Skip

OMG, it's a CPU not a sandwich!
June 6, 2006 7:32 PM   Subscribe

MacBook! Hold the Mayo! Apple's new MacBook is a hot item. At 65C internal running temperature, quite hot indeed. A MacNN reader discovered the MacBook service manual and shared the "Thermal Paste Mayo" pictures. Yikes! [more inside]
posted by cavalier (67 comments total)

 
I just found these pictures fascinating. For the non computer hardware enthusiast, the concept of thermal paste is simple enough.

[HEATSINK]
~~~~~~~
[---CPU----]

By layering a very thin layer of paste between the CPU and Heat Sink, you are looking to fill in all the little nooks and crannies -- pockets of air that are preventing the CPU and Heat Sink from touching each other. Used sparingly, the paste helps conduct the heat AWAY from the CPU and into the Heat Sink where it can further disperse the heat.

However, if these pictures are indeed bonafide then it begs the question... WTF? Using a significant amount of thermal paste likely turns it into a blanket, instead of a conduit. Apple's Legal team has already gone after Something Awful for posting a direct link to the Service Manual, but you can still find it if you google hard enough. DSL Reports forum with some more details.

While I was looking this up, I found this ArsTechnica journal which had some intersting points about the fan behavior. Apparently Apple was going for 'quiet' more than 'cool running', and accidentally unplugging the heat sensor ended up giving the writer a much cooler -- albiet louder -- MacBook. Maybe some firmware will help run the fans more often?
posted by cavalier at 7:32 PM on June 6, 2006


Once you put the heat sink on, all of the surplus will squeeze out the sides.
posted by betaray at 7:37 PM on June 6, 2006


That's still a crazy amount of thermal grease.
posted by Lord_Pall at 7:37 PM on June 6, 2006


Yeah, Apple screwed up. Maybe they'll fix it in their next revision.
posted by danb at 7:37 PM on June 6, 2006


What are those 2 cpu's? I assume the bottom one is the core duo. Is the top the gpu?
posted by Lord_Pall at 7:38 PM on June 6, 2006


The paste squeezing out the sides does not mean that the excess is gone and no longer a problem for the interface, there will still be an excess amount of paste between the CPU and the sink. The excess paste squeezed out from the sides could have a number of downstream consequences, restricting ambient airflow around the pcb, conducting heat to the PCB and occasionially making the interface from the processor to the sink less efficient than it otherwise would be without any paste at all.

Heatsink paste isn't like sex or free beer, you can have too much of it, especially if it's going everywhere but where you want it.
posted by iamabot at 7:46 PM on June 6, 2006


Sex and beer, on the other hand, always go exactly where I want them.
posted by nebulawindphone at 7:53 PM on June 6, 2006


one of the things I've learned from buiding computers is that they are *incredibly* sensitive to the thermal bridge between the CPU and the heatsink. I've had computers which would not POST boot fine *simply* because I pressed down on the heatsink, increasing the thermal transfer.
posted by unSane at 7:58 PM on June 6, 2006


It reminds me of toothpaste commercials where they always depict people piling a two-inch sundae on the brush. You only need an amount equivalent to a pencil eraser but they make a lot more money that way. But too much toothpaste doesn't completely defeat the purpose the way this does.
posted by George_Spiggott at 8:00 PM on June 6, 2006


Old news, and not really much different than this deleted post.
posted by b1tr0t at 8:00 PM on June 6, 2006


That tube is about 30-50 properly installed heat sinks worth of compound. That much compound is insane.

Remember, kids -- heat sink compound is an insulator. It's just much less of an insulator than air. Ideally, you'd lap the processor surface and the heat sink surface to such flatness that they'd stick together from the Casimir Effect. Not happening anytime soon.

So, we use heatsink compound to displace the air that would be in the space between the not-flat surfaces. Nnon moving air is a very effective insulator. This is why down coats are warm. However, you want to use just enough to displace the air, and no more -- because the thicker the layer of compound, the higher thermal resistance.

The best way to heatsink something is to bolt it to the heat sink, so the pressure compresses the gap as much as possible. They don't give us this option with CPUs and GPUs, so we're stuck with heat sink compound.
posted by eriko at 8:13 PM on June 6, 2006


It tastes like burning.
posted by fenriq at 8:16 PM on June 6, 2006


What b1tr0t said. Getting worked up over a stupid posed photograph is pretty stupid. Well done, you're all so much smarter than Apple.

