Specific Heat Capacities in relation to heatsinks....

[RedDeathDrinker]

It's been muttered in Cooling about the different SHC's of metals and its effect on the performance of heatsinks.

As you are probably aware, the specific heat capacity of a solid or liquid is defined as the heat required to raise unit mass of substance by one degree of temperature. This can be stated by the following equation: ∆Q=mc∆T, where ∆Q is the heat applied to the substance, m=mass of substance, c = Specific Heat Capacity, and ∆T the rise in temperature.

As ∆T is the factor we (as overclockers) are interested in keeping as low as possible for a given heat output (the burny, burny CPU), there are two possible options for improving performance of our heatsink. Either increase the mass, which leads to motherboard stress etc etc, or use materials with a higher Specific Heat Capacity.

If we take Copper as a standard base material, we can compare its SHC to that of other materials. Here is a link to a periodic table with regard to SHC.

Why do we stick with Copper? There are many more materials with better properties. Nickel and Iron both have higher SHC's and would (technically) make better heatsinks.

One reason is Thermal Conductivity. Although there are materials which have more suitable SHC's than Copper, Only Silver is actually able to transfer heat at a faster rate than Copper. In descending order:- Silver, Copper, Gold, Aluminium,Berrylium, Calcium, Tungsten,Magnesium, Rhodium and Iridium have the best Thermal Conductivity.

Have metal heatsinks had their day? Is exotic liquid cooling the way forward?

Or should we start looking at special cooling alloys to get the best out of our heatsinks?

 

[BillA]

you might want to catch this quite excellent article by Dave Smith (myv65)

General Heat Transfer Guide

be cool

 

[JoeC]

You assume metals are the only viable heatsink material - think outside the box.

Joe

 

[BillA]

thinking outside the box is the easy part
I too like water, as does anyone who digs into this subject
and the heat pipe is the embodiment of all that we could wish for

so where are they ?

when I first encountered Thermacore's 'flat plate heat pipe' (what in marketingspeak they now call a 'vapor chamber) several years ago I was immediately on the phone to them to place an order
but, their min run is 50k units, and they prefer licensing agreements

so where are the heat pipes ?

I heard that the Sparc IIIs (?) had such, but could never source one;
but pipes or a flat plate ?

there are some real apparently unresolved problems with the flat plate units
(localized overheating, physical orientation, cost also perhaps ?)

pipes are much easier to do but do not seem to offer any performance advantage
(due to the increased number and physical configuration of the thermal joints)
look at JoeC’s test results

there are literally dozens of companies on the 'net trying to peddle their embryonic (translate unproven even in the lab) 'technologies'
but no good products for sale

I would like to make such
I am aware of available capital to form a company to make such
I know existing companies who would like to hire someone who could set up a production line to make such

somehow I'm getting the idea that the technology, at least for flat plates, is not quite there

anyone got a link to a vendor ?
(Thermacore was selling 4"x3" engn eval samples for $400)

be cool

 

[The Overclocker]

You assume metals are the only viable heatsink material - think outside the box.

Joe

yes, i agree that you should think outside the box - but dont go as far to ignore conventional and probably the best options

there are a few reasons metals are used for heatsinks/waterblocks ect - most of the reasons have been covered above, another reason is that 40% of the heat transfer is done by free eletrons - something that makes metals even more appealing

i belive there are only two options for a non metal heatsink material - and that is :

graphite - but cannot be used due to the atomic structure that it forms which only allows heat transfer on a 2d plane

diamond - the best conductor of heat - but ruled out due to the difficulties in machining and price

so there yuo go - seems that we MUST use metals.

 

[Breadfan]

Well I think Joe is also referring to synthetic materials. I'm sure there are synthetics we have yet to "stumble" upon that will provide greater heat transfer rates than precious natural substances that are not viable options due to cost/difficulty in machining.

The interesting study might be if a particular molecular structure or pattern of molecules within the material (such as a certain type of crystalline structure) could aid the in the heat transfer process.

Mike

 

[nihili]

I think there are some design possibilities even with existing materials, or perhaps with some derivatives thereof.

Take graphite for example. It only transfers in one direction, but that just means we have to be more creative in design. I can think of a number of designs that would benefit from transfer along a graphite channel. For example, think about a large surface area copper base plate inlaid with graphite radiating from the center. The graphite should basically increase the ability of the plate to spread heat out away from block, thus allowing more surface area for heat transfer to the water. Combine that with a plate of variable thickness to optimize the conductivity of copper over the core and you should be able to improve over existing designs. Now it may not be economically feasible, but it's not a materials problem. If someone could make graphite in flexible strands, that would be even better.

