TEC+PROCESSOR, calculate your temp.

[Paxmax]

Edit!: This calculation is oversimplified and will not get you "exact" temps. It also contains some errors. Please read post from Since87 and Doc to learn where the errors/deviations are.
The biggest errors are my optimistic efficency increase between interfacing a TEC with cooler (as most data are found for cooler on cpu die). I compensate with overly negative figures for TECs efficiency. Further down the posts you'll find a handy Excel spread sheet for TEC calcs, made by Since87.

Enjoy! /Paxmax


I going try to explain how to calculate approximate processortemps using a peltier(TEC).

The answer we can get from the calc is Tc, the temp of the coldside of the TEC. The actual processor temperature will be warmer due to thermal interface losses.
The TEC should be looked at as a heat pump. It MOVES heat from coldside to the hotside, if you have observed the polarity correct!

The processorheat will be moved and dumped on whatever is there to cool the hotside. If there isn't anything there to cool it, or the cooling is insufficient, then the TEC will burn up, usually when it reaches around 80 Celcius.

Besides dumping the moved processorheat to the hotside, it will also dump the heat from the power it requires to be able to MOVE the heat.
The TEC is not a magical device, it's a pump, it can't move more heat than input power. At best it moves about 90-95% of supplied power.

The calc I use is in three steps, like this:

Tdif =(1-(Ph/Pc))*Dmax

Th=Tamb+(C/W*(Win+Ph))

Tc =Th - Tdif

Tdif = Temperature difference between the hotside and the coldside, in Celsius

Ph =processorheat, in Watts

Pc = TEC cooling power, in Watts

Win =The power needed to operate TEC, to produce the Pc. Measured in Watts

Dmax = A sort of efficiency and "quality" measure of a TEC running on max without ANY heatload. The value is usually between 60 and 90.

Th =Temperature on hotside of TEC, in Celsius

Tc =temperature on coldside of TEC in Celcius

Tamb = Ambient temperature or room temp in Celsius

C/W = The efficiency of a cooling device.

The C/W value will depend on many factors. Some are: what kind of paste is used, the contact area, clamping pressure, flatness/roughness of surfaces and such.
A really good aircooler today will yield a C/W rating under 0.30 when used with overclockers.com processor simulation device.

When the same aircooled heatsink is used with a TEC the C/W value will be better (for the heatsink/TEC interface) due to larger contact area.
To pass all heat from the processor through the small footprint of it, is like passing high current through a thin wire, there will be high resistance and losses involved.

I don't know exactly how much better so I'm doing a bit of guesswork. The footprint of a TEC is around 14 times bigger than a processor, that doesn't of coz mean it's 14 times better though.
The aircooler efficiency might double or something due to the better distribution of heat and lower heat per contact area.
Any waterblock should also yield better C/W with a peltier than on a processordie.

An example with TEC then....
Lets say we have 70W heat from processor, and we use a 226W TEC. Ambient temp =25 celcius
The TEC's data is: Dmax=67c, Pc=226W, Vmax=15.2, Imax=24
To get the 226W Pc cooling power, we have to input 15.2 volts * 24 ampere = 365 Watts
We cool it with an aircooler Swifttec MCX 462 it has a C/W rating of 0,26. I will be calculating with a C/W of only half of it, since the heatsink will be working more effciently with a large heatsource.

Tdif =(1-(70/226))*67 = 46,25 C Tdif Temperature difference between Coldside and the hotside

Th = 25 + (0.13* (365 + 70)) = 81,5 C at the hotside !!!! Thats a hot heatsink, and that peltier is on the verge to destroy itself.

Tc = 81,5 - 46,25 = 35,25 Celcius!!

Ooops... The cold side of the TEC will be 35,25 cecius, thats not exactly "cold". Not what you expect when you use a good TEC.
The 226W TEC / aircooled heatsink solution isn't going work too well at the TECs maximum rating since the heatsink can't handle the thermal load.
If we run the TEC at 12 volts(80% of Vmax) instead of Vmax we will run the TEC more efficiently.

