Many modern heatsinks seem to be fundamentally flawed in design

[will_maltby]

:eek: Basically, IMHO, almost all of the modern heatsinks seem to be fundamentally flawed in design. Why ?


Well, their fins are usually WAY too thin.


Take this example. You're standing by your gas cooker with flame burning. In one hand you have a sewing needle, in the other you've an allan key. Right, put them both in, what happens ?

Well, the sewing needle goes red hot at the tip very quickly, but that's the only place that it's hot. Your other hand starts burning 'cos the allan key has conducted the heat much more greatly.


I think this is why the swiftech sinks do so much better than other sinks since they have thick enough 'fins' (albeit pins/sticks). All these new sinks with an impressive array of ultra thin slides of copper are probably not very good and would be better if they had less but thicker fins.

I notice that even on my swiftech MCX370, the top of the pins are cool whilst the bottom is warm. So even on that thickness, there's a dissonance in heat along the pin.


Is copper really that good a conducter that it works very well in ultra thin slices ? I don't know, but if it's anything like a sewing needle, then thin fins on a heat sink is a bad design.

[WillysNut]

There are a lot of factors that need consideration here. The MCX 370 is a huge overall design, with much more mass and area then the "average" thin fin heat sink. The design has a copper base with alumnium pins. Alumnium sheds heat faster than copper. Most ultra-thin-fin designs are for all-copper HSF's

If you compare a thin fin, all copper heatsink with a non thin-fin, all copper heat sink, the thin fin design performs better because the thickness, or rather thinness, of the fin size allows the heat to be disappated at a faster rate. Copper is better at alumnium then absorbing heat but not at shedding it.

Example. The CAK38 vs. the CAKII-38.

[teezer]

ENTER THE REASONING OF INSTALLING A FAN, IN THE EARLY DAYS ~ BIG AND THICK HEATSINKS AND NO FANS ~~~ WOULD THE THICKNESS OF THE FINS SUGGEST WHETHER THE FAN WOULD PUSH OR PULL? ~~~~ WHICH IS BETTER? PUSH OR PULLING OUT THE AIR? ~~~~~~~~ ONE THING I'VE NOTICED IS THE EXCESSIVE AMOUNT OF VIBRATION IN @5000+ FANS ~~~~~~~ DOES ANYBODY BLUEPRINT OR BALANCE FANS?

[vandersl]

Its all a matter of balance.

Predominantly, heat transfer through a heatsink is via conduction.
Both the thermal conductivity and thickness of the HS material affect the transfer of heat within the metal itself. The thicker the metal, the more heat that can be transferred for a given temperature differential. It's much like current in an electrical conductor.

Transferring heat through the heatsink is only a part of the problem. Transferring heat from the heatsink to the air is another part. Here, the predominant mechanisms are conduction (from the HS surface to the air in the direct vicinity of the HS) and convection (moving the air away from the surface).

Conduction in air is many times lower than that in metals like copper. The only way to really compensate is to increase the surface area. Technically, one could probably calculate the optimal ratio of metal cross area to surface area to balance the heat flow rates. Apparently, HS manufacturers are, and the result is thinner fins and greater surface area.

Note that copper conducts heat better than aluminum, and is also more expensive. So the manufacturers have more of an interest in optimizing the ratio than previously.

[The Overclocker]

eh? what is the thinking behind this? if people throught like this then strapping a large copper block on your cpu is good. it is not. a heatsink purpose is to increase the surface area so enough heat can be dissapated to maintain a chip/transistor/anything else at a certain temprature. The reason swiftec heatsinks work well is because they use round pins which offer the most surface area.

[AKDUDE]

I find it interesting that the Glaciator 2 seems to work so well with the thicker fins, at a reduced surface area.
AKDUDE

[Sonny]

Originally posted by the overclocker
.....round pins which offer the most surface area. I need some clarification:) If we take a look at a section of a square & circle pin we get;

Square = 4(# of side) X 4mm(legth of each side) = 16mm
Circle = 3.14 X 4mm diameter = 12.56mm

Granted these are large # but that doesnt matter.

