Apple Announces New Liquid-Cooled 2.5 GHz G5's.

[Sudhian Staff]

Apple Announces New Liquid-Cooled 2.5 GHz G5's.

[Sudhian Staff]

For those of you that haven't heard yet, Apple has announced its new 2.5 GHz G5 models.  Though these systems are quite a bit short of the 3 GHz Steve Jobs promised this time last year, a 500 MHz clock gain is actually fairly respectable when you compare it to both Intel and AMD.  AMD's Opteron has gone from 2 GHz to 2.4 GHz in the same time period, while the P4 went from 3.2 GHz to 3.4 GHz.  Not exactly a year for clockspeed. 

The truly stunning part about Apple's new G5's is that they are liquid-cooled.  That's right folks--the first computer manufacturer shipping liquid cooling standard isnt' Alienware or Falcon-Northwest, its Apple.  Apple, of course, has spun this as a move made to quiet systems, but it seems more likely that IBM yields on the G5 aren't nearly what they'd like them to be, and that this was a necessary move to ship volume.  When fans like the Thermaltake Silent Boost can be had at whisper-quiet noise levels, there's no need for liquid cooling on a CPU.

One thing to note.  Apple uses a 1,250 MHz bus that's 64-bits wide and, according to them, [supports] "up to 10 GB/s data throughput."  Apple's HyperTransport linkage system is 32-bits wide per channel, with simultaneous upstream / downstream capability.  Actual clockspeed on the bus is 625 MHz (1250 DDR).  This compares very well with AMD, who offers a 1 GHz (2 GHz DDR) bus at 16-bits, for a total of 8 GB/s of bandwidth per CPU.  L2 cache remains the same at 512K per chip.

Apple also seems to have dumepd legacy PCI in their new cases in favor of not PCI-Express but PCI-X.  One 133 MHz / 64-bit channel and two 100 MHz / 64-bit channels are now offered.  At roughly 1 GB/s of bandwidth, 133 MHz PCI-X is a much more powerful standard than PCI, but is corespondingly more expensive to implement--its slightly surprising to see Apple going this route.

If there's anything disapointing about these new Apple's its the RAM and video card capabilities.  $3000 may get you God's own CPU's, but they'll be data-starved--only 512 meg of RAM ships with the system at that price.  Another 512 costs you $150, a full 2 gig will run $450 total.

This monster system ships with only the 9600XT as default and selecting the 9800XT apparently leaves you unable to use the first PCI-X slot.  If its true that these boards only offer three slots, and that the first slot is the 133 MHz one (as it would usually be), this is quite a kick in the face.

A regular 160 GB SATA drive rounds everything out. 

The new Apple rigs may say nothing good about IBM yields, but they may be good enough to put Apple on the competition map.  Frankly, the x86 market has largely stagnated in terms of higher clock speeds while the G5 has leapt ahead, and while its VERY true that clockspeed isn't everything, it does count for SOMETHING.  The idea of dual 2.5 GHz G5's in real competition with dual systems from AMD or Intel has a validity that dual G4's running at a third the clockspeed with a tenth the bus speed never did.  Could Apple be on its way to regaining real legitimacy in the market?

Well, I'm not holding my breath.  But it could happen.

[Jamawass]

Excuse my ignorance but how are high operating cpu temps linked to yield?

[Joel]

Jama,

The link is correlative and works out like this.  (I'm going to generalize--so be forewarned).

First you need to understand that a Megahertz is not TRULY a measurement of speed, it is a measure of FREQUENCY.  In the computer industry, the two have been combined.  A "hertz" is defined as one cycle per second. A kilohertz is a thousand cycles per second, a megahertz is a million cycles per second.  Thus, a 2 GHz system is cycling two million times a second.

A "cycle" in this case is defined as the time needed for a transistor to change states--from open to closed, positive to negative, on to off, call it what you like--its a state-change.  Remember Intel's "double-pumped" Arithmetic Logic Units" in the P4?   That meant they ran at double clock speed, or double the number of cycles.  