226W TEC @ 12 volt draws approx 19 A (instead of 24 A)

Pin will be 12 * 19 = 227 W. As a result of lowering input power the Pc will drop to about 90% of input power .9 * 227 =204 W
Doing the calculation again then:

Tdif =(1-(70/204))*67 = 44 celcius (we lost 2 C in temperature difference)
Th = 25+(0.13*(227+70)) = 63,6 C (we got 18 C colder at the hotside)
Tc =63,6 - 44 = 19,6 Celcius on the coldplate

19.6 is the calculated MAX temperature your TEC's coldside will be during full 70W heatload if we ignore a few losses here and there.
It's only 5 C below ambient, thats much better than what you'd get with the aircooled heatsink alone!
With the heatsink alone you'd get 25C + (0.26C/W * 70W ) = 43 C

A cool feature is that you can check theese results with an test done by overclockers.com see http://www.overclockers.com/articles646/
Another cool thing is, I have been using this formula for 2 years.

The C/W value is a true guesswork and it will vary from cooler to cooler, the other values can easily be calculated. Just exchange your numbers with mine in the calculation to test your config.
Remember that the Cooling power(Pc) drops when you run your TEC on other voltage than Vmax. Remember to also calculate new Win(power supplied) for the TEC when you operate it on other voltage than Vmax.

If you drop Volts by 80% you have to recalculate the Amperes by using Ohm's Law.
U = voltage in volts, I = Current in Amperes, R = resistance in Ohms.
Rules are: U/I=R and U/R=I and R*I=U
Also the power. U*I=P in watts


The problem with using TEC isn't to get it cold, it is to provide enough cooling for the hotside.

Mating TEC's with watercooling is necessary to get substancially below ambient temp, or use a puny processor!!

Good speed = Overclocked speed

And please flame me if needed, and prove me wrong, cuz after all this is all just theory.....

/Paxmax

 

[Since87]

And please flame me if needed, and prove me wrong, cuz after all this is all just theory.....

Well you asked for it.

J/K Someone who actually asks to be shown where he is wrong won't get any flames from me. Heck, I'll get you a job if you want to move to Lafayette IN.

For the most part your understanding is accurate. There are a few areas you are off though.

The TEC is not a magical device, it's a pump, it can't move more heat than input power. At best it moves about 90-95% of supplied power.

As you point out, the TEC is a pump. Does a pump need to draw 100W of power to move 95W in the form of heated water? Of course not. Neither, does a TEC need to draw more power than it is moving in the form of heated (excited) electrons.

Don't take my word for it though. Download the Kryotherm software. Under "Thermoelectric System" in the software there is a calculator that lets you vary all the operating parameters for a selected pelt. (Although it doesn't allow you to select a 226W pelt unfortunately) Anyway, if you play around with it, I think you'll convince yourself that a pelt can pump more power than it is consuming. (Or that Kryotherm doesn't know what they are talking about.)

Another good resource is this paper. It takes a bit of study to get a handle on how to use the graphs, but once you understand it, you have a much better idea what you can get out of pelts.

The calc I use is in three steps, like this:

Tdif =(1-(Ph/Pc))*Dmax
Th=Tamb+(C/W*(Win+Ph))
Tc =Th - Tdif

These are a little oversimplified, but good enough for ballpark numbers. If you study that paper you will see what some of the missing factors are.

heavily edited by Since87 for brevity
When the same aircooled heatsink is used with a TEC the C/W value will be better (for the heatsink/TEC interface) due to larger contact area.

I don't know exactly how much better so I'm doing a bit of guesswork.

We cool it with an aircooler Swifttec MCX 462 it has a C/W rating of 0,26. I will be calculating with a C/W of only half of it, since the heatsink will be working more effciently with a large heatsource.

By analyzing Joe's die simulator data for the MCX462+ and MCX462+T I calculated the on-pelt C/W to be 87% of the on-die C/W. That's a very thick based heatsink though. I would guess that you'd get more benefit from the wider surface contact, with a thin based waterblock. I don't know of a good source for comparison data for a thin based waterblock though.

If you drop Volts by 80% you have to recalculate the Amperes by using Ohm's Law.

There's a fair amount of curvature to a pelt's voltage vs current curve. You can see this in the document I linked. Assuming the pelt is a resistor doesn't throw you very far off though.

U = voltage in volts

I think you are a long way from Indiana. Right?

 

[Paxmax]

Whew !

