Get Your Cold Plate Thickness Calculation Here

[BillA]

lol
can anyone say 3/8" ?
btw, 30°C for the hotside is quite cold considering the coolant rise over amb plus the thermal resistance of the wb plus the wb TIM joint

 

[matttheniceguy]

well 12.5 mm is closer to 1/2, but your right, you'd be nuts to make it more than 3/8 in a commertial product.

Does anyone have any links to tests done with multiple cold plate thicknesses? I saw one like a year ago where a guy used 4 or 5 different thicknesses on the same setup and graphed the results. I would love to see that and try his numbers in my model to see if I get the same results.

 

[doc]

Is that for pure copper?

 

[matttheniceguy]

K = 390 w/m^2k, so yes, oxygen free copper.

The thickness would vary with hot side temperature, pelt and cpu power, and cpu size, but I think the model is pretty good at predicing the best thiskness. The actual temperature will vary a bit from the thermal resistances of the TIM and other things, but I think this model tells us the optimal thickness pretty accuratly.

 

[Les56]

I think you may have to use real TEC data to get meaningful results.
Last time this was discussed I presented(Post 25) link.
I now use this model - linked at the end of the linked thread.
Model and method(copied and pasted from Post 29):-
"Cold-plate Spreading Resistance revisited.
I think* Bill is 90% right indicating h~2w/c .However this should be expressed as h~868w/m2*c for calculation purposes.
The difference is mostly academic :

Have included the Total Resistance of the Coldplate.
This is obtained via:
Coldplate Resistance(c/w)= Total Resistance(Waterloo) - Contact Resistance
Contact Resistance= 1/(Area x h) , where h is expressed in w/m2*c.
The effect on Die Temperature :

Kryotherm is used.
Rc is the Coldplate Resistance + 0.0833c/w( TIM resistance for a 11x11mm layer)

* Think I am wrong in equating h to the heatsink's Thermal Conductance(w/m2*c).
The Film Coefficient(h) is the Thermal Conductance of the Coldplate's cold surface. This is taken as the [Reciprocal of the sum of Resistances] x Area.
Resistance(to ambient)= PeltierC/W + Heatsink(C/W to ambient) +TIMs(C/W)
Resistance(to ambient) is minimum when Heatsink(C/W to ambient)=0 and TIMs(C/W)=0.
The Film Coefficient(h) is maximum when Resistance(to ambient) is minimum
The Film Coefficient(h) will always be less than the inherent h~868w/m2*c Thermal Conductance of the Peltier.
Kryotherm suggests that
PeltierC/W ~ 0.5(actually 0.420) and hence the Maximum Film Coefficient(h) ~ 868w/m2*c (actually 1033w/m2*c) :

"

.......................................................................................................


The same procedure for the more powerful(Higher Flux density) 172w,40x40mm TEC gives the comparison shown in "this link" plus die temperatures:-

Some additional bits and bobs:-

EDIT: Bill, 3/8".

 

[doc]

The reason I asked about the purity of the copper is that it don't take much of any thing to change the thermal conductivity in copper, less than 1% of almost anything can bring it down. I did not use O2 free ( went with the cheap stuff ) this will change the thickness of the cold plate as well. I started getting some results that looks closer to what you and les56 have. I agree with les56 about needing use a real TEC or you must have very good data on the devices. If you use a voltage regulated PS, the TEC will use less current as dT gets larger around the outer areas, larger dT means a larger Seebeck voltage. Which to the PS will look like the device resistance has changed. If the current drops a little, so does the heat pumping capacity. Some equalibrium will be reached in short period of time, but the heat pumping will be different than if the whole surface of the TEC is at one temp.

 

[LabRat23]

I don't quite understand what the thick wall pipe equation is for

The thick wall pipe equation is an equation to calculate the thermal resistance in a radial direction. It is called the thick wall pipe equation because that is about all it's used for. Also a thin walled pipe can be aproximated by a one dimentional heat flow becuase the cross sectional area does change much, but in a thick walled pipe the change in area must be acounted for.

I find it interesting that my equation seems to error toward the thinner side, if these other models are correct. I would not doubt it since I probably eversimplimied things in the derivation of the equation. This has got me curious enough to write a program to calculate the heat flow and the temperature drops across the plate, taking into account the the rate of heat transfer to the pelt is dependant on temperature.

Hard lesson learned: the sooner you eat your crow the less horrible it tastes.