[res0r9lm]

Originally posted by sonny
I need some clarification:) If we take a look at a section of a square & circle pin we get;

Square = 4(# of side) X 4mm(legth of each side) = 16mm
Circle = 3.14 X 4mm diameter = 12.56mm

Granted these are large # but that doesnt matter.
as long as you are talking about square pin and not fins yes the square ones would have more surface area but would be difficult to manufacture. if you are looking for a shape that would give you the best surface area star shaped pin would be better but harder to implement than square ones

[jboy]

It seems to be that for all the chat that takes place on these forums, no one's done the fundemental home work that'll answer a couple of questions I have. For instance, 1.- why arent heatsinks tested without fans, so you can get a feel for their heat-wicking potential ? And, 2.-why hasn't someone figured what approximate percentage of cooling comes from the fan. I suspect the fan is much more important in the equation than the sink. For example, I swapped out the Delta fan that came w/my Alpha 6035 with the puny one that came w/my AMD retail sink, and, the AMD/Delta combo consistently cools better than the Delta,-(Tried this 3 times).Yes, I know I'm using my crappy on-board MB sensor to measure these differences, but they do seem to be consistent and repeatable, so I therefore have to accept them as valid. It seems that most effective heatsinks simply get their ability with large surface area /mass, and brute force fans blowing lots of air onto them. Where's the art in that ? I can imagine a HS/Fan combo, a tuned system, that would work in concert better than either of the components would suggest, but I dont see this on any of the current designs.

[The Overclocker]

heatsinks arn't tested withoput fans because:

1. the cpu would fry

2. the heatsink is not designed to work without a fan, the fins are too densly packed to work without a heatsink and some heatsinks (thermo engine) are designed around a fan.

a fan is a must, because of air's poor thermal properties hot air must be moved away from the heatsink. without a fan the air around the heatsink would get hotter and hotter

[neo86]

Ummmm... errr... I think companies have a lot more money to throw around researching this kind of stuff. They hire people with a degree in this stuff. The people they hired probably don't have to dust off a book to know anything about thermodynamics.
That's why I think you should leave these things to people whose job it is to design heatsinks.


But here's my thought:

Suppose you have a block of metal on the cpu. It won't take long to heat up, but to get rid of the heat from the block it'll take a while.

Now instead of a solid block put a base plate with a bunch of pins in it. The pins will absorb the same amount of heat at a time, but will cool down wayyy quicker, because the heat doesn't have to travel far to get to the edge of the needle where the air will absorb it. Now it can absorb more heat because it's cooler.
Cooler things absorb heat quicker than warm things...

Basically, it's not just about absorbing heat, but getting rid of it as well.

[will_maltby]

I'm not saying a big block of copper is best, it certainly isn't.

But I think there is an optimimal thickness of fin/pin with respect to material. True, the fan blows down a thin fin so it doesn't just cool from one end, BUT; if the fin is too thin, then there's little point in having it very high/long since it won't conduct the heat efficiently.

And I don't think we should immediately assume that heat sink manufacturers necessarily know the true ins and outs.

Swiftech and the alpha sinks are reknowned for being the best and they use thick pins.


If ultra thin is best, why haven't we seen a sink with an incredible array of sewing needle sized pins on a base ?

[Malakai]

i agree, my glaciator 2 has 21 fins a lttle thicker than an allan key, and a "slow" and quiet 5800rpm fan, and it works almost as good as swifty's with 8000rpm delta screamer's. better conduction through thicker metal slices i guess.
-Malakai

[teezer]

how come industrial servers, which are on 24/7 use only heatsinks, no fans(no fan failure)?? ~~~~ these have some major heatsinks ~~~~ after testing a few various hsf setups, the air from the fan is always pushed thru the hs (pushing hot air from the tip down?) ~~~~ thusly, wouldn't it be better to enclose an oversize, thin or thick hs, and extract the air from the hs (the heat at the tip leaves first) using a proportional size fan, tube diameter, and length??? ~~~~ gets fun trying to find that combination, like a custom exhaust header that you speed the air out faster by the size of the tube and special length and bigger is not always better

[{PMS}fishy]

Originally posted by teezer
how come industrial servers, which are on 24/7 use only heatsinks, no fans(no fan failure)?? ~~~~ these have some major heatsinks

Cause servers aren't overclocked:D

[Sonny]

Originally posted by jboy
1.- why arent heatsinks tested without fans, so you can get a feel for their heat-wicking potential?Who says they arent running a whole number of test that includes one with out a fan. They do not include this in their documentation the design will have been optimized for use with a fan. The manufacturers of the higher quality HSFs do their job in RD or we wouldnt get better performing components.Originally posted by jboy
2.-why hasn't someone figured what approximate percentage of cooling comes from the fan. It seems that most effective heatsinks simply get their ability with large surface area /mass, and brute force fans blowing lots of air onto them. Where's the art in that ?Have you tired reading the frontpage of this site that have a few articles about airflow thru the HSF & case? You can never rule out the case airflow. When you start to go for extreme cooling on air finesse is not always a consideration since systems are all different & you will need a lot of headroom to have consistent performance.