Thanks for understanding Since87!
You are darn right in all your arguments concerning my halfassed method
I guess it was a coincidence with the temps ending up just like the review.
Yup, I'm far away from Indiana, I live in Sweden. I was in the States this autumn, visiting a few friends in CA. What triggered you? The symbol "U" for voltage? What? do you use V?

I haven't been digging really deep into those documents, but glancing at them I saw that this is going to take some time!! I dl'd the kryo thingy, and didn't understand an iota as of how to use it. I guess I need more time to read and comprehend.

About the pump thingy, well I'd be damned. I thought of the peltier as a such inefficient device as it would never be able to pump more than it consumes.

I'd really like to homogenize the whole industry of calculating TEC cooling effect of processors, really dig in and solve the puzzle somewhat.
It would be cool if we all could "chip in" and get a more complex formula, that explains most of the "dark" areas of TEC's.
I(for one) don't strive to be exactly on the 10:th of a degree precise, just to at least close within 5 degrees C (or whatever that is in F).

I could almost bet there's more formulas out there than TEC's !!

Don't you Since87 have a formula on excel sheet or something?
Maybe it's a beginning or maybe it's the end of this debate ??

And please forum members (or others) do chip in if you have some key points that you think is important.

We need to find out some method of finding out the C/W when using the heatsink or waterblock with a TEC. Ehrrm... Wouldn't I be sweet if the nice staff of overclockers.com would make a peltier die to their cpu-die test equipment??? So that every test with the CPU-die on an W-Block or aircooled heatsink they would include a peltier die test aswell??

It's also my guess that we need to focus on how to use the graphs that concerns the particular TEC one would use.

But, a feeling that I've had for long is that peltiers are out, phase change is here to rule. So maybe the intrest in pelteirs are dropping.
Oh well, maybe CoolChips(tm) can change that
I don't know however how big they will be.

/Paxmax

 

[Since87]

Originally posted by Paxmax
Yup, I'm far away from Indiana, I live in Sweden. I was in the States this autumn, visiting a few friends in CA. What triggered you? The symbol "U" for voltage? What? do you use V?

Yes, I see U used predominately in stuff out of Europe.

I dl'd the kryo thingy, and didn't understand an iota as of how to use it. I guess I need more time to read and comprehend.

Here's a list of the input parameters and their relationship to a watercooled pelt CPU cooler. (A lot of this will be obvious, but I figured I might as well make a fairly complete list.)

Select Module - Choose one of the pelts Kryotherm sells. High Effective Single Stage is about the only category of interest to overclockers.

Number of Modules - Set this value to 1. (Once you know how to use it for one pelt, how it works for multiple pelts will be pretty obvious)

Connection - Only used for multiple pelts. (Allows you to choose electrical connection setup.)

Power source - Choose whether you are using a Voltage Regulated source or a Current Regulated source. Virtually all OC'ers will want to choose Voltage. Also, choose the Voltage (Current) applied to the pelt.

Ambient Ta (C) - Ambient temperature in degrees C.

Object heat rejected Wob (W) - CPU power dissipation in Watts.

Hot side parameters Rh (K/W) - C/W of hot side cooling. Good aircooling like the MCX462+ with an 80CFM fan gives about 0.26C/W. Values as low as 0.05C/W are probably reasonable for very good watercooling on a pelt. Because of the extemely wide variations in WC setups most people can't do much more than guess at the C/W of their WC system. Other issues come into play as well. The quality of thermal compound application and the clamping pressure on the pelt have a big effect on this number.

Cold side parameters Rc (K/W) - C/W of the stack consisting of: CPU to coldplate TIM, coldplate, coldplate to pelt TIM. Clamping pressures also play a role here. Because there are two TIM joints influencing this number, and TIM joints are a major source of variation in thermal transfer, this number is somewhat shaky as well. Les who posts on overclockers.com.au and procooling.com calculated a ball park number of 0.1C/W for this. When I used this number to simulate JoeC's die simulator test of the MCX462+T my numbers matched JoeC's very well.