Lord_Pall: The other chip is the system controller (aka northbridge and/or southbridge). It connects the CPU and memory to all the other parts of the computer. It's often a much more significant bit of engineering than the CPU or GPU.
posted by cillit bang at 8:29 PM on June 6, 2006



posted by Heywood Mogroot at 8:30 PM on June 6, 2006


Mini review: I got the lowest-end MB from amazon last week; I decided the plastic wasn't worth $150. 512MB simply isn't enough; matching DIMMs is important due to the lack of graphics card so I decided to go for 2GB and just solve the problem.

Keyboard is relative good, but only because all previous Apple laptop keyboards since Portable have sucked mightily.

Sadly, the same MBP buzz issue is audible on this guy; it goes away when I open Photobooth.

The machine runs hot, but not noticeably hotter than my now-retired 800Mhz PBG4.

Main point of dissatisfaction is that the fans make this weird cycling sound coming on for a second, going off, then coming back on for a second about 3 seconds later. HIGHLY annoying.

Sadly3, Apple doesn't have working XP drivers yet, but I assume this will be fixed.

Overall, worth $1000, but I suspect the Merom models will be quite good. This is kinda a tide-me-over untl they come out I guess.
posted by Heywood Mogroot at 8:38 PM on June 6, 2006


a little dab 'll do ya ... or

"what's this White Stuff , wee man?"
posted by celerystick at 8:59 PM on June 6, 2006


matching DIMMs is important due to the lack of graphics card

What happens if you have, say, a 1GB DIMM and a 512MB DIMM?
posted by Armitage Shanks at 9:57 PM on June 6, 2006


Chuck Norris ejaculates thermal paste.
posted by HTuttle at 10:13 PM on June 6, 2006


barefeats.com says asymmetric dual-channel is marginal for normal apps (just ~3% slower than matched pair(s)) but graphics apps take a bigger hit. Basically I want to keep the CPU satisfied since it i doing all the rendering now.
posted by Heywood Mogroot at 10:18 PM on June 6, 2006


I thought the thermal paste myth had been debunked for a few weeks now.
posted by Mr. Six at 11:37 PM on June 6, 2006


Very old news.

Overall, worth $1000, but I suspect the Merom models will be quite good. This is kinda a tide-me-over untl they come out I guess.

Yes. Instead of iPhoto opening in 2 seconds, it will open in 1. If only you can make it 'till then.
posted by justgary at 12:02 AM on June 7, 2006


My boyfriend boought one around a month & a half ago based on the reccomendation of his best friend, who is a rabid Mac fanatic. I am the resident geek half of our couplehood.

No whine but it does get very, very hot even placed on top of a jerry rigged piece of cardboard with 4 pieces of felt adhesive at the corners keeping it off the table. And the SO does not do processor intensive work. Max 4 windows open (email, safari, iTunes, & OpenOffice) and I could nicely fry an egg on either side of the keyboard. The latesest firmaware upgrade, contrary to others experiences in the Mac forums, did little to abate this problem.

I can fully admire an inch thick design while realising that heat dissapation is going to be an issue, but damn. I've warned him not to close the LCD screen unless its cooled down for at least 5 minutes.

I've built a grand total of two pcs of my own, so I'm no expert but that amount of thermal paste in the photos is a bit shocking. It might be old news, but I'd be interested to know if Apple is going to do something about the MBP's running hot/whining/what have you as I remain unimpressed with the tech support at the authorised Apple dealer where he bought it.

something similiarly posted at MeCha, but he'sI've done the firmware upgrade since
posted by romakimmy at 1:35 AM on June 7, 2006


I have a MacBook and I'm very happy with it. On maximum speed setting it does run hot, but not excessively so. On maximum battery setting it runs pretty cool. Don't be put off buying one of these babies because of the whingeing in various forums.

Controlling power and heat is the number one priority for chip designers now, IMHO. Just about any chip is fast enough for everyday use.
posted by salmacis at 1:45 AM on June 7, 2006


My ATI X850XT has managed to hit 90° CELSIUS on a bad day, when the drivers weren't telling it to turn its fan on, and I was playing BF2. Artifacting and good times all over.

That's *way* too hot.
posted by disillusioned at 2:44 AM on June 7, 2006


ok, ya know, i just don't buy any of this "too much thermal compound" shit that's been going around lately.

based on what fucking physics does some material that's a good thermal conductor suddenly become a poor thermal conductor when there's too much of it? what in the hell?

