For what it's worth: 7448 info

tht

Some clues about Crolles' 90 nm fab. From Freescale's website:

Code:

1.3 Volt 7448 CPUs



1000 MHz 1420 MHz 1600 MHz 1700 MHz

Typical ? Nominal 65 °C 15.0 19.0 20.0 21.0 W

Typical ? Thermal 105 °C 18.6 23.3 24.4 25.6 W

Maximum 105 °C 21.6 27.1 28.4 29.8 W



1.0 Volt 7448 CPUs



1000 MHz 1400 MHz

Typical ? Nominal 65 °C 9.5 11.0 W

Typical ? Thermal 105 °C 12.0 13.7 W

Maximum 105 °C 13.9 15.9 W









Those are the highest MHz CPUs advertized by Freescale. Scaling to 2 GHz at 1.3V, a 7448 would draw 35 Watt which is ok. If Freescale could produce a 1.1V 2 GHz part, it'd draw 25 Watt which would make a good and thin laptop chip.



The key for every CPU fab is producing a low voltage high MHz processor, but its doubtful that they can do it, at least with any volume. Maybe in 2H 06.



This also implies that a 1.7 GHz 8641D, the dual core, will likely draw in the neighborhood of 40W, and a 2 GHz one somewhere around 50 W.



So, good move Apple in going with Intel Core Duo.

backtomac

[QUOTE]Originally posted by THT

Some clues about Crolles' 90 nm fab. From Freescale's website:



Code:



1.3 Volt 7448 CPUs



1000 MHz 1420 MHz 1600 MHz 1700 MHz

Typical ? Nominal 65 °C 15.0 19.0 20.0 21.0 W

Typical ? Thermal 105 °C 18.6 23.3 24.4 25.6 W

Maximum 105 °C 21.6 27.1 28.4 29.8 W



1.0 Volt 7448 CPUs



1000 MHz 1400 MHz

Typical ? Nominal 65 °C 9.5 11.0 W

Typical ? Thermal 105 °C 12.0 13.7 W

Maximum 105 °C 13.9 15.9 W









Those are the highest MHz CPUs advertized by Freescale. Scaling to 2 GHz at 1.3V, a 7448 would draw 35 Watt which is ok. If Freescale could produce a 1.1V 2 GHz part, it'd draw 25 Watt which would make a good and thin laptop chip.



The key for every CPU fab is producing a low voltage high MHz processor, but its doubtful that they can do it, at least with any volume. Maybe in 2H 06.



This also implies that a 1.7 GHz 8641D, the dual core, will likely draw in the neighborhood of 40W, and a 2 GHz one somewhere around 50 W.



So, good move Apple in going with Intel Core Duo.



I thought I saw where Intel is about to move to 45nm fab. Seems to me that Intel is about to put some distance between themselves and the competition. At least IBM/Freescale.

kim kap sol

Quote:

Originally posted by THT

Some clues about Crolles' 90 nm fab. From Freescale's website:



Code:

1.3 Volt 7448 CPUs



1000 MHz 1420 MHz 1600 MHz 1700 MHz

Typical ? Nominal 65 °C 15.0 19.0 20.0 21.0 W

Typical ? Thermal 105 °C 18.6 23.3 24.4 25.6 W

Maximum 105 °C 21.6 27.1 28.4 29.8 W



1.0 Volt 7448 CPUs



1000 MHz 1400 MHz

Typical ? Nominal 65 °C 9.5 11.0 W

Typical ? Thermal 105 °C 12.0 13.7 W

Maximum 105 °C 13.9 15.9 W

Those are the highest MHz CPUs advertized by Freescale. Scaling to 2 GHz at 1.3V, a 7448 would draw 35 Watt which is ok. If Freescale could produce a 1.1V 2 GHz part, it'd draw 25 Watt which would make a good and thin laptop chip.



The key for every CPU fab is producing a low voltage high MHz processor, but its doubtful that they can do it, at least with any volume. Maybe in 2H 06.



This also implies that a 1.7 GHz 8641D, the dual core, will likely draw in the neighborhood of 40W, and a 2 GHz one somewhere around 50 W.



So, good move Apple in going with Intel Core Duo.



I agree.