Insulation Rins (K/W) - For our purposes, this is a catchall C/W that is dominated by the heatflow through the following path: Solder bumps on the CPU die, Copper traces and fiberglass from solder bumps to CPU pins, CPU pins to socket, socket to MOBO, MOBO copper traces and fiberglass to air. Complicating the issue is that the MOBO has substantial heat sources other than the CPU. There are lots of other paths such as the foam used for insulation, dielectric grease in the socket, etc. I won't even venture a guess for this number. By looking at JoeC's comparison of die simulator to CPU data you can see that this heat path is probably very substantial. Although as I mentioned earlier and JoeC mentions in that article, actual CPU power consumption is not a well known quantity. If you are targeting a CPU temperature near ambient with your design, you can neglect this value. (Set it to its maximum of 1000) Otherwise this is going to be a significant source of error.

Calculate system button - once you've got the previous values setup, click this button and it will calculate the results. The main result of interest is: Object Temperature (Tob). This represents the temperature of the CPU die. (Assuming Athlon. The P4 heatspreader makes thing even more complicated.)

IMO one significant factor the Kryotherm software is missing, is the C/W of the heatpath through the pelt clamping hardware. The steel screws we use for bolting the coldplates to the the waterblocks have a low thermal conductivity compared to copper, but it's a lot higher than air. One thing I don't like about the Maze3-1 design is that there is no space to use fiber shoulder washers to insulate the screws from the coldplate. I don't know whether any other CPU pelt systems handle this better though. (BTW my spreadsheet doesn't account for this either.)

I'd really like to homogenize the whole industry of calculating TEC cooling effect of processors, really dig in and solve the puzzle somewhat.
It would be cool if we all could "chip in" and get a more complex formula, that explains most of the "dark" areas of TEC's.
I(for one) don't strive to be exactly on the 10:th of a degree precise, just to at least close within 5 degrees C (or whatever that is in F).

I could almost bet there's more formulas out there than TEC's !!

Don't you Since87 have a formula on excel sheet or something?
Maybe it's a beginning or maybe it's the end of this debate ??

I've seen hints that the Kryotherm software uses lookup tables rather than equations to some extent. (This is based on very skimpy evidence.) If so, that would imply that even Kryotherm may not rely on equations.

I've never come across in depth equations for doing these calculations. They may well be out there though. The fact that TE Tech provides that paper I linked, as their reference for engineers, does not make me optimistic that such equations are publicly available though.

I definitely don't think my spreadsheet is "the end of this debate". I would like to tweak my equations to match the Kryotherm calculator better. If I could stay within 2C over a wide range of pelts and operating conditions, I'd consider the spreadsheet a reasonably good simulation. I also would like to add an Rins factor, and a factor for heat conduction through the clamping hardware.

I'd be much more motivated to work on this if I knew of someone with the skills and motivation to make a web based calculator out of this information. Anyone?

We need to find out some method of finding out the C/W when using the heatsink or waterblock with a TEC. Ehrrm... Wouldn't I be sweet if the nice staff of overclockers.com would make a peltier die to their cpu-die test equipment??? So that every test with the CPU-die on an W-Block or aircooled heatsink they would include a peltier die test aswell??

That would be nice, but that kind of testing takes a lot of time to do well. It sounds like BillA is going to be reporting some data on WB testing soon. That may be of some use, although the WB itself is just a component of the WC system C/W.

It's also my guess that we need to focus on how to use the graphs that concerns the particular TEC one would use.

That may be, but considering how well my spreadsheet has done calculating results for multiple pelts against measured data and the Kryotherm calculator, I'm inclined to stick with the equations. I find using graphs tedious, otherwise I wouldn't have created the spreadsheet in the first place.

But, a feeling that I've had for long is that peltiers are out, phase change is here to rule. So maybe the intrest in pelteirs are dropping.
Oh well, maybe CoolChips(tm) can change that
I don't know however how big they will be.

I agree. As things are progressing now, pelts don't have a long term future in CPU cooling. Phase change may be standard equipment for maintaining ambient temperatures five years from now. I know electronics and have access to electronics tools, so pelts are what I play with.

I don't quite see how the CoolChips stuff is going to help. My understanding is that the technology requires a very thin "sheet" of vacuum which high excitation electrons can cross. I'm not sure how you maintain that vacuum while clamping the device between two heatconductors. The major value of the vacuum is that it acts as a thermal insulator. If you build in the mechanical supports to maintain the evacuated space, you are bypassing that insulation with something with relatively significant thermal conductivity. The CoolChips stuff is "wait and see" for me.

 

[Toysrme]

Just a quick question.