sure, maybe using too much is bad idea because it's wasteful, and can spill out and goop over the edges. but until someone can explain this new science of silver colloids where it somehow doesn't work because there's too much, i call BS.
posted by sergeant sandwich at 4:03 AM on June 7, 2006


Remove all plastic wrapping before use. Apple says that there is an over enthusiastic plastic wrapper installer that has places a possibly difficult to detect piece of sticky plastic over the vents. www.powerpage.org/archives/2006/06/the_apple_core_apple_comes_clean_on_macbook_vent_cover_up.html
posted by Gungho at 4:20 AM on June 7, 2006


If only your heat sink and cpu casing were somehow precision laser etched into perfectly flat surfaces, then they would have 100% direct contact and you wouldn't need thermal paste at all.

But despite what their mirror-smooth polish and a ruler will imply, they aren't perfectly flat. Thousands of microscopic air pockets will remain trapped between the two irregular surfaces, and air pockets are surprisingly good insulators.

Thermal paste is used to fill up these gaps and displace that air with some kind of metallic particle mix that at least conducts heat better than air does. But it's never going to come close to conducting as well as a direct contact would.

And that's what fucking physics the proper application of thermal paste is based on. Another solution would be to heat up the bottom of your heat sink to melting temperatures and compress it onto your cpu before cooling, but that's rather hard to undo later on.
posted by ceribus peribus at 4:26 AM on June 7, 2006


There's a discussion of this at MacInTouch. While I'm certainly not any kind of expert, my sense after reading various discussions of this issue is that people are incorrectly extrapolating from their experiences manually applying thermal grease. Apparently, in production, the heat sinks are attached by machinery that applies sufficient pressure and motion to ensure that any excess thermal grease squirts out the sides. I don't know if this excess can cause other problems, but it does sound like it's unlikely that too much thermal grease would remain between the chips and the heat sinks.
posted by klausness at 5:01 AM on June 7, 2006


based on what fucking physics does some material that's a good thermal conductor

Okay. Thermal Conductivity, in watts/mK

Copper: 401 watts/mK

Aluminum: 237 watts/mK

Artic Silver 5 metallic heat sink compound(in 25 μm
layer) : 20 watts/mK

Artic Alumina ceramic heat sink compound (in a 25 μm layer) : 4 watts/mK

Silcone/Zinc Oxide heat sink compound: .4 watts/mK

Air: 0.025 watts/mK

Now do you get it? .4, 4 or 20 is much greater than 0.025, but is much less than 237 or 405. You want as much metal to metal, or ceramic to metal, contact as you can possibly get. Heat sink compound is designed to replace the air in the spots that don't touch with some that's more conductive, but too think a layer means that the heat sink and die never touch at all -- instead, all of the contact is throught the heat sink compound.

Worse, as those layers get thicker, the conductivity drops, because the thermal resistance of heat sink compounds is much higher, thus, a thick layer of heat sink compound can acutally be worse than no compound at all.

Note the thickness of the layer -- 25 micrometers, or about one thousandth of a an inch.

And, note the manufacturers specs on this stuff. For example, Artic Silver 5 states that The 3.5 gram tube contains enough compound to cover at least 15 to 25 small CPU cores, or 6 to 10 large CPU cores. Total area, in a 75μm layer, 16 square inches, from one tube.

15-25. Not one, as Apple apparently used -- and that's assuming they're using Artic Silver 5, the most expensive compound I know of. Artic Alumina, which is what I use on most things, covers 48 square inches with 3.5 grams of compound, or (as I said before) 25 to 25 small CPU dies, with a few TO-220 heatsinks thrown in. Note that these are at .003 inch layers, not the .001 layer the specs call for -- you could triple the coverage if you could manage to spread the stuff just right.
posted by eriko at 5:18 AM on June 7, 2006


based on what fucking physics does some material that's a good thermal conductor

Okay. Thermal Conductivity, in watts/mK

Copper: 401 watts/mK

Aluminum: 237 watts/mK

Artic Silver 5 metallic heat sink compound(in 25 μm
layer) : 20 watts/mK

Artic Alumina ceramic heat sink compound (in a 25 μm layer) : 4 watts/mK

Silcone/Zinc Oxide heat sink compound: .4 watts/mK

Air: 0.025 watts/mK

Now do you get it? .4, 4 or 20 is much greater than 0.025, but is much less than 237 or 405. You want as much metal to metal, or ceramic to metal, contact as you can possibly get. Heat sink compound is designed to replace the air in the spots that don't touch with some that's more conductive, but too think a layer means that the heat sink and die never touch at all -- instead, all of the contact is throught the heat sink compound.

Worse, as those layers get thicker, the conductivity drops, because the thermal resistance of heat sink compounds is much higher, thus, a thick layer of heat sink compound can acutally be worse than no compound at all.