If you need a simple analogy for why this generates heat, go to the wall and start flipping the light switch on and off rapidly, rub your hands together, use a hand or power sander on wood, or do any motion that involves back and forth movement.  What happens?  Energy is needed to create the motion, and this, in turn, creates friction--which is released as heat.  

Lets keep with the simple analogies for a moment.  If you create too much friction between your hands you'll stop rubbing them out of pain.  Using a sander too hard, too fast against the wood could theoretically catch it on fire through simple friction.  Flipping the light switch back and forth quickly enough could create enough energy to break the switch or melt the wiring.  The point, in general, is that when the moving components become too hot, they cease to function properly.  Why?  Because increasing heat increases resistance.

When the resistance becomes too high--when the transistor is no longer capable of moving from open-to-closed, or closed-to-open, it fails.  

Now, one way to overcome this is to increase voltage--"push" the electrical current harder, which will jump across the increased resistance.  Problem is, increasing voltage increases temperature, and increasing temperature....increases resistance.  Sooner or later, voltage pushes become useless.  

Conversely, LOWERING temperature DECREASES resistance, but on a similar scale.  Works like this:

Slap an OEM cooler on a CPU, you'll get a 10% overclock with an operating temperature of, lets say, 60'C.  Go out and buy a top-of-the-line, uber-expensive air cooler (at, say, $50) and you'll push 25% more speed out of that CPU and it'll be a little cooler even--lets say, 56'C.

But if you want that last 15%, you'll need a liquid nitrogen cooling system that runs the CPU at 0'C or less.  

The rule of thumb is this:  The closer you approach the theoretical maximum frequency a CPU is capable of operating at, the more cooling you'll need to run it at that frequency.  Since liquid cooling rigs are much harder to build and safely implement from an OEM perspective than a simple air cooler, you can bet Apple is paying a high price for this.  Original Pentium 3 on .18 micron was barely capable of 1.1 GHz reliably even with ice-water cooling.  Today you could take that same ice water cooling, slap it on a hypothetical .09 micron P3 (basically Dothan) and take that sucker up close to 3 GHz, more like as not.  The cooling didn't change--the CPU design and capabilities did.

Hopefully this explains it.  Keep in mind, though, that different CPU's have different operating temperatures.  An Athlon 64 is only specced for a maximum operating temperature of 70'C, and under use stays around 48-51.'C.  Prescott tops out at around 75'C and runs perfectly stable there.  So it all depends on what chip you've got as to what temperatures it can withstand.

 

[Jamawass]

I get your point that apple is essentially selling overclocked chips to get a 2.5 Ghz, but isn't yield the number of cpu's per wafer?

[BananaNutBread]

Quote
Originally posted by: Sudhian Staff
For those of you that haven't heard yet, Apple has announced its new 2.5 GHz G5 models.  ...
The truly stunning part about Apple's new G5's is that they are liquid-cooled.  That's right folks--the first computer manufacturer shipping liquid cooling standard isnt' Alienware or Falcon-Northwest, its Apple.  ...

I am shocked and dismayed by this statement, particularly coming from a Sudhian staffer. From what little is yet known about Apple's liquid cooling mechanism, from their website it looks like nothing more than heatpipe technology that Shuttle has been shipping for a couple of years now. How is it that Apple is the "first"??

While I do not mean to knock the Apple product at all, I simply do not agree that they are the "first"

[Joel]

Banana,

You raise a good point.  From Apple's diagrams, it LOOKS like a fair amount of fluid is flowing--but, upon closer examination, this may be nothing more than a standard heatpipe.  I'm subsequently less impressed.

Jama,

Yield can refer to several different things.  We can talk about "yields" which means the number of CPU's per wafer, or we can talk about yields at a certain speed grade.  There's a timely example of this with Prescott.