Is his equation above correct? Just thought I'd make sure before I start doing the math and seeing if they're viable? (or atleast fun to play with <g>) In the future after christmass.

-Toysrme

 

[Since87]

Which equation?

 

[Toysrme]

Nevermind I just read your post closer. Sorry, skipped it out of lazyness. Hey! So I'm atleast honest! Can I have a job too!

-Toysrme

 

[Since87]

Nevermind I just read your post closer. Sorry, skipped it out of lazyness. Hey! So I'm atleast honest! Can I have a job too!

-Toysrme

LOL

 

[Paxmax]

Okay Since87, I'll try the Kryo software a little later, thx a million for your quick-start instructions !
An easy test to get a hint of how much cooling power is lost through the heatpath down through the socket to the motherboard would be to run the pelt with a totally well regulated voltage source, processor and motherboard off, but fans blowing on the motherboard itself. You add some insulation around the socket and backside of mobo.

What you know is how much power it takes to keep the coldplate at maybe 5 degrees below ambient, then 10 degrees below and so forth. You could calculate the amount of "lost" cooling power due to traces/insulation
It would be useful to know the magnitude of that loss, or at least find out if it is significant.
A thing I remember "from the old days" is that it was known to be easier to cool SLOT processors compared to socket.

Of coz there are (as you said) other heatsources on the motherboard that will influence the end result, but for the moment I think the passive cooling area of the traces might be the biggest contributor. The second bad influence would be the traces leading heat from the voltage regulators for the processor.

And, the numbers one get is only good for that particular motherboard type.
I have no clue as of how much it will vary form the next MB. I'm thinking/hoping "not too much".

I totally agree with you Since87, graphs are tedious. I had a dude from "SuperCool.se"(a manufacturer of TECs) mail me an powerpoint presentation of how to use the graphs with current/voltage/Tdif/Cooling Power. I tried to apply it on my own hideous creation of P2 300 celeron + 40W peltier. I just never got the graph to "work" for me......

I have a similar feeling towards peltiers like you, stick to what you know. And like you I'm also an electronics(work with it aswell) fan with a thumb in the hand when it comes to soldering high pressure DIY compressor hoses/gasblocks. Neither have I got the money (or the friends) to solder me a fine phase changer system. So, I'm stuck with peltiers and powersupply(a little problem) instead. Luckily I have access to a milling equipment to make the neccessary WB's and coldplates. My biggest foe however is time. I got most of the other stuff to make me an "nice" high powered TEC cooling rig. I'm thinking like running dual 226W TECs.

Would you(Since87) share your excel sheet or is it to "rough" yet?
/Paxmax

 

[doc]

TECs are not linear devices. If you put 2 times the power in, they do not pump 2 times the amout of heat. The curves are second order. The equations used are only a few and when properly manipulated are a gold mine of information. I have worked with TECs for many years.

 

[Since87]

TECs are not linear devices. If you put 2 times the power in, they do not pump 2 times the amout of heat. The curves are second order. The equations used are only a few and when properly manipulated are a gold mine of information. I have worked with TECs for many years.

Can you post the equations you've used? I've had to infer equations from rough graphs. (No gridlines) Now that I know how to use the Kryotherm software, I can use it to tweak my equations. Having the equations spelled out would save time though.

 

[Paxmax]

quote:
Originally posted by doc
TECs are not linear devices. If you put 2 times the power in, they do not pump 2 times the amout of heat. The curves are second order. The equations used are only a few and when properly manipulated are a gold mine of information.

Exactly, also from what I've read, TECs are(could) also optimized for a specific temperature on the hotside. Some may be optimized for 120 C and others 40 C.
Another thing I've read is that they are more effective when operating with 0 C temperature difference between coldside/hotside, thats a bit dissapointing but what the heck if we can get a factor included about that aswell it'll still be intresting. Please DocJ spill your knowledge on us poor fools

Also Since87(or DocJ) during what circumstances does a TEC pump more heat than input power? (yeah yeah, I still haven't quite absorbed that truth yet )

 

[doc]

Take a look at this graph, tell me what you see. Sorry the graph did not upload correctly.

 

[doc]

Maybe this time!

 

[Since87]

Take a look at this graph, tell me what you see.

I see that for a 36.7 degree C deltaT, (Th-Tc) you can't pump power equal to the input power.