Note the thickness of the layer -- 25 micrometers, or about one thousandth of a an inch.

And, note the manufacturers specs on this stuff. For example, Artic Silver 5 states that The 3.5 gram tube contains enough compound to cover at least 15 to 25 small CPU cores, or 6 to 10 large CPU cores. Total area, in a 75μm layer, 16 square inches, from one tube.

15-25. Not one, as Apple apparently used -- and that's assuming they're using Artic Silver 5, the most expensive compound I know of. Artic Alumina, which is what I use on most things, covers 48 square inches with 3.5 grams of compound, or (as I said before) 25 to 25 small CPU dies, with a few TO-220 heatsinks thrown in. Note that these are at .003 inch layers, not the .001 layer the specs call for -- you could triple the coverage if you could manage to spread the stuff just right.
posted by eriko at 5:18 AM on June 7, 2006


eriko, wow, thanks!
posted by cavalier at 6:04 AM on June 7, 2006


Apparently, in production, the heat sinks are attached by machinery that applies sufficient pressure and motion to ensure that any excess thermal grease squirts out the sides.

Unfortunately, the picture linked in the FPP is from the service manual. In a service situation, the service technician would affix the heat sink with his or her bare hands.
posted by zsazsa at 6:08 AM on June 7, 2006


eriko, you're comparing apples and oranges. obviously it's not as good a conductor as a solid metal, but so what? clearly having a heat sink welded to the cpu would result in better heat conduction. big deal. this is a red herring in this context.

have a look at the units of the specs you pasted in - watts per meter per kelvin. think about what that means - i'll use the middle-of-the-range alumina compound as an example - it's 4 watts per meter per degree of temperature difference across the junction.

let's say you have a millimeter thick layer that's conformally coating both cpu and heatsink; the thermal conduction of the layer is then 4000 watts per 1 degree of temperature difference. if it's half a milllimeter (8kW/K) or 2 millimeter (2kW/K) doesn't really make a difference - your cpu isn't generating kilowatts of heat. further, i dunno what temperature a typical heatsink will be at, but surely it's more than a degree cooler than the core when the fan's on.

also, clearly there's an upper limit on the thickness that's imposed by the clip that holds down the heatsink. when you clip it down it comes out the sides if there's as much as shown in the apple pictures. so no matter how much goop you spray in there, you aren't going to get more than a mm or so between source and sink. (i'm also inclined to believe that it actually is better with more, since having enough compound in there to squeeze will ensure a nice tight fit when it's clipped down. but this is just hand waving.)

the point is, having a thicker layer of compound is no worse for conduction than having a thicker layer of any given material, if said material stays the same. saying " too much of this stuff will insulate your cpu because it's not as good a conductor as solid copper" is silly.

also, don't be so fucking condescending.
posted by sergeant sandwich at 6:15 AM on June 7, 2006


also, don't be so fucking condescending. .. Pot, calling Kettle, this is Pot! Over! Kettle, what color, over..
posted by cavalier at 6:42 AM on June 7, 2006


the point is, having a thicker layer of compound is no worse for conduction than having a thicker layer of any given material, if said material stays the same.

1) If the layer is thick enough to completely separate the die from the heatsink, you've lost a great deal of conductivity. I also didn't bring into play thermal resistance, because that's dependant on other factors, such as area.

2) Note that the compunds specificy the conductivity for a given thickness. There's a reason for that.

3) Theory, schmery, do the experiment. Don't use an older Athalon, it won't survive. It is trivial to show that a thick layer of heat sink compound significantly decreases the thermal conductivity between the heat sink and the die, thus raising the die temp. The number of CPUs and semiconductors I've seen killed by the "more is better approach" is scary. More is not better.

I've got years of experience, plus the spec sheets and installation instructions of every reputable manufacuture of temperature senstive semiconducters *and* the same info from the manufactures of heat sink compounds, and they all say the same thing: Thick layers of compound seriously impede the function of heat sinks.

You want to use only enough to fill the holes, and any more trades metal-to-metal contact for metal-to-compound contact. Properly used, the compound decreases the thermal resistance of the junction by replacing air, and only air, with compound, while not displacing any metal to metal contact. Too little, you get air gaps. Too much, you lose metal contact area.

