The most efficient way of watercooling?

[MadHacker]

Originally Posted by jaydee116

It is not even close to. Waters thermal conducticity is .6 while coppers is 390. Yes that is a point in front of that 6!

http://www.hukseflux.com/thermal%20c...ty/thermal.htm

that starts to make sense to me...
.6 wow.. copper is 650 times more conductive... I guess i'm convinced
I can C why mebe putting nano particals of copper in the water would be helpful... but wouldn't that aid in corrosion? duno.. off topic anyways...

[AveMORphine]

What strikes me as odd is that most ppl start talking about the thermal conductivity when it comes to this question.

The 1 thing most seem to forget is that the water carries the energy (water can hold a bit more energy (heat) than copper can) and since the water is the carrier, everything btw the water and the heat source becomes a thermal resistance.

[MadHacker]

Originally Posted by AveMORphine

What strikes me as odd is that most ppl start talking about the thermal conductivity when it comes to this question.

The 1 thing most seem to forget is that the water carries the energy (water can hold a bit more energy (heat) than copper can) and since the water is the carrier, everything btw the water and the heat source becomes a thermal resistance.

OK things were starting to make sense ...
but now I'm confused again... :shrug: nothing new...
so by not using copper as a heat spreader you lower the thermal resistance?
so there must be a point where having a higher thermal conductivity outweighs thermal resistance?

[superart]

Originally Posted by MadHacker

that starts to make sense to me...
.6 wow.. copper is 650 times more conductive... I guess i'm convinced
I can C why mebe putting nano particals of copper in the water would be helpful... but wouldn't that aid in corrosion? duno.. off topic anyways...

When I first read the article and discussed it with pH, he had the same concern. I emailed the professor doing the research and asked him if he noticed any corrosion either to his pumps or his blocks.

He replied saying that since the particles are so tiny, on the order of microns, after extended use and testing, he has seen no sign of excessive wear or corrosion. BTW, the guy I'm referring to isn't using copper, hes using Aluminum Oxide, but that's beside the point.


....and yes, your'e right...off topic :shrug:

[Althornin]

Originally Posted by AveMORphine

What strikes me as odd is that most ppl start talking about the thermal conductivity when it comes to this question.

The 1 thing most seem to forget is that the water carries the energy (water can hold a bit more energy (heat) than copper can) and since the water is the carrier, everything btw the water and the heat source becomes a thermal resistance.

Yes, but without the high thermal conductivity (or high surface area) of a copper waterblock...the temperature differential must be higher between the two (die and water) before all the heat is transfered.

You cannot look at one, without the other.

In direct die, the low thermal conductivity of water becomes the barrier - the delta between the die and the water must be higher in order to transfer all the heat produced.
With a water block, the thermal resistance of the copper and TIM become the issue - Because the high surface area makes up for the low thermal conductivity of the water/copper interface.

Remember the units - that one problem in this thread. I see tons of numbers being thrown around, most of em don't have units....and are thus worthless.

Look at JD's numbers for example. .6 and 390 - what good are they? Without the units, thermal conducitivity number mean nothing. Because thermal conductivity changes based on temp delta and surface area of contact...

[AveMORphine]

Indeed the larger area will make up 4 a lot of it, otherwise we would proborably allready been discussing inlet designs to our directdie blocks instead of the numourous designs on normal block`s.

But I am looking forward to the results from http://www.ocshoot.no `s tests

They are currently testing direct-HSF, by modifying an WC Antarctica. Those guys have a lot of experience with cooling (Normal Watercooling, direct die, direct TEC and so on).

Perhaps we are about to get an eye-opener?

http://www.ocshoot.no/antarctica_xtreeme1.htm

[Cathar]

Originally Posted by AveMORphine

Perhaps we are about to get an eye-opener?

http://www.ocshoot.no/antarctica_xtreeme1.htm

With the direct-die block set up like that? Not a chance.

[jaydee]

Originally Posted by Althornin

Look at JD's numbers for example. .6 and 390 - what good are they? Without the units, thermal conducitivity number mean nothing. Because thermal conductivity changes based on temp delta and surface area of contact...

Dosn't change that much....

[Incoherent]

Originally Posted by Althornin

...Look at JD's numbers for example. .6 and 390 - what good are they? Without the units, thermal conducitivity number mean nothing.

OK, he doesn't give units, I'll give them to you, standard, almost always used in a metric world, the unit is W/m*°C. Watts per metre for one degree celsius. Even if it was given without a unit the proportion is enough to tell you what you want to know, Copper is ~650 times better than water. Heat transfer is directly proportional to k. Irrelevant for a comparison when we have no idea of the convection coefficient.