 

[matttheniceguy]

I don't think anyone has really had to eat any crow here. We have all been coming up with relativly accurate ways of trying to determine the best cold plate thickness with different models. None of these models are compleatly correct, but they are at least a starting point. The only way to really determine the best thickness for a perticular setup is to just try a few thicknesses and see what works best. At least these models can tell you roughly where to start.

 

[LabRat23]

The only way to really determine the best thickness for a perticular setup is to just try a few thicknesses and see what works best. At least these models can tell you roughly where to start.

I agree. I would bet that you would get your answer faster that way too.

I forgot to ask, what is that software you are using and how could I get my hands on it?

 

[doc]

If I can, I try to make the crows small. They go down easier. But like I said you guys keep me on my toes, plus a little less multi-tasking at work won't hurt either. Live and learn!

 

[matttheniceguy]

I was using ANSYS to do the modeling. I have access to it because I am a mechanical engineering student at UBC, and one of our proffs is largly behind all the actual FE modeling and the theory of how the program works. He managed to get a licence for the university.

If you want to run it yourself, I believe to get a one year licence for a decent workstation (1 modern precessor) it is around 20K a year.

Another good program is COSMOS. It is made by the same people who make solidworks, so you can do your modeling in solidworks and them do the FE stuff with COSMOS. I don't think that program is exactly cheap either.

 

[LabRat23]

I figured you were a mechanical engineering student. I am a student at Virginia Commonwealth University, and we have a very nice computer lab with full SolidWorks and COSMOS. I havent used COSMOS yet but I am as pretty darn good on SolidWorks. I'll give it a try.

 

[eobard]

Stuck.

 

[matttheniceguy]

Wow... our mad ravings have been stuck.

I'm still looking for some actual experimental results where someone has tried various cold plate thicknesses on one setup. If we can find this, we can validate and adjust our models so they will be much more accurate. Let me know if you can find any.

If anyone is looking for the best thickness for a perticular setup, PM me with your specs and I'll throw them at ansys. I'll need to know the delta T and Qmax of the pelt (at the voltage you run it at) and it's size, as well as what chip you have and the intended speed and voltage or wattage. Info on your water cooling will also help, like an estimate of hot side temp.

If you want the ansys results to be more accurate, find me the results I was asking for above.

 

[3vil_l337]

i know this thread has been dead for awhile now, but im curious if this data holds true for the newer large procs and tecs? moastly thinking about opteron sized chips pushing 180-200 watts, and also tecs from the 50-70 mm size? jsut curous as im about to slap a tec and water onto my opty. so um yeah any help would be nice. thank you
3vil

 

[BillA]

science has not been repealed

 

[exhausted mule]

i'm curious if the use of cold plates is purely for using larger sized tec's on smaller die.

if not, then is there actually a temp difference with using one and not?


cause if there isn't. i'm going to try and find a way to slap a large tec on the die using direct contact with copper leading the surrounding edges.

thanks for any help.

 

[matttheniceguy]

The reason for a cold plate is to use a pelt on a cpu which is smaller than it. If you had a pelt that was same size as your core, that would be ideal, unfortunately pelts can't compete with processors as far as watts/area.

You don't want to just touch the cpu to part part of the pelt, because that part of the pelt wouldn't be able to take the power load. If the whole pelt is 200 watts, 1/10 the pelt will only be 20 watts. If your pelt is bigger than your processor (and I will be) you need a cold plate.

 

[greenmaji]

The reason for a cold plate is to use a pelt on a cpu which is smaller than it. If you had a pelt that was same size as your core, that would be ideal, unfortunately pelts can't compete with processors as far as watts/area.

You don't want to just touch the cpu to part part of the pelt, because that part of the pelt wouldn't be able to take the power load. If the whole pelt is 200 watts, 1/10 the pelt will only be 20 watts. If your pelt is bigger than your processor (and I will be) you need a cold plate.

Ive been trying to wrap my head around this for a while (more to help out others intrested then anything else), related quesiton I think..

Would it stand to reason that you would need a *hot* plate in conjunction with the cold plate if your cooler's surfase area (water block, air cooler what have you) is smaller then the pelt?

 

[matttheniceguy]

It certainly would. If part of the cold side of your pelt is not touching anything, that part of the pelt is useless. If part of the hot side of the pelt is not being cooled, that part of the pelt, and likely the whole pelt as a result, will overheat and burn out.