When Intel launched Prescott, they made a lot of noise about having completed a very quick transition to .09 micron, and 2.8E and 3.0E chips were easy to find.  3.2E's were scarce, and 3.4E's were non-existent.  Even reviewers didn't have them.  About six weeks later we got 3.4E's, and now (six months after launch) the 3.4E should be on the market.

What happened here was yield.  Intel may have transitioned their equipment to .09 micron quickly, but 3.4E yields weren't good.  Took time for that to change.  So IBM might be makign TONS of 2 GHz G5's, but 2.5's?  Lots harder.

Keep in mind, when you manufacturer a die you don't know how good it is until you test it.  AMD doesn't have a "2 GHz line" or a "2.4 GHz line."  They may have one area of a fab that is producing better, but its not like Toyota, which might have one line building Camry, one building LExus, one building Corolla, etc.  You take a core, you test it--adn that's what it is.  

[gotshuttle]

banananutbread,

are you saying that the apple liquid cooling is just a heatpipe like ones in shuttles? it sure looks like it but theres fluid in those pipes...
if thats not liquid cooling, what is?

[GTX]

Quote
Hip To Be Cool

Take it up a notch without losing your cool. The top-of-the-line Power Mac G5 with dual 2.5GHz processors squeezes outrageous performance into tight quarters. To cool down those steaming circuits, Apple designed a sophisticated liquid cooling system that takes off the heat without bumping up the noise. Mac OS X dynamically adjusts the flow of the fluid and the speed of the fans based on temperature.

The dual 2.5GHz Power Mac G5 features an innovative liquid cooling system that’s more efficient than a traditional heat sink. This system provides a continuous flow of thermally conductive fluid that transfers heat from the processors as they work harder. The heated fluid then flows through a radiant grille, where air passing over cooling fins returns the fluid to its original temperature.

Doesn't sound like a heatpipe implementation.

Cooligy was supposed to be working with  Intel, AMD and Apple on a liquid cooling solution.
 

[GTX]

Apple Liquefies G5

Quote
The Cupertino, Calif.,-based computer maker introduced the three new Power Macs Wednesday. The fastest features two 2.5 GHz processors and a 1.25 GHz per processor front side bus, and it is partially cooled by a combination of water and propylene glycol, a clear liquid used in automobile antifreeze.

[MTP]

I mentioned this here in the new apple section.

What bothers me is that if the artist rendering shown on the apple site is accurate to the flow and connection of the cooling system it is already set to fail.

Anybody who has run water cooling in a dual cpu system knows you must "Y" the cooling into the cpus so each cpu is fed roughly the same temperature cooling. From the Apple picture, it appears one cpu block feeds the other cp block.  Therefore the secondary cpu in the loop will always be way hotter than the first.  They need to feed each cpu at the same time and remove from each cpu at the same time.  
Plus with just a simple radiator and fan setup this will not be a truly cool running system.  I can not see it being much cooler than the fan system they had before.

I am very interested in seeing how it really works for them. I do see this as a possible RMA nightmare for them.
Plus I am curious to why they did not do this on all the new G5 systems annouced?

[TitaniumAngel]

I don't think that it will be a heatpipe solution. The standard G5 cooler already is a heatpipe solution. If my image hosting was still up I would post pictures.

[JhuFrank]

Quote
Originally posted by: MTP
I mentioned this here in the new apple section.



Plus I am curious to why they did not do this on all the new G5 systems annouced?

I think Apple has a tendency to reuse older architecture for their lowend models. Just swap out the processors, and voila! a higher revved G5. Only the top of the line get the new cooling systems.

[Ackatack]

A couple of notes:

1. The G5s have 4 PCI slots in them.
2. IBM helped out with the design of PCI-X which may have something to do with why Apple is now using it as their standard.
3. The cooling system that is displayed is either incomplete or incorrect.  If it were purely liquid-based and not a heatpipe, it would require a motor of some kind to drive the water cycle.  At least that's what my limited understanding of physics and thermodynamics allows me to reason out.