Can you generate the same graph for a 20C deltaT doc?

 

[Paxmax]

Take a look at this graph, tell me what you see. Sorry the graph did not upload correctly.

Well.. as usually I see that you always have to input alot of power to maintain a decent deltaT with a medium heatload.

I never see it pump more heat than input...

The purpose with this graph is??? (beside being a really good graph )

/Paxmax

 

[Since87]

An easy test to get a hint of how much cooling power is lost through the heatpath down through the socket to the motherboard would be to run the pelt with a totally well regulated voltage source, processor and motherboard off, but fans blowing on the motherboard itself. You add some insulation around the socket and backside of mobo.

That would be a good test. I thought about posting an ad on the "Wanted" board to see if I could get a dead CPU and MOBO to do that sort of test.

Wayne Harrow said in this thread that he would be willing to do some testing that would help determining a rough value for the Rins factor. By subtracting the simulator's curve, from the curve generated by Wayne's data I should be able to come up with a correction factor that improves accuracy substantially. (As long as the error curve is reasonably close to a straight line. If it's not a fairly simple curve then there are probably substantial factors being missed.)

Wayne's data would have the advantage of including other heat sources in the picture. One disadvantage of using the kind of testing I talk about in that thread, is that the calibration of the temp sensors is very suspect. It's still better than the total lack of data I have now though.

There would certainly be some variation from MOBO to MOBO, (for example I've heard the regulators run hot on Epox boards) I think the variation would be small enough to neglect and get pretty good predictions though.

My biggest foe however is time. I got most of the other stuff to make me an "nice" high powered TEC cooling rig. I'm thinking like running dual 226W TECs.

Time is my problem too. (I spend too much of it reading and writing posts here.)

I was going to go Maze3 on top of a 172 Watt pelt, but after reading about Cathar's White Water block over at OCAU I decided to buy one of his waterblocks. It looks like his block is the best way to move heat off the CPU short of direct die watercooling. I'm going to build a dual 172 Watt TEC water chiller to cool the water going through the White Water block. I've got one Maze2 block to use for the water chiller, and I'm looking to trade my Maze3-1 for another Maze2.

I don't want to deal with condensation issues, so I've got an electronic humidity sensor, and I'm going build a system to keep the water temperature just above the dewpoint. The dewpoint in my computer room is probably around 7C most of the time here, so I should be able to keep the CPU subambient with no worry about condensation. I'm using 172 watt TEC's because their higher voltage and lower current are easier to deal with in a switching controller.

Would you(Since87) share your excel sheet or is it to "rough" yet?

Give me a few days to get it tweaked in a little closer to the Kryotherm results (and maybe put in an Rins factor) and I'll post it.

If I don't get around to making the changes soon, I'll post what I've got.

 

[Paxmax]

Cool Since87 ! Way cool !
*eagerly looking forward* for the excel sheet.

Of coz, white water blocks are good... they are designed "in parallell" just as all blocks should.
I haven't figured out why ppl use spiral/maze design... all you end up with is a long single channel with high resistance for the water. Ok, ok... it looks nice with a spiral/maze... have to give it some credit

I wish DocJ would show up with some wisdom.

/Paxmax

 

[doc]

Hi guys, sorry I couldnt get back to ya yesterday. That graph was not the best choice, so I found a better one. TECs pump more heat per watt input at lower power input. This ratio is called COP or Coefficient Of Performace, the COP depends on power input, heat being pumped and delta T. These conditions dictate what the COP will be. Someone wanted to use ohms law to calculate the power input at other voltages. That would not work, because of the seebeck effect. These devices will generate a voltage and current if you have delta T on them. If you dont believe this try it. Heat one side and cool the other, while measuring the voltage output. Anyway the seebeck effect is why just measuring the resistance and using ohms law will not work. You need to measure the seebeck coefficient, which is just volts per kelvin. But this can change as a function of temp so its not so easy. The equation I made mention of before goes like this:

Qc=aITc-.5I^2R-K(Th-Tc)

Qc=heat pumped (watts)
a=average seebeck coefficient (volts/kelvin)
I=current (amps)
Tc=temp on cold side (kelvin)
Th=temp on hot side (kelvin)
K=thermal condutance (watts/kelvin)
R=resistance (ohms)

 

[doc]

new graph.