970 vs. 970fx

[osprey76]

i was under the impression that less power consumption meant less heat generated, but it seems that that's not correct as that graphic states that the junction temp. range, whatever that means, is larger for the 970fx. anyone know what's up?

You're confusing heat and temperature. The chip does produce less heat, but does apparently run at a higher temperature.

The concept of heat versus temperature confuses a lot of people. A rough parallel would be a gallon of water at 200F versus a cup of water at 212F and cooling each to 100F. Sure the cup of water has a higher temperature, but the gallon of water contains more heat. This is why you read of fusion experiments that reach temperatures of millions of degrees (F or C), but they don't contain enough heat to boil a cup water. There are a few dozen atoms at millions of degrees, and several million water atoms at room temperature. You could toss the fusion core in the water and it would barely sizzle for a microsecond as the fusion atoms are cooled very quickly by the enormous (relatively) amount of water.

The heat a chip produces is directly proportional to the power it uses. So, the 970 will probably produce about twice as much heat as the 970FX. However, the smaller surface area of the 970FX chip means that it is harder to pull the heat out and results in higher temperatures. Surface area tosses in another important aspect of heat transfer and I hope this helps everyone understand better rather than confuse them!

 

[alexf]

You're confusing heat and temperature. The chip does produce less heat, but does apparently run at a higher temperature.

The concept of heat versus temperature confuses a lot of people. A rough parallel would be a gallon of water at 200F versus a cup of water at 212F and cooling each to 100F. Sure the cup of water has a higher temperature, but the gallon of water contains more heat. This is why you read of fusion experiments that reach temperatures of millions of degrees (F or C), but they don't contain enough heat to boil a cup water. There are a few dozen atoms at millions of degrees, and several million water atoms at room temperature. You could toss the fusion core in the water and it would barely sizzle for a microsecond as the fusion atoms are cooled very quickly by the enormous (relatively) amount of water.

The heat a chip produces is directly proportional to the power it uses. So, the 970 will probably produce about twice as much heat as the 970FX. However, the smaller surface area of the 970FX chip means that it is harder to pull the heat out and results in higher temperatures. Surface area tosses in another important aspect of heat transfer and I hope this helps everyone understand better rather than confuse them!

Thank you, that does clarify things quite a bit.

But the real question: what does this mean in terms of cooling systems? Some have written that the 970fx chip requires more work to be cooled off (probably resulting in a noisier machine), while others have written the exact opposite.

What do you think?

 

[alexf]

The new iMacs are already months late. The PPC 970FX has supposedly been in products since January 2004 (when the XServe G5 was announced and shown at MWSF). The fact that it is now almost September and we are still waiting for the iMac G5, and supplies of high-end PowerMac G5s are still highly limited indicates that the 970FX is not being produced in large numbers yet. Apple said as much in their most recent financial results conference call where they blame IBM for the delay in shipping iMac G5s.

So, what I still don't understand is that, if the newer 970fx chips do in fact get hotter and require a more elaborate cooling system (if this is true), why don't they just put the older 970 chips into the new iMacs?

Am I missing something?

 

[Mord]

It *WAS* supposed to be cooler, etc. It's not.

The only upside to a 90nm chip right now is for the manufacturer: more chips fit on a single wafer of silicon, increasing fab capacity. For everyone else (computer manufacturers who bhave to design fancy cooling systems for these chips and consumers who have to live with hotter, noisier computers), 90nm chips just plain suck.

In the case of the 970fx, IBM is seeing such poor yields that this capacity increase hasn't materialized - 970fx yields are ultra-low, and this is why the G5 imac is nowhere to be found. Not enough CPUs.

Basically the 970fx is a motorola-grade fiasco.

-vga4life

dude there was a thing on xlr8yourmac which had the recorded power consumption of the 970fx and it did average at 25w and peak at 35w, and this was a single 2.0GHz xserve.

as for the dude that was confused as to why ibm states that the 970fx is stated as a higher temp cpu as a 970 it means it needs less cooling and will run at 105 degrees stable.

 

[shadowband]

Thank you, that does clarify things quite a bit.

But the real question: what does this mean in terms of cooling systems? Some have written that the 970fx chip requires more work to be cooled off (probably resulting in a noisier machine), while others have written the exact opposite.

What do you think?

The real answer will require a bake-off between a 2.0GHz powermac using PowerPC G5 version 2.2 CPUs (PPC970) and one using PowerPC G5 version 3.0 CPUs (PPC970fx). The Hardware Monitor (shareware) records the CPU die temperature, CPU voltage, and the speed of the various fans. The fan speed comparison is ultimately what is required to determine whether or not the 970fx results in a quieter system. Anyone want to start a pool

 

[osprey76]

Thank you, that does clarify things quite a bit.

But the real question: what does this mean in terms of cooling systems? Some have written that the 970fx chip requires more work to be cooled off (probably resulting in a noisier machine), while others have written the exact opposite.

What do you think?

That sounds valid to me. With a higher heat density (lower power, but a much smaller area to extract it) you would need another wind tunnel machine or a new cooling system. A la liquid cooling in the 2.5 machine. Water is a much better cooling medium than air. You use the water to pull the heat out of the processor and then you can use a radiator type device to spread the heat out and use relatively low velocity air to vent the heat out of the case.

