More Liquid Cooling Diagrams

[Sun Baked]

But the dude from Apple said (and this is where I have problem): that it is the higher density PowerPC 970FX chips that make it necessary to use the liquid cooling. Well...if all the new dualies have the 970FX chips (i.e. the density is the same because the chip design is the same -- same masks, etc), then why only liquid cooling for the top end? Speed (i.e. clock rate), not "chip density" (i.e. process shrink to 90nm) is "apparently" the source of the heat problem.

The 2.5GHz chip DOES operate at a higher energy density than the 2.0GHz PPC 970 or 970FX.

The 2.5GHz PPC970FX operates at a maximum power dissipation between 100-110W -- the 2.0GHz PPC970FX (50-55W) and below chips operate at much lower maximum power dissipation numbers.

Somebody did a quick calculation based on the surface area, and it does operate much higher than the other chips.

 

[tazznb]

......hhmmmm

Cooling is related to processor speed and surface area. Obviously only the 2.5 reaches the threshold where the old heatsinks can't handle the load.

Something from the barefeats article that caught my eye was the comment that the new dual 2.0 is identified as 7,3. I would assume the dual 2.5 is also. Which makes one wonder what 8,1 is....???

X-Station 3.0 Ghz

 

[stcanard]

Is this diagram thing getting old or is just me??

Maybe not just you but people with your interests...

Until they actually ship I can't get enough diagrams to see how it's working. A no-maintenance, (relatively) low cost water cooled consumer system. Definitely worth looking at as many design specs as possible.

But then again I've used a soldering iron on a 6 month old laptop motherboard ... the inside of all my electronics is just too much to stay away from.

 

[ethernet76]

By comparing them to Cubase and Avid Media Composer.

Cuebase runs on a mac too, quite wonderfully. My powerbook 1.25 ghz kicked the crap out of my friends overclocked athlon.

Why wouldn't they just use Cuebase for both, as cuebase has more features, and will probably be slower on a PC and a mac.

Apple benchmarks are shady.

 

[eSnow]

But the dude from Apple said (and this is where I have problem): that it is the higher density PowerPC 970FX chips that make it necessary to use the liquid cooling. Well...if all the new dualies have the 970FX chips (i.e. the density is the same because the chip design is the same -- same masks, etc), then why only liquid cooling for the top end? Speed (i.e. clock rate), not "chip density" (i.e. process shrink to 90nm) is "apparently" the source of the heat problem.

Well, the point is that the old 970 were not able to reach 2.5Ghz - the higher density allows for the speed increase - which brings a higher power dissipation. Furthermore the smaller structural size leads to the same amount of transistors on a smaller surface area, thereby exacerbating the problem.

So, the Apple guy was right in that the structure shrink is part of the problem - the other part is the higher speed you pointed out.

 

[shawnce]

The 2.5GHz chip DOES operate at a higher energy density than the 2.0GHz PPC 970 or 970FX.

The 2.5GHz PPC970FX operates at a maximum power dissipation between 100-110W -- the 2.0GHz PPC970FX (50-55W) and below chips operate at much lower maximum power dissipation numbers.

Somebody did a quick calculation based on the surface area, and it does operate much higher than the other chips.

Can you show documents that talk about the 2.5GHz version burn over 100W of power.

I doubt it to be true if the 2.0GHz 970FX is only burning 55W (which I found a document online a while back stating). If the cores are the same, the process is the same, then the power dissipation should roughly scale linearly with clock speed. In other words I would expect the 2.5GHz version to be burning roughly 70W.

 

[ethernet76]

Right, so the surface area of the 970FX (90nm) chips are all the same. Check. And this surface area is smaller than the 970 chips used before (130nm) in the older PMacs. Check.

That leaves ONLY processor speed (clock rate) as the differentiator (for heat generation and therefore cooling system requirements) on the 970FX machines (1.8D, 2.0D, 2.5D). Yeah, I expect the faster chips to produce more heat that the slower chips.