A Yonah running at 1.6GHz is probably almost as fast as a G4 running at 2GHz (possibly slightly slower)...but the kicker is that, with the Yonah, you can have 2 cores.



I'd rather have a two processors running slightly slower than one processor running slightly faster. I think almost everyone does.

electric monk

Quote:

Originally posted by THT

Some clues about Crolles' 90 nm fab. From Freescale's website

I have to wonder about that fab. I mean sure we haven't heard the same kind of horror stories IBM had at Fishkill but they certainly seem(ed?) to be perpetually a quarter or two away from volume for a while now. Plus the 7448 was supposed to be shipping last quarter in volume, and the 8641 should have been sampling/limited by now. So.



And I know the advertised speeds are at least ~300 MHz short if Apple wanted to buy the chip but still. If the 7447b can be overclocked safely to 1.92 and 2 GHz...



I'm actually worried about what kind of speeds the 8641/D will debut at. It doesn't matter much to their target audience, but the kind of horrified reactions the Mac community might have had if it comes in at 7448 speeds improved bus or no. Shudder.

tubgirl

Quote:

Originally posted by THT

Code:

1.0 Volt 7448 CPUs



1000 MHz 1400 MHz

Typical ? Nominal 65 °C 9.5 11.0 W

Typical ? Thermal 105 °C 12.0 13.7 W

Maximum 105 °C 13.9 15.9 W

apple might put one of these bad boys in the next ipod...

anders

That would probably be a downgrade

dh87

Quote:

Originally posted by THT

Some clues about Crolles' 90 nm fab. From Freescale's website:



Code:

1.3 Volt 7448 CPUs



1000 MHz 1420 MHz 1600 MHz 1700 MHz

Typical ? Nominal 65 °C 15.0 19.0 20.0 21.0 W

Typical ? Thermal 105 °C 18.6 23.3 24.4 25.6 W

Maximum 105 °C 21.6 27.1 28.4 29.8 W



1.0 Volt 7448 CPUs



1000 MHz 1400 MHz

Typical ? Nominal 65 °C 9.5 11.0 W

Typical ? Thermal 105 °C 12.0 13.7 W

Maximum 105 °C 13.9 15.9 W

Those are the highest MHz CPUs advertized by Freescale. Scaling to 2 GHz at 1.3V, a 7448 would draw 35 Watt which is ok. If Freescale could produce a 1.1V 2 GHz part, it'd draw 25 Watt which would make a good and thin laptop chip.



The key for every CPU fab is producing a low voltage high MHz processor, but its doubtful that they can do it, at least with any volume. Maybe in 2H 06.



This also implies that a 1.7 GHz 8641D, the dual core, will likely draw in the neighborhood of 40W, and a 2 GHz one somewhere around 50 W.



So, good move Apple in going with Intel Core Duo.



I know that extrapolating power consumption has been discussed before, but the 1.3V data here are linear in power vs. frequency, leading to 33W at maximum load and 24W at typical - nominal (whatever that means). Applying this linear extrapolation to the 1.0V numbers gives 19W at maximum and 13W at typical. Why is my extrapolation incorrect?

tht

Quote:

Originally posted by dh87

I know that extrapolating power consumption has been discussed before, but the 1.3V data here are linear in power vs. frequency, leading to 33W at maximum load and 24W at typical - nomimal (whatever that means). Applying this linear extrapolation to the 1.0V numbers gives 19W at maximum and 13W at typical. Why is my extrapolation incorrect?

Because reality, ie, physics, gets in the way.



Semiconductor manufacturing is like other types of manufacturing or farming. The products that come out of the fab are of varying quality that operate under a varying set of conditions. The variation in quality is like a normal distribution (well, asymptotic depending parameters) where the true high quality parts that operate well under more extreme conditions are rare parts while the vast majority of parts coming out of the fab work ok under a set of looser conditions.



A 1V 2GHz 7448 would be one of those rare high quality parts to be able to operate at 2 GHz at 1 Volt, extreme conditions for that CPU, while the vast majority of 1V 7448 CPUs operate from 1 to 1.4 GHz.