Clips don't apply much pressure, and have a great deal of deflection. The right answer is a real amount of mechanical force -- bolts connecting the die and heat sink together, but CPU manufacturers don't give us that option. Many semiconductor/heat sink interfaces use no compound, because the compression caused by the strong mechanical bond puts enough metal in contact to drop the thermal resistance enough to allow the required heat transfer.
posted by eriko at 6:50 AM on June 7, 2006


This article from MacDevCenter suggest that the excess thermal paste isn't really the cause of the heat problem.
posted by pmbuko at 6:58 AM on June 7, 2006



posted by pmbuko at 7:04 AM on June 7, 2006


Thanks, pmbuko. I knew I had read that the thermal paste myth had been debunked somewhere.
posted by Mr. Six at 7:30 AM on June 7, 2006


If a thicker layer of compund is no worse for conduction, then why do the makers of Artic Silver only reccommend (emphasis mine):
The flatter the mating surfaces, the thinner the layer that is required. Stock processors and/or heatsinks with normal surface irregularities will require a layer 0.003" to 0.005 thick as shown below to fill the resultant gaps. (Equal to the thickness of about 1 sheet of standard weight paper.) Properly lapped heatsinks with mirror finishes will only require a translucent haze.
instead of saying "Hey buddy, just glop it on." Thicker layer=less suqare inches covered per tube=more sales, would it not?

From pmbuko's article:
So, maybe the reports of cooler laptops from people who replaced their thermal paste are really reports of laptops whose temperature control units are simply responding to a disconnected temperature sensor. Or maybe they really did do a better job than I did. I can neither prove nor disprove any of those statements. I can only report on my findings. However, since several internet reports of cooler laptops are accompanied by reports of louder fan noise, it's a strong possibility in my book.
I'd love to crack open the SO's MBP give a squizz at the thermal paste, but I think the relationship would probably end up being voided along with the warranty :D

Frankly, my guess is that the issue is multi-tiered, judging from the varietiey of 'Foo fixed it for me/Foo didn't work for me' sprinkled throughout the Internet. But a more detailed response from Apple would be nice.

That reminds me - does Apple often issue firmware upgrades without a list of what the upgrades actually fix?! No, really. Just 'Install this' ?!
posted by romakimmy at 7:54 AM on June 7, 2006


Although the article pmbuko links to is interesting, it's far from definitive.

On the plus side, the article suggests an alternative explanation for the fact that some people have seen cooler temperatures after disassembling their computers and reapplying the thermal paste. It is apparently fairly easy to accidentally disconnect the heat pipe sensor from the motherboard when you disassemble-and-reassemble your computer. If you do this, your system goes into emergency "Oh my God it's too hot mode!" and runs the fans constantly at full blast, resulting in a cooler-but-much-noisier computer. However, this has nothing to do with the change in the paste; you could get the same effect just by opening up your computer and deliberately disconnecting the sensor.

The article's main drawback is that the author didn't think to take the temperature of his computer before scraping off the paste. As a result, it is impossible to know if removing the paste actually had any effect! Alas, that makes this article much less useful, even as anecdotal evidence.
posted by yankeefog at 7:57 AM on June 7, 2006


The article's main drawback is that the author didn't think to take the temperature of his computer before scraping off the paste. As a result, it is impossible to know if removing the paste actually had any effect!

Even had he compared his laptop pre- and post-modification, it is possible that disassembly and reassembly have an effect on temperature, so that is not enough on its own either.

I understand your point, but he does compare his laptop with another stock laptop of the same build. Though the laptop belongs to his friend, it could be taken to be a randomly selected sample.

Assuming unmodified laptops have a certain mean temperature, if there is a significant effect from removing paste, he would have been more likely to see a marked temperature difference between his modified laptop and a randomly selected, unmodified laptop.
posted by Mr. Six at 8:12 AM on June 7, 2006


I also didn't bring into play thermal resistance

thermal conductivity and thermal resistivity are reciprocals of each other, just as are electrical conductivity and resistivity.

any more trades metal-to-metal contact for metal-to-compound contact

i think you've totally missed the point of my previous post, which was to point out that even a purely metal - compound - metal junction several millimeters thick will still have plenty of conductivity to not be a bottleneck in the heat transfer. yes, it's not as good as copper, but good enough that thickness up to several millimeters isn't a factor in these cases.

Pot, calling Kettle, this is Pot! Over!

ok, fair enough. my apologies for being jerky. i still think this concept is wrong though.
posted by sergeant sandwich at 8:39 AM on June 7, 2006


That's not thermal paste!

/obligatory
posted by mr_crash_davis at 8:43 AM on June 7, 2006


Mr. Six, check out the Intel Mac Temperature Database. CPU temperature variance between different stock MBPs is quite high.
posted by zsazsa at 9:00 AM on June 7, 2006


Having never opened up a computer I'm sure the idea of thermal paste is quite amusing to Mac users.
posted by wfrgms at 9:04 AM on June 7, 2006


Thanks crash. I now have a mental image of Steve (H.?) Jobs standing over the production line personally 'applying the paste'.