But the dude from Apple said (and this is where I have problem): that it is the higher density PowerPC 970FX chips that make it necessary to use the liquid cooling. Well...if all the new dualies have the 970FX chips (i.e. the density is the same because the chip design is the same -- same masks, etc), then why only liquid cooling for the top end? Speed (i.e. clock rate), not "chip density" (i.e. process shrink to 90nm) is "apparently" the source of the heat problem.

Was the Apple dude just talking jive marketspeak, or was he misquoted or paraphrased incorrectly? (I can't find the original quote from the links above).

Any chipheads with knowledge out there?

---MacUnit

Ok. Here is how it works, as I use to overclock my P3 to ridicous speeds and saw heat increase significantly.

When you increase the clock cycle, you increase the amount of energy flowing through the chip and more and more of the energy escapes as heat. This is some sort of non-linear equation deal.

Theoretically, if you ran the processor slow enough, you wouldn't need a heatsync. For example, when I halved my P3 600 down to a 300 I could turn off the processor fans and see temperatures of about 100 degrees F. Compare this to a Pention 2 at 300 and you might of overheated because the P3 design was a lot more efficient than the P2.

Now compare this when I got the 600 to run at 933. To keep it at safe running temperatures I had the side panel door open and a 14-inch industrial fan blowing inside the case. It still ran at about 170 degrees F.

While the 970FX is more efficient than the original 970, it isn't a huge improvement.

Now considering the following. Take 1000 people put them in a 130 by 130 foot room. Then, take the same 1000 people and put them in a 90 by 90 room. Which room is going to be hotter?

So in effect, the chip generates more heat because the high clock speed in a smaller space. The chip has an effective surface area of 8100 nm squared. The original has twice that surface area.

So in reality liquid cooling was going to happen at some point. Apple has the unique ability to start the move into the off-the-shelve computer liquid cooling because they build the computers and set the pace for the mac world. Intel however is at the mercy of what Dell puts in there, and has hindered pentium 4 clock increases. Along with the pentium 4 being a ****ty chip.

 

[ethernet76]

Can you show documents that talk about the 2.5GHz version burn over 100W of power.

I doubt it to be true if the 2.0GHz 970FX is only burning 55W (which I found a document online a while back stating). If the cores are the same, the process is the same, then the power dissipation should roughly scale linearly with clock speed. In other words I would expect the 2.5GHz version to be burning roughly 70W.

The 970FX at 2 ghz consumes far less energy.

Typical power 51W@1.8GHz<- 970
25W@2.0GHz<- 970fx

Source: http://www.ibm.com/chips/techlib/techlib.nsf/techdocs/7874C7DA8607C0B287256BF3006FBE54/$file/PPC_QRG_2-22-04.pdf

 

[BrianKonarsMac]

It's not just you.

It's only water cooling, nothing to get too over excited about.

is it water though? i personally dont think it is. they seem to make a point aboud calling it liquid cooled, not water cooled. no proof obviously, unless u want to crack it open and do a taste test for me

 

[g4cubed]

This is supposed to fit into a PowerBook eventually? Wow, good luck...

Yea, that radiator hanging off the back would make them a bit heavy

Is this diagram thing getting old or is just me??

It's not just you.

It's only water cooling, nothing to get too over excited about.

And to those less interested or bored with the subject, just don't read those threads. It's your choice.

 

[BrianKonarsMac]

Well, the point is that the old 970 were not able to reach 2.5Ghz - the higher density allows for the speed increase - which brings a higher power dissipation. Furthermore the smaller structural size leads to the same amount of transistors on a smaller surface area, thereby exacerbating the problem.

So, the Apple guy was right in that the structure shrink is part of the problem - the other part is the higher speed you pointed out.

tje reason the density is such a problem is because you can longer cool the heat with a fan, its so hot in such a small area the air would never cool the center of the chip, just the edges. a liquid cooled device is able to pull heat off the entire chip.

And to those less interested or bored with the subject, just don't read those threads. It's your choice.

but that would require far more common sense than they have...the simplicity of this idea boggles the mind

 

[g4cubed]

is it water though? i personally dont think it is. they seem to make a point aboud calling it liquid cooled, not water cooled. no proof obviously, unless u want to crack it open and do a taste test for me

It is a combination of water and propylene glycol, a clear liquid used in automobile antifreeze.