To get to 2 GHz, you can increase the voltage, but that's a catch-22 because power would increase quadratically (or worse if current leakage isn't under control which could mean cubic increases) with voltage increases. Hence, Freescale has a "nominal" operating voltage of 1.3V with operating frequencies up to 1.7 GHz and 30 Watts. Nominal represents their high volume processors. They can probably get it to 2 GHz fairly easy with good quantities by using 1.35-1.4 Volt, but that'll mean a 40 Watt 2 GHz chip.



So, 2 things work against your extrapolation. To get higher operating frequencies, voltages have to be increased [or advancing to the next node ], and those higher frequency lower voltage parts tend to be lower yield, rarer parts making a <30 Watt 2 GHz 7448 part difficult to achieve in real quantities.



Keep in mind that the 7448 is a short pipeline design. It will take more power to drive it to 2 GHz compared to the longer pipeline designs of the Pentium-M and Turion. It should take more power to drive the 7448 to 2 GHz than the Pentium-M or Turion. Not entirely sure, but that's what I'm thinking right now. It also could be that Crolles is still immature too, who knows.



Another interesting question is why does Daystar use a 7447A, a 130 nm part, for their 2 GHz Powerbook upgrade instead of the 7448? The only answer that comes to mind is that the 7448 is too expensive or the 7448 is unavailable. I'd probably go with the former.

dh87

Quote:

Originally posted by THT

Because reality, ie, physics, gets in the way.



Semiconductor manufacturing is like other types of manufacturing or farming. The products that come out of the fab are of varying quality that operate under a varying set of conditions. The variation in quality is like a normal distribution (well, asymptotic depending parameters) where the true high quality parts that operate well under more extreme conditions are rare parts while the vast majority of parts coming out of the fab work ok under a set of looser conditions.



A 1V 2GHz 7448 would be one of those rare high quality parts to be able to operate at 2 GHz at 1 Volt, extreme conditions for that CPU, while the vast majority of 1V 7448 CPUs operate from 1 to 1.4 GHz.



To get to 2 GHz, you can increase the voltage, but that's a catch-22 because power would increase quadratically (or worse if current leakage isn't under control which could mean cubic increases) with voltage increases. Hence, Freescale has a "nominal" operating voltage of 1.3V with operating frequencies up to 1.7 GHz and 30 Watts. Nominal represents their high volume processors. They can probably get it to 2 GHz fairly easy with good quantities by using 1.35-1.4 Volt, but that'll mean a 40 Watt 2 GHz chip.



So, 2 things work against your extrapolation. To get higher operating frequencies, voltages have to be increased [or advancing to the next node ], and those higher frequency lower voltage parts tend to be lower yield, rarer parts making a <30 Watt 2 GHz 7448 part difficult to achieve in real quantities.



Keep in mind that the 7448 is a short pipeline design. It will take more power to drive it to 2 GHz compared to the longer pipeline designs of the Pentium-M and Turion. It should take more power to drive the 7448 to 2 GHz than the Pentium-M or Turion. Not entirely sure, but that's what I'm thinking right now. It also could be that Crolles is still immature too, who knows.



Another interesting question is why does Daystar use a 7447A, a 130 nm part, for their 2 GHz Powerbook upgrade instead of the 7448? The only answer that comes to mind is that the 7448 is too expensive or the 7448 is unavailable. I'd probably go with the former.

Thanks for clarifying this.

skatman

Quote:

If Freescale could produce a 1.1V 2 GHz part, it'd draw 25 Watt which would make a good and thin laptop chip.

25 watts is way too much for thin and light. Pentium M (Dothan) ULV draws 7 watts and even then the laptop peeps have trouble with heat removal.

Also, that is the idle power consumption on those chips? That is a lot more important than then full throttle number.

pb

Quote:

Originally posted by kim kap sol

I agree.



A Yonah running at 1.6GHz is probably almost as fast as a G4 running at 2GHz (possibly slightly slower)...but the kicker is that, with the Yonah, you can have 2 cores.



I'd rather have a two processors running slightly slower than one processor running slightly faster. I think almost everyone does.

And now guess what, Apple does not offer at this moment Intel based machines with lower than 1.83 GHz clock frequency. And all of them are dual core. We will probably see the 1.6 GHz single core Yonah in iBooks or the Mac mini in the not so distant future. Ouch!