ew.
posted by romakimmy at 9:05 AM on June 7, 2006


If you think about an electric circuit, a resistor causes a voltage drop across it. Voltage is analogous to temperature (both measure a potential energy difference), and the thermal compound acts as a resistor. In a situation like this, you want to minimize the temperature drop across the joint. According to this article and this article thermal resistance is directly proportional to joint thickness. Twice as much thickness doubles the resistance.

Now, is this resistance substantial enough to cause a large temperature difference between the CPU die and the heat pipe inside the MacBooks? To know, we'd need to find out what kind of paste Apple is using (most likely standard silicone/zinc oxide) and measure the thickness of the default application of compound. Unfortunately, I think only the folks like those who write for Electronics Cooling Magazine have the equipment to determine that positively. Which then leads us to anecdotes and gut instincts. According to years of anecdotal evidence from overclocker and tweaker types around the world, yes, it is substantial. According to many other's gut instincts, the layer should be thin enough given the pressure of the heatsink.

While the first article I linked to dealt with phase-change thermal interfaces, rather than grease, it displayed a definite correlation between both viscosity of the interface and mounting pressure to thermal resistance, as with lower viscosity and rather substantial mounting pressure, the interface can flow out from between the device and the heat sink. I'd like to see how the viscosities of those phase-change interfaces compare to standard thermal paste.

Me, I've got a hot-running, mooing MacBook and a fresh tube of Arctic Silver. I also have a warranty. I think I'll sit tight for now.
posted by zsazsa at 9:42 AM on June 7, 2006


Just to clarify, it's not that it's easy to accidentally disconnect the heat sensor from the mobo (the connector holds it pretty well and you have to remove it to get the mobo out), it's that it's easy to forget to reattach it. When I re-did the paste on mine (I don't think it made a difference...) I almost made this mistake.
posted by rbs at 9:47 AM on June 7, 2006


CPU temperature variance between different stock MBPs is quite high.

Are they running under the same CPU usage conditions when temperatures are measured? I don't doubt you, I'm just curious.
posted by Mr. Six at 10:05 AM on June 7, 2006


ergo proxy
posted by bobo123 at 11:06 AM on June 7, 2006


watts per meter per kelvin

This is an incomplete view, and might be the source of confusion here. Thermal conductivity is (watts * distance) / (area * degrees). So the heat conducted varies inversely with the thickness of the layer.

A 5cm square junction covered with a 1mm layer of 4watts/mK compound, assuming the compound stays effective at that thickness, would conduct only 10 Watts per degree Kelvin instead of 400W*K for a 25μm layer. I'm not really sure what a typical steady state temperature difference would be across the junction.
posted by ceribus peribus at 12:09 PM on June 7, 2006


sergeant sandwich: ...let's say you have a millimeter thick layer that's conformally coating both cpu and heatsink; the thermal conduction of the layer is then 4000 watts per 1 degree of temperature difference.

This can't be right. What happened to the junction's area? If area didn't matter, then even a microscopic heat-sink could dispose of 4000 watts.

I see ceribus peribus has beaten me to the punch with the math, so I'll stop there. I'll only add that 5cm x 5cm is pretty large, as chips go - the chips in question in the MacBook look a lot smaller. That could conceivably make the MacBook situation considerably worse. A 1.5cm x 1.5cm chip has less than a tenth the surface area of a 5cm x 5cm one.

It's also worth pointing out that the message-board thread that started the ruckus included before/after case-temperature measurements, showing a significant improvement from thinning the compound.
posted by Western Infidels at 12:27 PM on June 7, 2006


My 1.67 Ghz PowerPC G4 is running at 26°C
posted by Mean Mr. Bucket at 12:44 PM on June 7, 2006


For materials which are uniform down to molecular scales of length, such as solid silver (assuming glass or single crystal ordering), thermal conductivity does not change with film thickness.

For colloids, such as Arctic Silver 5, the situation is drastically different. As the film thickness approaches the diameter of the suspended silver particles, which ought to be as constant as you can make it, thermal conductivity rises sharply. That's the explanation for the specs eriko quoted, and the reason for using these sophisticated and hard to manufacture compounds in the first place.