I challenge your theory on the fans with both set up. Send them to JoeC. & have him do some tests on it. You talk about them being repeatable but you never mention the fact that there could be another cooling path towards the thermistor that can influence temp monitoring. How different are the designs of both HSF combinations & was the case optimized for their specific requirements? Why do you have to optimize? Well it's because different designs require different airflow. Test each at there best if you want accurate numbers.

[soil]

Originally posted by the overclocker
heatsinks arn't tested withoput fans because:

1. the cpu would fry

2. the heatsink is not designed to work without a fan, the fins are too densly packed to work without a heatsink and some heatsinks (thermo engine) are designed around a fan.

a fan is a must, because of air's poor thermal properties hot air must be moved away from the heatsink. without a fan the air around the heatsink would get hotter and hotter

comparison based on the same fan, or same model at least...?

[JigPu]

Originally posted by WillysNut
Alumnium sheds heat faster than copper
Yet annother on the side of aluminum shedding heat faster than copper :rolleyes: ... The idea that aluminum gives up it's heat faster than copper is a long standing myth started by some village idiot. There have been several posts about this idea, and all the physics guys here agree. Aluminum dosen't do that. I started a post just after I joined called "Copper vs. Alumimum". This is where I first learned that my poor belief was wrong...

Now.. About vandersl's statement. From what I read, he is not saying that a big ol' block o' copper will solve all your problems. Basicly, what you want in the end is to get rid of the heat as fast as possible. To do this, you want maximum USED surface area. The farther you get from the heat source, the cooler it will be. This cooler area will not be doing it's job effectivley. In order to maximize the ammount of metal that will warm up, you want to make your pins thicker so that the heat transfers up to the colder regions easier. However, doing this will invariably decrease the surface area that you have to get rid of the heat. So a balance must be struck between utilizing ALL of the heatsink (verticaly) and actually getting the heat off of the sink when it gets there.

This feels like a rant too me... but I never turned rant mode on! :p
JigPu

[res0r9lm]

Originally posted by JigPu

Yet annother on the side of aluminum shedding heat faster than copper :rolleyes: ... The idea that aluminum gives up it's heat faster than copper is a long standing myth started by some village idiot. There have been several posts about this idea, and all the physics guys here agree. Aluminum dosen't do that. I started a post just after I joined called "Copper vs. Alumimum". This is where I first learned that my poor belief was wrong...

Now.. About vandersl's statement. From what I read, he is not saying that a big ol' block o' copper will solve all your problems. Basicly, what you want in the end is to get rid of the heat as fast as possible. To do this, you want maximum USED surface area. The farther you get from the heat source, the cooler it will be. This cooler area will not be doing it's job effectivley. In order to maximize the ammount of metal that will warm up, you want to make your pins thicker so that the heat transfers up to the colder regions easier. However, doing this will invariably decrease the surface area that you have to get rid of the heat. So a balance must be struck between utilizing ALL of the heatsink (verticaly) and actually getting the heat off of the sink when it gets there.

This feels like a rant too me... but I never turned rant mode on! :p
JigPu
I think more people know that copper dissapates better than aluminum but when you tell some that it would take a force of nature to convince them. so people just don't say anything.

[neo86]

Maybe somebody should try the pin idea!:beer:

You won't know till you try.:burn:

[Brant]

Alluminum disipates heat faster that copper because in general it has a rough texture which gives a greater surface area. look at a aluminum sink vs copper comparing the texture.

[errorz]

Aluminum is also a lot LESS DENSE then copper...

example: your gotta ITTY BITT sponge, verses a BIG ONE... itty bitt can only handle so much... where as the big one sucks a ton.... now squeeze then... the itty bitty one looses all its water with a quick squeeze, whereas the big one takes both hands... and sitting on it.

Copper vs. aluminum is the same way. Sure when you turn off your computer your aluminum sinc will be cooler quicker, but it doesn't matter.

turning on a computer, your aluminum would suck the heat up real quick.... thus giving the temps a stable jump in temp... copper may take a while to suck up all the heat, but creates a nice slow increase in temps... they will not be as jumpy

[donny_paycheck]

Originally posted by jboy
It seems that most effective heatsinks simply get their ability with large surface area /mass, and brute force fans blowing lots of air onto them. Where's the art in that ?

That's basically the definition of an effective heat sink.

[Brant]

Errorz, I am not sceculating that copper or aluminum can hold more heat. Or which one conducts heat better. All i said is that aluminum transfers heat better to the air because it generally has more surface area.

Your sponge analogy really sucks.

The specific heat of Al is .9 J/g*K, Cu is .379 J/g*K Meaning Al Holds more heat per gram. However Al holds 2.43 J/cm^3 and Cu holds 3.43 J/cm^3

The amount of heat that the material can hold doesnt matter. It has nothing to do with how well the material disipates heat. The purpose of a heat sink it to disipate heat. Storing heat is not disipating heat.