Whether this ends up being noisier or not is dependent on several factors. Velocity of the air in the case, fan speed, heat sink, etc. It just all depends on how Apple engineered the cooling solution and if they made any changes from Rev. A to Rev. B on the air-cooled machines.

 

[alexf]

It just all depends on how Apple engineered the cooling solution and if they made any changes from Rev. A to Rev. B on the air-cooled machines.

Did they make any changes? Someone here must know...

 

[Flynnstone]

I doesn't make a big difference between the Rev A & B.
The differences from the user point is likely very little.
The rest is ... academic. A roughly 50 watt difference will not make a small rodent's buttocks.
The problem with 90nm chips is the leakage current is excessive.

I think the more important question is if the power supply clicking is gone in the rev Bs.

 

[alexf]

A roughly 50 watt difference will not make a small rodent's buttocks.

Do small rodents really have buttocks?

 

[oingoboingo]

So, what I still don't understand is that, if the newer 970fx chips do in fact get hotter and require a more elaborate cooling system (if this is true), why don't they just put the older 970 chips into the new iMacs?

Am I missing something?

You've got me there. I don't know. At this stage maybe Apple should just rename the MPC 7447A to 'G5' and make the launch.

 

[Abstract]

Heh, yeah, or go back to 180 nm chips, hike the 970 nm up to 3GHz, and away we go!

Of course, this depends on whether its true that these new 90nm cpu's produce less heat, but are technically more difficult to cool. But hey, just keep making the die sizes larger and larger, and according to the information posted in this thread so far, we could be at 4GHz by next week!

 

[alexf]

Perhaps the reason that Apple doesn't put the original 970 chips in the iMac is simply cost - the newer chips are simply cheaper to make.

In any case, I will be very curious to see the cooling system on the new iMacs... If the new chips really do run hotter and need to be cooled off better I am curious as to how Apple will manage this in a small enclosure without turning the machine into another windtunnel.

 

[alexf]

The real answer will require a bake-off between a 2.0GHz powermac using PowerPC G5 version 2.2 CPUs (PPC970) and one using PowerPC G5 version 3.0 CPUs (PPC970fx). The Hardware Monitor (shareware) records the CPU die temperature, CPU voltage, and the speed of the various fans. The fan speed comparison is ultimately what is required to determine whether or not the 970fx results in a quieter system. Anyone want to start a pool

Yes, this would be a good idea...

Anyone?

BTW, I think I may have read somewhere that there is a slight speed difference between a similar GHz 970 and 970fx.

Can anyone confirm this?

 

[Sun Baked]

Re: from M. Isobe

Here's an IBM graphic on the 2.5GHz PPC970fx...

Figure 2. Maximum power envelope from 0.8 to 1.3 V showing the power reduction possible through power-tuning methods.

Figure 3. Maximum power and nap power envelopes showing transitions from maximum power to nap power at f, f/2 and f/4, and then to deep nap at f/64.

 

[alexf]

Here's an IBM graphic on the 2.5GHz PPC970fx...

Figure 2. Maximum power envelope from 0.8 to 1.3 V showing the power reduction possible through power-tuning methods.

Figure 3. Maximum power and nap power envelopes showing transitions from maximum power to nap power at f, f/2 and f/4, and then to deep nap at f/64.

It's all Greek to me...

 

[Sun Baked]

It's all Greek to me...

IBM is stating that the Maximum Power Number is twice the 50W Typical number for the PPC970FX -- 100W.

Which is much higher than the numbers they gave us for Maximum Power Dissipation for either the 1.8/2.0 PPC970 which had 51W/66W Typical.

---

EDIT -- If you look here you'll see some more evidence of the problems IBM has been having with the PPC970 (aka, low voltage operation & reliability at higher clock speed).

 

[alexf]

IBM is stating that the Maximum Power Number is twice the 50W Typical number for the PPC970FX -- 100W.

Which is much higher than the numbers they gave us for Maximum Power Dissipation for either the 1.8/2.0 PPC970 which had 51W/66W Typical.

---

EDIT -- If you look here you'll see some more evidence of the problems IBM has been having with the PPC970 (aka, low voltage operation & reliability at higher clock speed).

Right, understood - thanks.

Yet the question stilll remains: would me, as the end user, notice any performance difference at all between the 2 chips? Or is this still unknown?

 

[Catfish_Man]

IBM is stating that the Maximum Power Number is twice the 50W Typical number for the PPC970FX -- 100W.

Which is much higher than the numbers they gave us for Maximum Power Dissipation for either the 1.8/2.0 PPC970 which had 51W/66W Typical.

---

EDIT -- If you look here you'll see some more evidence of the problems IBM has been having with the PPC970 (aka, low voltage operation & reliability at higher clock speed).

Did they actually ever list the max power for the 970? I know they had 66 watts typical but I never heard any confirmed max power numbers. This graph would suggest >120 watts, but the ratio may have changed between 970 and 970fx. The lack of 0.8 volt operation also may be a reason why the iMac was delayed. If IBM does a revision of the chip that can support that it could significantly lower power consumption at low frequencies.

alexf: The chips should be identical from a user perspective. Any performance gain on the fx (minor errata fixed, stuff tweaked, whatever) will be too small to notice, if it even exists.

 

[alexf]

alexf: The chips should be identical from a user perspective. Any performance gain on the fx (minor errata fixed, stuff tweaked, whatever) will be too small to notice, if it even exists.

Thanks for the info.