Sergeant sandwich, I disagree with your calculation. Say your CPU is running at 10 watts and has a footprint on the heat sink of 2cmX3cm, or .0006 sq. m. Then the heat flow you must cope with is Watts/Area, or 25000 watts per sq. meter. Plugging this value and the Arctic Alumina k of 4, along with the film thickness of 1mm you posit, back into the thermal conductivity equation, I get a temperature difference between CPU and heat sink of 6.25 deg. C. With 2mm, the difference is 12.5 deg. C.
posted by jamjam at 1:57 PM on June 7, 2006


Having never opened up a computer I'm sure the idea of thermal paste is quite amusing to Mac users.

Ever seen a G4 Tower? Why there's a little latch and hinge system that lets you oper 'er right up with a pinky. In fact going back to the CX's Macs have always been an order of magnatude easier to crack open than a PC. I work on both and I relish the ease of getting into a Mac...
posted by tatnasty at 3:12 PM on June 7, 2006


For those dissing toothpaste as a thermal compound, check out Dan's review of goops, including toothpaste and (that old standby) -- VEGEMITE!
posted by coriolisdave at 4:48 PM on June 7, 2006


people... remain calm :)

one thing to keep in mind: the arctic silver tech notes are about installing heatsinks on desktop PCs. anyone who has ever assembled a PC knows that the heatsink pushes down on the die with an absolutely ridiculous amount of force.

i'm just guessing here, but i dont think such a force is possible in the macbooks, because there is no place to put a huge lever. as a consequence, one might need a bit more thermal paste than you would use in the desktop PC situation. i agree that they appear to be using too much, but i think that a microscopically thin layer might not be enough in the case of a laptop. people need to be tearing apart dells, etc. to see how its done there to make a fair comparison.

i think the other thing, which has been noted above, is that apple (steve jobs in particular) assigns a huge premium to quiet computers. i noticed that when the processors are heavily loaded, the first thing they try is scaling down the processor frequency. if the load is still around after a few seconds of that, then they finally start the fan. and when the fan gets going, man, is it ever loud.

i'll bet that a dell or toshiba with core duos runs the fan all the time, and is probably thicker such that they have a vane-style heatsink which they are pushing air over (and out the back of the case) all the time. the macbooks really only have heat spreaders; there is no traditional heatsink in there.
posted by joeblough at 7:49 PM on June 7, 2006


i'm just guessing here, but i dont think such a force is possible in the macbooks, because there is no place to put a huge lever.

Bad guess, you don't need a lever to maintain pressure, just to initiate it. You could obtain the same pressure with a screw.
posted by b1tr0t at 8:16 PM on June 7, 2006


i'll bet that a dell or toshiba with core duos runs the fan all the time, and is probably thicker such that they have a vane-style heatsink which they are pushing air over (and out the back of the case) all the time. the macbooks really only have heat spreaders; there is no traditional heatsink in there.

Also likely wrong. My 1" thin TiBook had plenty of room for vane-style heatsinks. It accomplished this with heat piping.
posted by b1tr0t at 8:18 PM on June 7, 2006


well i'm just full of wrong stuff today.

the macbooks are even thinner than the tibook. regardless, what i'm talking about are heatsinks with a whole boatload of vanes, which i can see exposed in the back of all the dell laptops at work. the tibook does not have these. the more surface area the heatsink has the more efficient it's going to be at removing heat, and the "dell-style" heatsinks are definitely better than the flat ones that apple has going on.

this is what is inside a TiBook. you can see the fan tucked under there. what you are looking at is the main heatsink. yes its a heatsink, but i was talking about is something more like this (crappy picture, but tall vane-style heatsink is right in the middle of the image, and there is a fan right behind it.

and your rebuttle about the screws is wrong too, but that's my fault - the lever itself isnt all that important. what the desktop PC has that the laptop doesnt is actually an enormous clip, which functions as a spring, and puts the force right over the center of the die. screws around the edge of the heatsink will not necessarily apply as much force to the die, since the heatsink will deform. if screws were all that were needed, don't you think AMD and intel would just have you screw down the heatsink instead of engineering such elaborate sockets and heatsink clips? it would certainly be cheaper, and cheap is what its all about in the PC market.
posted by joeblough at 11:25 PM on June 7, 2006


the macbooks are even thinner than the tibook.

Not according to apple: Perhaps the most profound design element of MacBook is what it lacks: bulk. At 1.08 inches thin — more than a quarter of an inch thinner than the iBook G4...

While my 1GHz TiBook was specified at 1.0".

The problem with screws in a desktop application is that end-users can't be expected to (1) own a torque wrench and (2) know how to use it properly. You need both in order to implement a screw-based pressure system. Such a system could be acceptable to engineer into a notebook production line.
posted by b1tr0t at 11:41 PM on June 7, 2006


if screws were all that were needed, don't you think AMD and intel would just have you screw down the heatsink instead of engineering such elaborate sockets and heatsink clips?