 

[Sun Baked]

The 970FX at 2 ghz consumes far less energy.

Typical power 51W@1.8GHz<- 970
25W@2.0GHz<- 970fx

Source: http://www.ibm.com/chips/techlib/techlib.nsf/techdocs/7874C7DA8607C0B287256BF3006FBE54/$file/PPC_QRG_2-22-04.pdf

We were talking Maximum Power Dissipation which is the actual design criteria -- not the typical number.

And the Maximum number is about twice the Typical.

And here is an IBM Presentation on Power Dissipation

---

>shawnce

Ask this guy where he got the number for this post

Though it probably isn't too far off (or even a conservative number) -- it's the leakage that's killing us at 90nm.

 

[g4cubed]

but that would require far more common sense than they have...the simplicity of this idea boggles the mind

Something a lot of people seem to be lacking...

 

[alexf]

I was wondering whether these liquid-cooled machines will actually be quieter than the other models...

Also, is a newly-released Rev. B any noisier / quieter than a similarily configured Rev. A?

 

[g4cubed]

I was wondering whether these liquid-cooled machines will actually be quieter than the other models...

Also, is a newly-released Rev. B any noisier / quieter than a similarily configured Rev. A?

I'm not sure about the Rev.
But generally the water cooled are alot quieter because of less fans

 

[Chryx]

But the dude from Apple said (and this is where I have problem): that it is the higher density PowerPC 970FX chips that make it necessary to use the liquid cooling. Well...if all the new dualies have the 970FX chips (i.e. the density is the same because the chip design is the same -- same masks, etc), then why only liquid cooling for the top end? Speed (i.e. clock rate), not "chip density" (i.e. process shrink to 90nm) is "apparently" the source of the heat problem.

Was the Apple dude just talking jive marketspeak, or was he misquoted or paraphrased incorrectly? (I can't find the original quote from the links above).

Any chipheads with knowledge out there?
---MacUnit

yeah, the surface area is all the same, but the higher clocked one is going to put out more heat, perhaps quite a lot more than the 25% clockspeed rise indicates (someone at IBM claimed 50w typical for the 2.5 which is DOUBLE the 2.0Ghz 970FX)

So you have the same surface area, and twice the heat, you need to more aggressively get that away from the die, hence the watercooling.

 

[dongmin]

tje reason the density is such a problem is because you can longer cool the heat with a fan, its so hot in such a small area the air would never cool the center of the chip, just the edges. a liquid cooled device is able to pull heat off the entire chip.

but that would require far more common sense than they have...the simplicity of this idea boggles the mind

Well, not exactly, Mr. Common Sense.

The air does not directly "cool the center of the chip" in the Rev-A machines ,as you imply. It never has. Both use a system to move the heat from a small concentrated area (the chip) to a larger surface area (heat sink, radiator) which is then cooled by moving air over it via a fan. It's basically the same principle, except in the liquid cooling system, a mechanical pump actively moves the material around. I supposed it's more efficient than just using conduction via metal.

 

[Abstract]

Okay from a non-computer guy point of view, I can understand how MacUnit is confused, and I think I can explain it to him. If the "wires" (ie: die size)are smaller than before, and the amount of power that must flow through the chip is the same as before, then the amount of power flowing through every length of the wire (ie: every metre, cm, nm, whatever) at any given time has increased. This is what's causing the increase of heat.

My problem with even my own explanation (as well as theirs) is that I thought that the smaller the process became, the less heat that would be generated from these chips. If the first gen 2.0GHz chips didn't need liquid cooling at 130nm, then I expected that by going to 90nm would make a 2.5GHz much easier to do. I would understand if they found it difficult to make a 2.5GHz cpu using the 130nm process, but since they shrank the die size, it should have been quite easy to ramp the speed up a bit, right?

 

[jsw]

Yea, that radiator hanging off the back would make them a bit heavy

On a related note, anyone know how much weight the LCS adds? Apple's site says the G5s weigh 44.4 lb but then goes on to qualify that with "Weight varies by configuration and manufacturing process."