[errorz]

I was simply explaining Brants idea alittle moe thouroughly. I have seen both aluminum and copper... I notice littel tecture differences. Anywas it would be SOOOO little it wouldnt affect disipation greatly.

[Brant]

The differnece is little but that little diference could double the surface area.

[JigPu]

Not to sound stupid here, but couldn't you rough up the surface of copper to increase the surface area?? Also, just how is aluminum rougher than copper (in general)? In my brain, one metal would be just as rough as annother (in general) when they were cut... In the microscopic scale, they would have been cut the same way, and the only diffrence I can think of would be that the electrons of the aluminum would be in a diffrent arrangement than the electons of copper :rolleyes: . Doesn't seem like there is room for much diffrence to me...

JigPu

[res0r9lm]

the aluminum alloy used to make forged pistons in a car is supposed to be better than copper from what I heard but is to hard to machine (wear & tear on equipment)

[Patchmaster]

I hope a moderator deletes this entire thread pretty soon before some unsuspecting overclocker comes along and tries to use the vast quantity of misinformation contained herein.

[teezer]

can you stack heatsinks? ~~~ use square or round aluminum heatsink to a slab ending in thin copper fins? ~~~ then exit the heat out of a blowhole with a high speed fan? ~~~ using the best of both worlds

[Brant]

Copper is a heck of a lot softer than aluminum and therefore makes a smoother cut. If you polished aluminum and copper they would be very similar. Most copper heatsinks have a mirror sheen(flat and smooth) to them aluminum does not even come close to a mirror(not flat and not smooth).

I think that you could rough up copper to increase the surface area.

Dont take any of this as 100% fact.

[Brant]

Originally posted by res0r9lm
the aluminum alloy used to make forged pistons in a car is supposed to be better than copper from what I heard but is to hard to machine (wear & tear on equipment)

Why is it better than copper, heat transfer, disipation, capacity?

[JigPu]

Originally posted by Brant
Copper is a heck of a lot softer than aluminum and therefore makes a smoother cut. If you polished aluminum and copper they would be very similar. Most copper heatsinks have a mirror sheen(flat and smooth) to them aluminum does not even come close to a mirror(not flat and not smooth).

I think that you could rough up copper to increase the surface area.

Dont take any of this as 100% fact.
Thank you! Now it makes more sense to me... I would have never thought about the softness issue...

JigPu

[Sonny]

Originally posted by Patchmaster
I hope a moderator deletes this entire thread pretty soon before some unsuspecting overclocker comes along and tries to use the vast quantity of misinformation contained herein. There is also good information here & deleting a thread that equally voices out opinions of HSF design is good for everyone would be a mistake. Everyone should hear both sides.

Cu vs Al = What could be the reason why the top HS manufacturers choose a Cu base & Al pins? There seems to be a marked performance increase with this combination so dont be too quick to judge the benifits of Al. Just because somebody whips out a book to show you the laws of physics & explain relativity to you doesnt mean they are right. Just take a look at what really happens. Dont forget that the world is round:beer:

[Brant]

Yes good point. Although Al has a higher heat disipation rate copper conducts heat nearly twice as well as Al. 3.9 W/cm*K to 2.165 W/cm*K of aluminum. There!

[will_maltby]

if copper was cheaper, I wonder if we'd have aluminium sinks around at all!


I'm pretty sure copper is a better conducter since it's more dense, therefore the atoms are closer together and so their vibrations are transferred more quickly to the next atom.

I think that's the underlying notion to heat sink material; the denser the better. Though I dunno if that means lead or gold are even better than copper...

[Malakai]

actually, silver is the best conductor of heat, but prices for a silver heatsink would be nuts.

[vandersl]

Ok, I'm not going to speak on the Al vs Cu heat transfer to air issue (I think they are probably equivalent). However, I think people are confused on heat transfer and thermal conductivity.

Lets use a simple example - I have a 1cm square block of copper, which is perfectly insulated on 4 sides (i.e. no heat loss there). I need to move 60W of heat from one non-insulated side to the other. What temperature differential will I need across the block to do this?

The equation is Q = TC * A * dT/dx

Where Q is the heat flow, TC is the thermal conductivity of the material, A is the cross sectional area, and dT/dx is the temperature/thickness gradient.

The thermal conductivity of copper is 4.01W/cm*C. The area is 1cm^2. Q is 60W. The thickness is 1cm.

So 60W = 4.01W/cm*C * 1cm^2 * dT/1cm, and the temperature difference dT is 14.96C.