[Joel]

That's exactly why I figured it must NOT be water-block based.  I've seen some pretty elaborate water-cooling rigs, including one that had a full-size cooling tower built next to the PC--but every single one of them used some type of pump / motor / waterblock, etc.  But my waterblock knowledge is pretty limited....I know a fair amount about air cooling, which has always been what I used, but not so much about water.

Correct me if I'm wrong, but it WOULD be possible to cool the same CPU with one water stream, provided the liquid was cool enough--but--it seems like yes, your second CPU would always be hotter.  Though I know that in most dual rigs I've seen, first CPU is always 4-6' hotter than second CPU--not sure why.

Also, there are significant downsides to PCI-X.  While its a powerful standard, the reason that it stayed on the server / workstation was mainly cost.  The circuitry is difficult to build and the boards require additional controllers and traces--one for every PCI-X device, IIRC.  

There *is* a PCI-X 2.0 standard that allows for devices up to 533 MHz, for a total of 4.3 GB/s of bandwidth, but from what I've heard, its all but dead now that PCI-Express is coming.  A PCI-E x1 connector offers double standard PCI bandwidth in about a quarter the space.  Move to x4, and you've got bandwidth equal to that of a PCI-X / 133 slot--and a MUCH smaller connecter with much less trace-routing.

One other substantial disadvantage to these PCI-X systems?

If you put one 33 MHz PCI component in a PCI-X slot, it'll run--but it'll cause ALL the cards in the system to run at 33 MHz.  The benefits of that 133 MHz PCI-X slot?  Gone, at that point.

[MTP]

According to the info being released it uses a water/glycol mixture with the OS controlling the pump flow output and the speed of the fan.

Joel you are right, it is possible but never very smart.  When I used to water cool my duallies (PIII systems) my first system showed big temp changes between inline cooling and parallel cooling.

I always just beleived that even though the cpus are trying to share the wealth, cpu 1 generally still runs the mb and all the perips so it generally was a few degrees hotter.

Ack-  thats what I said in the other thread.  That rendering can not be correct.
Direct from the Apple site

Quote
Take it up a notch without losing your cool. The top-of-the-line Power Mac G5 with dual 2.5GHz processors squeezes outrageous performance into tight quarters. To cool down those steaming circuits, Apple designed a sophisticated liquid cooling system that takes off the heat without bumping up the noise. Mac OS X dynamically adjusts the flow of the fluid and the speed of the fans based on temperature.

[PaulS]

forgive me if what im about to say has already been said as everyone that's replied has said a bunch....

i was watching techtv's screensavers and they were talking about this new G5 and trumping steve jobs b/c he lied about 3g's ...anyways its apparently a closed-loop cooling system similar to watercooling but using an antifreeze type liquid to repel calcium buildups and algae.
there has to be a pump in the liquid circuit somewhere b/c "convection" per se just letting the hot liquid rise up and all would be too slow to get any heat exchange done with liquid thats not mercury (heatpipe)

[MTP]

Pauls-everybody uses glycol in their PC water setups and if they don't they should for all the reasons I already mentioned (thermal displacement, lubrication, lower resistivity).

Plus water cooling is always closed loop.  Closed loop refers to the fact in uses the same cooling fluid, vs a constant supply of new liquid coming in.

I thoguht you had played with some water cooling??

I was big into the "water-scene" a few years ago.  Even had some that keep the old Thunderbirds just above freezing with a full load on the cpu

Crazy fun.

[Joel]

MTP,

How'd you keep the water that cold?  And the whole idea of a heatpipe is basically that you can use liquid to cool a CPU if the liquid is in the pipe-system--but heatpipes don't actually have a water "flow" persay.  They DO, of course, in the literal sense, but not not in the sense that there's a volume of water literally flowing through the system.  So if Apple has enough water in that thing to actually have a flow, then its obviously water-cooled--

Which brings me back to my original two points.  #1.  This thing is going to be expensive.  #2.  Their obviously topping out hard on G5 yields ATM.  But compared to Intel and AMD, neither of whom have moved THIS much in the past 12 months, its a strong move--assuming they can ship it.  But I have to wonder....I've never heard of a CPU manufacturer speccing water cooling for a mainstream part, and yet if you think about it, that's what IBM has done (via Apple).  