Originally Posted by Althornin

Because thermal conductivity changes based on temp delta and surface area of contact...

Bullshit. Thermal conductivity changes based on absolute temperature (google Wiedemann-Franz Law). It has nothing to do with surface area and delta T is only relevant insomuch as k is varying slightly throughout the heat path. Slightly.
What I think you mean is that heat transfered, heat flux or watts changes based on temp delta and surface area of contact. Rearrange this to our scenario, temp delta is directly proportional to heat flux 'Q'. Surface area 'A' is a constant, heat path length 'L' is a constant and thermal conductivity 'k' is a constant.
EDIT: Thermal conductance, defined as the inverse of thermal resistance, is perhaps your intended meaning?
The formula, Fouriers Law is Q=k*A*dT/L or for us dT=Q*L/k*A. This applies to a waterblock, it is half the story. It is not applicable to direct die. (except within the die itself, this is the same for DD or conventional WB)
Direct die is almost purely governed by Newtons law of cooling. Water blocks are dominated by it as well, how much so depends on the design philosophy. It is similar. Q=h*A*dt. 'h' is the convection coefficient, units W/m^2*°C. It is a function of water flowrate, thermal conductivity, density, viscosity, specific heat capacity, thermal diffusivity and the main problem, surface geometry and localised water velocity...
We do not know it. Thats the problem.

[Althornin]

Originally Posted by Incoherent

What I think you mean is that heat transfered, heat flux or watts changes based on temp delta and surface area of contact. Rearrange this to our scenario, temp delta is directly proportional to heat flux 'Q'. Surface area 'A' is a constant, heat path length 'L' is a constant and thermal conductivity 'k' is a constant.

That is, indeed, what i meant.
appologies on incorrect terminology - it's been awhile since my PDE classes.

Note: Please say "bullshit" to everyone else who used the incorrect terminology before me in this thread.

[jaydee]

Originally Posted by Althornin

Note: Please say "bullshit" to everyone else who used the incorrect terminology before me in this thread.

Bullshit or not I would love to hear any corrections. Until someone takes the time to correct things it will always be incorrect.... :shrug:

[Albigger]

Ok, there seems to have been a bit of confusion in this thread, intermixed with good information. Hopefully this explanation will help out:

In comparing direct die vs a waterblock scenario, in a waterblock the heat flux must CONDUCT through the wateblock base, CONDUCT through the water boundary layer on the inside of the block, and CONVECT away from the (slowly moving) boundary layer in the bulk (fast-moving) fluid. In direct die, the heat flux must CONDUCT through the boundary layer of the water, and then CONVECTS away, carried in the bulk fluid again. And (as mentioned above by InCoherent) Fourier's and Newton's laws, respectively, govern conduction and convection. And so it is simply not enough to compare the thermal conductivity of copper versus water (unless one was talking about stagnant water cooling the core).

In general, the boundary layer of a fluid is conventionally defined as the distance from a surface where the fluid velocity is 99% of the free-stream fluid velocity, and inside this boundary, the velocity varies from this value to zero (at the surface - either core or waterblock).

Now it is easy to see (based on the numbers quoted for thermal conductivity of water and copper) why latest block designs attempt to (at least in some amount) REDUCE the boundary layer thickness (by increasing fluid velocity and/or turbulence). This reduces the distance the heat must conduct through the water, and convection can take place closer to the surface, further reducing the temperature difference between the incoming water and the heat source.

Now, in the poll I voted for direct die (although my vote was purely based on lowest temperaure achievable at the core and NOT on efficiency), but, after some rough (at least to the order of magnitude) calculations, I agree with Cathar about direct die at best falling somewhat short of the current best waterblocks (given the pumps currently in use).

To give an outline of what I thought about, there is something called stagnation point flow or Hiemenz Flow, which deals with a fluid flowing directly at a flat surface (such as maybe a single slit or jet over a cpu core). If water velocities are high enough then viscous forces may be neglected and a similarity solution may be found - probably this type of flow would be covered in any intro to fluid-mechanics course (or can be found in this book, one of my references: Incompressible Flow by Ronald L. Panton, 2nd edition). I'm sure you can also probably google for 'stagnation flow' or similar to find images and information concerning this.