That is, if I try to move 60W of heat through a block of copper 1cm thick and 1cm^2 area, I will need a temperature difference of around 15 degrees C. Aluminum would require a much larger differential.

Now, this is quite similar to a heatsink base, though the heatsink is much tougher to model (ack, calculus skills have abandoned me!). Heat enters at essentially one small point, and dissipates through the material in all directions. Forget this for now.

Now the unfortunate part is, you always need the same temperature differential to move the heat, no matter how hot the block becomes. If heat isn't removed from the other end of the block as fast as it is put in, the block temperature must increase. But the same differential is required, so long as 60W is being delivered (ok, for those nit-pickers out there, the temperature differential won't be 15C if not all the heat is being removed at the other end...so sue me). If the block temperature increases 10C, then the temperature at the driving end must increase 10C as well.

From this, I think you can see that improving the conductivity of the material does nothing unless you can draw the heat away as fast as it is being put in.

How fast you can remove the heat from the material, if using air or water/fluid (really the same thing), is a function of a lot of parameters, including surface area, air flow (since conduction is vastly improved at the surface by moving warm air away and replacing it with cool air), and the temperature differential (which is affected by air flow, which is affected by fan/pump).

If you have a given surface area and air flow, the primary remaining variable is the temperature differential. Eventually, the temperature differential MUST reach the level where the heat flow in equals the heat flow out.

Going back to the example of the 1cm block. Lets say that with the 1cm area, and some made-up fan blowing air, I can remove 3W per degree C of temperature differential (thats a good fan!). I need a 20C differential to remove the 60W.

This means that from the standpoint of the end where I am putting in the 60W of heat, I need to be 14.96C + 20C above ambient at equilibrium. If ambient is 25C, I need to bring the input side up to almost 60C.

Now, lets say I add fins to the end of the block. Nothing changes in the block itself - all previous calculations are valid. But now, due to increased area, I can dissipate 10W per degree C (not between the fin surface and the air, but still between the block and the air). For 60W, I now only need a 6C differential. the 14.96C drop in the block is unchanged, but my entry (ok, core) temperature now only needs to reach 20.96C above ambient.

Now, the neat thing here (in this idealized model) is that you can begin to see that the base and fins can be treated somewhat independently. I can change my fin structure and air flow values without changing the characteristics of the base. Similarly, I can change the base without changing the requirements for the fins.

Again, this is all simplified. But the concepts are valid (I think :)).

[Brant]

vandersl, I just want to point out one thing that you left out.

Copper, having a higher thermal conductivity can have a much larger surface area with the same heat conduction as Al.

Say your Cu 1cm block with 4.01W/cm*C has needs a temp differential of 14.96 C, Surface area=5cm^2 no base

But an Al block with a 2.165W/cm*C will need a temp differential of 27.71C, Surface area=5cm^2 no base

Or having an Al block .54*1*1 cm, 2.165W/cm*C you need a differential of 14.94 C the same as a 1*1*1cm copper block. The surface area is smaller however decreasing heat disipation to the air. Surface area= 3.16cm^2 no base, 1.84 cm^2 smaller than an equal conductively efficent copper block.


By the way, diamond has much better thermal conductivity than silver which is only about .2W/cm*C better than Cu

W/cm*C
Diamond (room temperature) 6.299
Diamond (77 K) 24.
Diamond (room temperature, isotopically pure) 50.
Aluminum 2.165
Copper 3.937
Gold 2.913
Silver 4.173

[flounder43]

I am sure all of this has been discussed, and fully figured out by engineers at the companies that design HSF's....

If you really think that they have something wrong, I would like to see a person's education and credentials when I evaluate the value of what is posted.

[jboy]

Hi Sonny, thanks for your input on my queries. Thanks for your offer to forward my results to Joe C. , but I've decided to wait for the SLK600 and stop wearing down the center lugs of my socket. Although both sinks have the same footprint, (tho one's copper and the other aluminum), you can never be sure that the mounting/unmounting process results in exactly comparable physical configurations, couple this w/the uncertainty of onboard sensors and it just seems like too much work for what essentially will be a not very solid measurement. I'm planning to buy a front-panel breakout box from kdcomputers.com, that includes a Dr. Thermo battery operated temp diode . Do you know if this particular diode will give me more reliable CPU temp info the the on board sensor currently being used ? Also, a review I found of this item said the diodes are "delicate". Are they physically delicate, prone to burnout, what ? Tnx..

[soil]

Is there any small on-line store fabricating copper heatsink based on customer sample/drawing/spec?

Not off-topic as sometimes I do agree with the toic of this thread.