Speaking of odd ways to use graphite.... What would happen if you formed a block of copper with short graphite fibers strewn randomly through it? I'm guessing that there would be an optimal density for the fibers at which heat transfer would be aided more than occluded. Does anyone know if that's a good guess?


As for where the heatpipes are, a lot of laptops use them, and I believe that one of Shuttle's mini boxes uses one.

nihili

 

[BillA]

I understand that Aesik's thesis involved using amorphous carbon fiber swedged into indium (?) for the conductive face.
He posted a bit about it in an old thread (which I do not recall)
- there is a fair bit of this technology about, just quite removed from the OCing world

perhaps Aesik will visit to describe his work

nihili
I'm after a flat plate heat pipe set vertically with a 100W 1cm² heat source
lots of vaporware, even pricing, but no hardware

a carbon/copper composite will behave as other composites do, the difficulty still being the anisotropy of the carbon tube - having high thermal conductivity along its length but far less transversely
homogeneity would likely be a bit*h

were I blue skying the subject I'd be more inclined to look at crystal lattice structures
- I think I recall reading that bucky balls had high thermal conductivity

be cool

 

[Greedy Guido]

I though that there was some form of carbide that had better heat transfer properties then silver.

 

[William]

I think in the near term the solutions are going to be what we keep seeing: bigger, louder, heavier. This will all take a step back when 64 bit hits the market at some lower speeds but they won't last too long. I am not sure how much we are going to need to go beyond where we are right now with newer materials. There are still enough unexplored methods for cooling with what we have that we should be able to stave off long enough for 64bit and ultimately smaller, lower powered transistors along with RISC.

I think anyone thinking synthetics is along the right train of thought. They have an electrically conductive plastic out now I believe so designing one that can conduct heat is certainly feasible.

But I think in the near term, the creativity of the heatsink designers is going to have to continue to allow heatsinks that will allow overclocking(desinging ones to run at stock speeds isn't too challenging) or some new mounting shemes stronger than the motherboard(motherboard tray for instance).

 

[pHaestus]

Also electrically conductive plastic presents the possibility of the heatsink or waterblock being clear with space for blue LEDs.

A more practical question: How the hell does one measure the heat capacity for these porous carbon materials (like the Poco foam etc)? Certainly not something I can do in the basement, but I have actually been getting some queries about experimentally determining heat capacity of heterogeneous (and porous) samples at work. Some sort of differential scanning calorimetry, perhaps? Maybe Aesik can enlighten...

 

[Breadfan]

I remember a synthetic material being tested at Spodes Abode, and Overclockers.com tested one way back.

If you look pretty far back, Joe tested some odd heatsinks from Russia that were some sort of polymer. They had design that looked like stacked cylindrical plates. If I recall, they perfrmed "ok" but could've done better if the fin design was one that promoted better airflow. Perhaps still not better than Aluminum, but stuff like that is a step in a new direction.

Mike

**Edit: That material was at Spode's Abode...I'm sure it's still there archived somehwere..

 

[William]

There are many ways to test specific heat, the most common way with metals is to heat a bit to some temperature, drop it into water with a calorimetor, and wait for equillbrium to establish between the two and record the final temp. You can figure out the q of the water which equals the -q of the metal and work from that. Thats a pretty rudimentary way, but it works quite well and is easy to do.

 

[RedDeathDrinker]

....and what about surface area? As CPU's are manufactured using smaller and smaller processes (0.9 microns in not too far away), the physical size of CPU's will shrink, leaving less area for heat to escape.

What about some form of internal cooling? Allowing some form of cooling medium pathways inside, and around the CPU die?

 

[JoeC]

I wish I could say more, but NDAs preclude me from jumping in. Suffice to say there are some very interesting "skunk works" out there. Many practical problems, but significant potential.

Love the discussions in Tech Forum!

Joe

 

[DocClock aka MadClocker]

Though technicly challenged, I was very excited when "Bucky Balls" were first discovered, and I wouldn't be surprised to see some new polimers infused with BB's as a solution to heat issues.
I also think that cpu makers are frantic over what comes next, as it is obvious that heatsinks aren't going to cut it for much longer, and if Moore's law still holds true then the next gen of proc's are already in trouble with respect to heat.
How about a Porus plastic with micro channels to let the heat escape?