Disclaimer: Please let me know if any of this seems wrong and I will attempt to correct:
Carrying on... from Fourier's law, when considering direct die, lets see what would be needed to achieve similar results to today's waterblocks (we would at LEAST want to be on par, correct?) Assuming the following values: Heat flux, Q = 100 Watts, Area (say roughly 170 sqr. mm), k = 0.6 W/mC, and temp-difference ~8 degrees celcius (water to core - seems reasonable). using Fourier's law, where Q = (k*A*delt-Temp)/distance where 'distance' is the boundary layer distance of the water, this gives B.L. dist = .0082mm - a VERY SMALL distance. From the results of the stagnation point flow analysis of momentum and mass conservation equations, if the reynolds # (based on some characteristic length) is much greater than 1, then it can be shown the boundary layer thickness = 2.4*sqrt(kinematic viscosity*charac. length/free-strem velocity) or BL = 2.4*sqrt(v*L/U) the characteristic length is taken to be half the length of one side of a 13mm square cpu core (or 6.5mm) - this corresponds to ONE jet for the die. using this and the kinematic viscosity of water at about 30 deg. celcius, the free stream velocity necessary, U, comes out to be about U=450 m/s (yes that's meters per second)! Obviously, way too high to achieve given the standard methods of today.

since thermal cond. of water, cpu size (surface area), temp difference we want to achieve, and heat flux are fixed, the MAXIMUM allowable boundary layer thickness (to achieve the desired temp diff.) is also fixed. From this, one improvement would be to add more jets which would reduce L (by the way the problem is set up), and then also reduce the required free-stream fluid velocity.

Ok, one glaring error I should probably note is that the actual boundary layer will vary in thickness from the stagnation point out towards the surface edge (if considering a flat cpu core), and the calculation is for the boundary layer at the assumed outside edge of the core (where it would be thickest) and so the actual heat transfer would be somewhater better than this.

Finally, I'm tired, its late, and I hope any errors I made will be readily pointed out by the members here.

[Incoherent]

Originally Posted by Althornin

Please say "bullshit" to everyone else who used the incorrect terminology before me in this thread.

LOL


Sorry, I was a bit grumpy yesterday morning. That comes off as a bit rude I guess, not my intention or normal nature.
It is kind of hard to refute a lot of this thread, much of it takes the form of questions and there's no such thing as an incorrect question. On the other hand it is very easy to make an incorrect answer, especially when it is on the boundary of one own knowledge. So I'll just hang out and watch, emitting the odd "bullshit!" when I see something wrong which is within my perimeter .

Albigger, I like your explanation of the boundary layer aspect, a minimum required thickness being very small. Since the waterblock geometry can be optimised to actually reduce this layer, over a larger area, it starts becoming clear (to me at least) that direct die can never be as effective within the working bounds of "normal" pumps. Direct dies' saving grace is perhaps the lack of TIM joint.

[Les]

Originally Posted by Incoherent

Albigger, I like your explanation of the boundary layer aspect, a minimum required thickness being very small. Since the waterblock geometry can be optimised to actually reduce this layer, over a larger area, it starts becoming clear (to me at least) that direct die can never be as effective within the working bounds of "normal" pumps. Direct dies' saving grace is perhaps the lack of TIM joint.

Suspect to understand Submerged Jet Impingement(SJI), you have to split into Wall jet region and Impingement region .
This approach is used here.
They consider the "Womac correlation" to best fit the heat transfer.
I use this paper as my baseline

Guess you have to refer back to Womac's work to gain a more fundamental understanding.
I have not done , partially because found little on-line.

[YaYappyDoa]

uhm... better to make some tests... same pump, radiator.
Enlarging the area spreading heat (using a copper plate, better with fins/microchannels) should be better than using the bare chip area. Air cooling teach that enlarging the area spreading heat by using well designed copper/aluminium heatsink and a good fan helps to make the job. To obtain an airflow effective enough to cool a bare cpu area you need a fan strong as a B52 helix, that's a bit complicated. Better apply an heatsink on it. Nearly the same should be for water cooling (with regard about density and uncompressibility, then using a waterblock with fins or microchannels).... Clearly, if you use a fireman's pump direct on cpu die you have no need of copper plates or waterblocks!!

[Kobuchi]

I'm intrigued that the pollster phrased it, "Direct DIE vs Waterblocks with solid bottoms", as though to leave no room for hybrids. Often, "the most efficient way" dovetails strengths of several archetypes.

[FL3JM]

Kobuchi, what i ment with solid bottoms was that the block is "closed", that it has no holes in it so that the water touches the core in some way. Sorry if i wrote that kinda blurry. >.<

[Kobuchi]

I'm sure everyone understood what you meant without giving it any thought. Besides, one must present the sides in a poll or debate in plain contrast.