Interesting that they're so exact (to the tenth of a pound). An average? The lowest weight?

Anyway, my Rev A dual-2 G5 weighs so much already that I am afraid to add anything else inside of it for fear that it will collapse into a miniature black hole. I wonder how much additional weight the LCS adds.

Not a real problem. Just a curiosity. And perhaps an issue for those of you who will have to carry one with its built-in flesh-cleaving flat handles.

 

[jared_kipe]

Here is the thing that should satify most of your thoughts. The 90 nM chips do make less heat. Thats the point of the reduction of process size. However, they are also smaller, and thus harder to take heat away from. So while they do make less heat, it is hard to cool them. There is where the enigma is.

As for a G5 in a laptop, as stated before me, just because it has a large heatsink doesn't mean it won't work in a laptop. First of all, I don't know if you all have looked at the heatsinks on any of the late G4 towers, but they were very large. What makes them fit in laptops is #1 you don't use dual processors, #2 you have them clocked down a bit (in the begining), and #3 you use the select processors that will reach a desired speed at lower voltage than the desktop CPU. I used to over clock pentiums for people who needed new hardware but couldn't afford it. Every computer overclocks different, and the main thing that raised heat for me when I overclocked was increased voltage not MHz.

 

[g4cubed]

Every computer overclocks different, and the main thing that raised heat for me when I overclocked was increased voltage not MHz.

I don't know anything about the P4s or over clocking for that fact, but I was lead to believe increasing the voltage, increased the MHz Maybe just a misconception on my part, but agree about increase voltage increase heat.

 

[jakemikey]

Not a real problem. Just a curiosity. And perhaps an issue for those of you who will have to carry one with its built-in flesh-cleaving flat handles.

No kidding. I call my G5 "old ironsides". Those handles have to be the most ergonomically incorrect handles ever...and they KILL when you have to move it even a moderate distance.

 

[Bigheadache]

I don't know anything about the P4s or over clocking for that fact, but I was lead to believe increasing the voltage, increased the MHz

Thats not true. Sometimes you need to increase voltage to get your overclock stable, but its not always the case.

On my Athlon XP, I overclocked from 1.8Ghz to 2.2Ghz without increasing core voltage, but some others have had to tweak the core by 0.05v. Generally the more MHz you try to overclock, the more likely you will need more core voltage, but as with all overclocks, each case can be different. What jared said is correct.

 

[shawnce]

>shawnce

Ask this guy where he got the number for this post

Though it probably isn't too far off (or even a conservative number) -- it's the leakage that's killing us at 90nm.

Well it looks like he is guessing at this number...

... (maximum wattage of 102 - I'm assuming this figure from the typical)
... Let's say that the 970fx at 2.5ghz is the same as the 130nm part at 2.0ghz

It would be great to see some real documentation in regards to the 970 and 970FX parts. To bad IBM doesn't appear to put in any out at the moment.

However IBM does have a little quick reference guide that states the following...

Typical power 970:51W@1.8GHz 970FX:12.3W@1.4GHz, 24.5W@2.0GHz

So if you take the typical and double it as a bench mark for maximum you get the 970FX at 2GHz is burning about 49W. Then note that the 2GHz is 43% faster (clock wise) then 1.4GHz and the power difference is about 100% or 2x then... you can guess that the 970FX running at 2.5GHz would be burning around 74W maximum (1.5x the 2GHz part).

Anyway I am surprised that if the cores are the same that the active power dissipation isn't scaling linearly (the static dissipation should be similar for the same core at the same core voltage). I wonder if a core voltage difference exists between the 970FX running at 1.4GHz and the one running at 2.0GHz. If so that would explain it.. ah yes the 1.4GHz part runs at a core voltage of 1V while the 2GHz part runs with a voltage of 1.3V.

So factoring that the active component will vary linearly with clock rate but to square of the voltage we are looking at a delta of about 70% caused by the voltage.

Anyway that easily negates my 2x multiplier in the above so the 970FX running at 2.5GHz likely runs more closely to 25% more then the 2.0GHz (assuming the same core voltage) or in other words more like 62W.

... but who knows without some better numbers on the 970FX.