 

[William]

64 bit will buy some more time running at slower clock speeds. It is also possible to go the route of AMD and do more per clock cycle rather than have more clock cycles which helps to reduce the heat. Also, prudent design on the part of either company will result in cooler chips if they want to focus on that, right they are primarilly focused on speed. Cooling a P4 and XP isn't terribly challenging, just cooling those overclocked puppies is. Lowering the voltage required for a CPU is certainly another way to do that as well.

There are also some other "ad hoc" solutions to this issue. Simply go to a dual CPU architecture to get more performance a la Apple(G4 is at the limit, so put two in to compete). While you don't get a doubling in speed from this, you can use two lower speed CPUs to get the same performance from a single faster CPU. While its the brute force approach to it, its certainly possible though it puts you down a bad road as well.

Of course, IBM has a line of technologies that will work wonders on this, particullarly in regards to transistor size. The smaller they get, the less power they require and the less heat they will produce. Although its a ways off, IBM has developed a 1 atom transistor.

Another method to achieve this would be to change the entire architecture of computing(another thing IBM is looking at). Ternary computing is possible in which a computer would have three and not two positions(off, on, and on/off or something like that) so you do more with calculation through the computer requiring fewer cycles.

Another approach would be to cool the die on two sides doubling the amount of area to cool it with right there. Simple have say a heatpipe going to the back side would do a ton for cooling instead of our current method of only cooling one side.

Of course, all these aren't that plausible in the near term. Its more likely that water cooling will go mainstream if air just can't keep pace. But there is still a lot of room to go with air. I remember when the Glaciator came out as a mid noise, high performance heatsink. It seemed to break every trend out there and did so by some elegant and brute force design(It was very massive).

As for internal cooling, that is certainly possible. Something like a heatpipe on a very tiny scale could do wonders though it would be very expensive.

 

[nihili]

Another method to achieve this would be to change the entire architecture of computing(another thing IBM is looking at). Ternary computing is possible in which a computer would have three and not two positions(off, on, and on/off or something like that) so you do more with calculation through the computer requiring fewer cycles.

Do you happen to have any links to this? My brain tells me there are some issues with 3 valued logic that would make this difficult. It's also not clear to me that it's merely a matter of doing more calculations per cycle.

Is it possible that they're doing ternary computing in the sense of using truth functions with three arguments rather than functions with a range of three?

After I get some more coffee in me, I'll see what I can dredge up from the recesses of my cranium.

nihili

 

[ookabooka]

Quantum computing uses the position of an atom for computations. Each atom has 3 position: up, down, and up/down. The up/down if I remember correctly, is a bit unstable. Also, there has been some talk of left and right being a possibility. Imagine, instead of 0 and 1, you would have 0 1 2 3 4.

 

[nihili]

Yeah, I've done some work with multi-valued logics. There are some generalizations due to Post that allow any denumerable number of values. Fuzzy logic allows non-denumerable values, though I don't know if it can go beyond continuum many. Why one would need more than continuum many isbeyond me, but I suppose it's possible. I think the finitely valued versions of qualtum logic are actually simplifications, but I'd have to double check that.

It's certainly possible to have logics of however many values you care to have. All you have to do is define the output functions. The question is what the utility of doing so is. In order to have a heat savings, you would need the new logic to be a conservative extension of a boolean algebra. That part isn't so hard. But furthermore you would need some general algorithm for turning a computation in the old logic into a shorter computation in the new logic. that part is very tricky. To take one example. Intuitionist logic is a three valued conservative extension of boolean algebra, but proofs are typically longer rather than shorter. Thus an implementation of intuitionist logic on a cpu would probably increase temperatures as more calculations would be required to do the same work.

The only way I can see to routinely decrease computation would be to explore implementing a logic that allowed 3 or 4 inputs. It shouldn't be to hard to algorithmically discover when a computation would be improved in such a case, and hence to build a cpu that would only use the the new gates when there was a reasonable chance of computational savings. Alternatively you could define a variable input gate that was a conservative extension of the boolean gate. This is trivial for the most common gates (AND, OR, IFF) but requires some finesse in other cases (IF, XOR, etc.). Because it's a conservative extension, it should work seamlessly with old programs. Ne programs that could take advantage of the increased inputs could save considerable computation. For example, If you want to return 1 just in case each of four inputs is 1, this requires three computations with boolean AND, but it would require only 1 computation with a 4 input AND.

nihili