If you assume that heat is the only problem, they'd *love* to use screws. The problem -- you have to screw on the heatsink without mangling the pins. Problem two: Screws cost money. Threading holes or bolts cost money. Screwing heatsinks to CPUS would take time, that takes money.

The clips and levers look like they're applying a lot of force, but when you take that force and apply it across the surface of the processor, you'll see that it is isn't that much. A screw can generate an amazing amount of pressure -- a single M6 screw can provide a clamp pressure over 1700 PSI. The limiting factor here is the strength of the screw and the nut or thread holding it, so you'd really want to use more than one.

Given that the surface of a P4 is around 2.25 inches square, if the heat sink and die are rigid enough, you could easily get around 500psi across the whole surface with one M6 screw tightend enough. Four of them would let you easily get 1000psi with very little effort, and 2000psi if you worked at it a bit. At this level of clamp force, you might not even need heat sink compound.

But it takes almost no time or talent to gob on a bit of compound and shove a heat sink on. Indeed, time costs so much that it is worth spending hours designing a clip system that will let a work install the heat sink in seconds. Time is money, and talented engineers realize that when they design systems like this.

Now, if the situtation was that clips just couldn't reliably do the job, they'd spend the time and use screws, despite the cost in parts and assembly time.
posted by eriko at 6:12 AM on June 8, 2006


screws around the edge of the heatsink will not necessarily apply as much force to the die, since the heatsink will deform.

Done correctly, that's a huge win. If the heat sink deforms so that it comes into closer contact with the die, that's more heat transfer.
posted by eriko at 6:15 AM on June 8, 2006


the macbooks really only have heat spreaders; there is no traditional heatsink in there.

What? No. What goes on top of the chips is a heat pipe which directs the heat to a real heatsink, which you can see in this picture and also the step 2 picture linked in the FPP. You can see the heatsink of the MBP in this picture.

These heatsinks are larger than that in my old Dell with a Pentium M 1.6GHz, which had a TDP not much lower than the Core Duos in the Macs. The Dell had the exact same heatpipe arrangement, and the heatsink was also screwed down on top of the CPU and chipset.
posted by zsazsa at 6:43 AM on June 8, 2006


i stand corrected on the macbook heatsink. i hadnt looked at that picture before. i guess the real problem boils down to "we don't want to run the fan until we really have to". just a tiny bit of airflow over any heatsink is infinitely better than none, which as far as i can tell is what the MB and MPB are doing until things get serious. i mean, as i pointed out above, they even try scaling the frequency/voltage before starting the fan!

on the topic of screws vs. springs, agreed that there is a user factor, and simplicity is also important. i dont think screwing down the heatsink would mangle the pins, since the heatsink would be connected to the socket/motherboard rather than the device. unless you mean that you might end up torquing the heatsink and twisting the part in its socket, but that can be taken care of with some mechanicals, like large plastic pins.

this is the most intelligent discussion of this problem that i have seen anywhere on the web. the thing that bugs me is that all these fanboys in random forums assume that they know more than apple's entire hardware team, which is ridiculous. if in fact the heatsink plates are able to apply 1000s of PSI onto the die, then it almost doesnt matter how much compound you put on - the majority of it is going to be squeezed off the side of the die as pressure is applied to the heatsink.

if the cooling system in the macbook was truly inadequate, then we'd have loads of fried macbooks around, and that doesnt seem to be the case. don't forget that in the end, the core duo dissipates something like 30W at full load, which really isnt that much power compared to a desktop processor. as far as i can tell, this 2GHz core duo has better performance than my athlon x2 4200+, which probably burns about 60-70W when its running full out on the same kind of job.
posted by joeblough at 9:58 AM on June 8, 2006


also, on the topic of deformation, yes if it is a soft metal and you end up with a die-shaped indentation in it, that is good.

i was thinking more along the lines of a stiff metal bowing up in middle and only coming into strong contact with the corners of the die, since all the forces applied are far to the edges of the heatsink plate itself.
posted by joeblough at 10:00 AM on June 8, 2006


This can't be right. What happened to the junction's area?

ah, yes. i'd forgotten the area. you guys are right. i stand corrected, and apologize for my snotty attitude. i think i was feeling cranky yesterday.
posted by sergeant sandwich at 2:17 PM on June 8, 2006


« Older History   |   Finders keepers? Newer »


This thread has been archived and is closed to new comments



Post