[MTP]

Joel, the water was not really at those temps, the cpus were.  I used to use 120W peltiers powered by an external psu.
The peltier sat on the cpus and the water blocks were used to cool the hot side of the peltier (the colder you make the hot side of the peltier the colder the codl side gets).
I made the cpu sockets airtight with rtv and used dielectric grease in the holes and in the center of the socket.
All the 1.4G T-bird systems I used a 12" tall multipass radiator that had 2 high cfm 120mm fans blowing across it and a single flow 115vac pump to move the coolant. A good 60/40 mixture and I added water wetter to it.

I had to have the system run 2 different DC programs 24/7 (SETI and Genome@home) that way if one finished the other would pick up.  This had to be done to keep a full load on the cpu.  If the system idle'd for more than a minute or so the cpu the cpu temps went to about -10F which is about the time the system would freeze (mbm5 always ran as well).  This would in turn to the water block into a cube of ice.
OS was Mandrake 7.

That was my most extreme cooling system.  The air coming off the radiator was so warm it was uncomfortable to sit infront of the machine.

Yes I agree it can not be a heatpipe in the Apple.  Mainly because you do not use a pump and can not control the flow of the fluid.  Thermal dynamics controls the fluid movement and as you know as it heats it moves up the pipe to the cooling area by natural "heat rises" and as it is cooled it travels via wicking material in the tube back to the hot zone.

I think this is a strong move by Apple and really it would be very easy for the cpu manufactures to set water cooling standards for the cpus.  They already set them for air cooling so it will only be a matter of conversion to enusre proper cooling.

My concerns with the setup is from my past experience, condensation under the cpus, condensation on the piping, etc...
 

[Ackatack]

I'm not entirely convinced that Apple is going the water cooling route because they *have* to.  Watching Apple over the last couple of years and playing with a few G5/G4/G3s has led me to believe that this move is more geared towards trying to maintain the near silence of the G5s; afterall, Apple does boast that the "G5 is 40% quieter than the G4."  Apple could probably have used some high-CFM fans to cool these new G5 CPUs but those tend to be quite loud.  In any case, Apple usually does a great job of spinning something that's really not good into something that's attractive to most end-users.

[Connor]

Quote
Originally posted by: Ackatack
Apple usually does a great job of spinning something that's really not good into something that's attractive to most end-users.

Attractive is a good choice of words, because there's probably an element of looks in there as well, adding water cooling removes too many unsightly fans and replaces them with senic water hoses.

[darksynth]

Who is this Cooligy?  

[Joel]

Ack,

I disagree with you.  If Apple wanted a quieter cooling system, they could, in fact, use a standard heatpipe ala Shuttle.  There are a *lot* of ways to build a cooling system that's quiet and cool without resorting to water when you have the room of a mid-tower and a proprietary form factor.  In fact, I'd say water is *usually* not the most economical system for dealing with noise.  MTP can back that up or shoot it down.

You use water to get drastically lower temps--and the only reason I know to do THAT is to keep CPU's running at the brink.  The simplest way to test this, of course, would be to buy a dual G5 2.5 and slap on some really high end air cooling--see what happens.  Anyone got a spare $3K?

[MTP]

Quote
Anyone got a spare $3K?

Damn..Sorry Joel left my wallet in my other shorts

I agree Joel, water cooling does not really reduce fan noice because you still need the fans to cool the radiator to properly cool the water.  Sure the radiator could possible cool it without a fan, but I think you would end up with thermal run away evidentually because the water would never proper cool back down to a "resting state".

Quote
MTP can back that up or shoot it down

I am so mis-understood why does everyone always think I am being arguementative

Why isn't anyone commenting on the design in this thread?