may not hold up to the new 1.7Ghz centrinos for no benchmarks have been released yet. If I remember correctly the last revision fell behind in many tasks, although not by all that much. The G4 in its current form is not going to be able to keep up with the centrino. It needs a faster Bus, a smaller die shrink, and a 1MB L2 cache. If Freescale/motorola can pull this off the G4 should be able to surpass the centrino. The problem currently stems from the slow bus which substantially affects the G4s ability to process large amounts of data.
The G3 is a good chip, but I don't believe for a second that it comes anywhere close to the 2.6Ghz P4M. The P4M may be a lot slower than its desktop sibling but it is not that slow.
And its getting WORSE!, 2GHz Centrino , and its shipping in May!.
So hopefully, the next bump in Jan is a bit more than 100MHz? Oh no let me guess, we dont need faster CPU's because its all about the OS???
Originally posted by hasapi
And its getting WORSE!, 2GHz Centrino , and its shipping in May!.
Did you notice the L2 cache size
Originally posted by PB
Did you notice the L2 cache size
I thought this was the most interesting statement:
By 2006, a Pentium M derivative called Jonah is expected to become the mainstay for the desktop line.
Wither Pentium4?
Originally posted by Matsu
Production represents reality. Looking at Xserves and PMs, I think it's reather obvious that while the G5 is cool for a chip of its intended market, it isn't close to as cool as a 1.5Ghz 7447A.
Not only that, but some recent real data **confirm** that this G5 would not go in a Powerbook. We will see perhaps one day a G5 Powerbook, but I seriously doubt it will have the G5 we all know today.
And I explain. xlr8yourmac posted yesterday some comments of a XServe G5 user. Let's see:
Xserve G5 owner comments on CPU Power Usage/Heat
- A reader with a new Xserve G5 wrote regarding the lower-power/wattage version (compared to the original G5 towers) of the IBM PowerPC970 it uses
"Re: observations G5 970FX in new XServe
hi mike,
we received our xserve G5 a few days ago. it is the model with one 2.0GHz processor.
theserver-monitor application is pretty much verbose: Vcore of this G5 is 1.32 Volts. when idling it draws about 15 watts (or 11.8amps) (15W/1.32V = appx 11.36A), under 100% cpu-load it draws 44.5 watts or 33.5 amps. (remember this is at 1.32V and using 44.5W/1.32V = appx 33.7A)
the maximum core-temp from the cpu under 100% load was 75 degrees celsius. ambient temp. was at about 25 degrees celsius.
best regards,
michael b.
P4 northwood draws about 18 amps when idle. "
Now, let someone explain to me how do you put a 44.5 W chip (processor alone--granted, at 100% CPU load) in a thin notebook like the Powerbook.
The G5 relies on high Mhz and a huge FSB for it's speed. Slowing it down to 1.3-1.6Ghz and limiting it to a DDR333/400 FSB may render it little better than a G4 in it's current state.
Basically, cooling it down enough for laptop use, may render it too slow. It isn't looking like a laptop chip, even in 970FX form.
Originally posted by Matsu
The G5 relies on high Mhz and a huge FSB for it's speed. Slowing it down to 1.3-1.6Ghz and limiting it to a DDR333/400 FSB may render it little better than a G4 in it's current state.
Of course, and even if it is worth to slow it down to, say, 1.6 GHz, then what will be the next upgrade step, when this chip is hot as hell (in the slim Powerbook case) once it passes beyond 1.8 GHz? I don't think Apple would try such a solution without having an upgrade path for 3-4 generations of the same chip (with light alterations eventually). Unless it is 100% compatible with something more efficient in its late development stages.
Once yields are better, IBM can take high-clocking 970fx's, and run them at 1 volt, for a 1/3 reduction in power and (approximately) clock rate. That won't be a cheap line, though, because it's taking the cream of the crop and clocking it low.
A low voltage 970 would, at least, hardly be worse than the 7400 G4s that went into the first TiBooks. But it leaves Freescale a lot of room to repackage a G4-like core into a low-wattage, high-voltage (i.e., cheap) notebook CPU. Move the memory controller on die (no MaxBus), and you're pretty much there. Add an FPU, and you have a sweet little chip. At 90nm, it could either run really cool, or Mot could take it dual core and it would still run cool enough.
Originally posted by Amorph
Add an FPU, and you have a sweet little chip. At 90nm, it could either run really cool, or Mot could take it dual core and it would still run cool enough.
You mean to have two FPUs? What impact would have this on performance and heat?
Originally posted by PB
You mean to have two FPUs? What impact would have this on performance and heat?
Right now, the G4 gets slaughtered on scalar FP because it only has one floating point unit. So all floating point work queues up to go through this bottleneck one at a time. Contrast this to the G4's four integer units. A second floating point unit and a little extra support for instruction scheduling and reordering would nearly double the G4's FP performance. This is an extremely common tactic. I believe it's the only "high-end" CPU left that doesn't have two FP units, although I could be missing a "mobile" CPU somewhere.
Basically, look at the difference between the 970's FP performance and the G4's: The G4 has a very nice FP unit, so that's not the problem. The remaining 970 advantages are bandwidth, dual FP units, and clockrate. An on-die memory controller and a second FP unit knock down two out of three, and a process shrink will at least keep the third under control.
As for heat and die size, the entire AltiVec unit on the 7455 takes up less than 10% of the die, so I wouldn't expect a single FP unit to have a more than negligible impact, even accounting for the additional circuitry that will almost certainly be needed to take advantage of the additional parallelism. Die size might not change at all: There's a lot of space on the 745x die, because a lot of the circuitry is spaced out generously to avoid hot spots on the die.
Originally posted by Amorph
Basically, look at the difference between the 970's FP performance and the G4's: The G4 has a very nice FP unit, so that's not the problem. The remaining 970 advantages are bandwidth, dual FP units, and clockrate.
Which CPUs, other than the 970, have two FPU units?
As for heat and die size, the entire AltiVec unit on the 7455 takes up less than 10% of the die, so I wouldn't expect a single FP unit to have a more than negligible impact, even accounting for the additional circuitry that will almost certainly be needed to take advantage of the additional parallelism. Die size might not change at all: There's a lot of space on the 745x die, because a lot of the circuitry is spaced out generously to avoid hot spots on the die.
From what you are saying, I understand that adding a second FPU in the G4 depends only on a decision, since there are no significant technical difficulties. Right? So, why the G4s don't have this second FPU? Is it again all about the well known Motorola specialisation in screwing up things?