38W pelt ok for northbridge, aircoolable?

[[SLC]Tachyon]

Would it be possible to slap it together on the NB with a Socket A hsf and air-cool it (and would 38W be ok for the NB?)?

[saaya]

875? i think so... macci uses a 30W tec on his nb as well, and even a old celeron hsf can cool it...

[Sick0fthelies]

Im sure it would be fine in terms of being able to cool down the nb.

[[SLC]Tachyon]

Sounds good. Once I have my phase-change hooked up I don't really want to have water hooked up at the same time just for the NB.

[KnightElite]

Yeah, should work good. Just make sure you pick a heatsink that's not so enormous it won't fit on the northbridge :D.

[Stang_Man]

what about a retail amd heatsink with a fast 60mm fan cooling a 38w pelt on the nb?

how cool will the pelt cool it too?

(nforce2)

[KnightElite]

Would probably cool it pretty good... a NB with 38W TEC will still produce less heat than a CPU, I would think.

[Aldus]

How heavy of insulation (if any) would a 38W TEC require if being cooled by a decent heatsink, such as a sk-6+ ?

[CCW]

I presume a 38W TEC can effectly cool something that dissiaptes up to 38W's of heat?

If so, will the hot side of the TEC only produce 38W's of heat? If so most heatsinks and fans can handle this so will be fine for cooling the TEC as most processors put out around 70W's of heat.

Craig

[KnightElite]

The hot side will produce voltage of TECx Current used by TEC watts + however much heat the NB is producing.

[Teus]

let's say it'll produce 60-70Watts. with a silent fan you won't need to use a lot of isolation, as the TEC prolly won't go below ambient and create condensation ( if it does, lower the fanspeed)

you can also put a powerful fan on it if you want to push it to the limit, but you'll certainly have condensation problems

[afireinside]

Where are you getting a 38 watt pelt? I'm intrested in pelting some ram :smileysex

[[SLC]Tachyon]

http://www.mpja.com/productview.asp?product=12323+PM

Havn't ordered one yet since I havn't found a small heatsink that'll fit on my 865PE NB.

[afireinside]

"L: 1-3/16" W: 1-3/16" T: .13 WT: .03" Um how big is that? Why can't they use frikkin decimiles?

[KnightElite]

1.1875. Just do 3/16 in a calculator, and add 1.

[afireinside]

Only a bit bigger than an inch eh? Maybe I'll pelt my video ram :D

[Holst]

Originally posted by CCW
I presume a 38W TEC can effectly cool something that dissiaptes up to 38W's of heat?

If so, will the hot side of the TEC only produce 38W's of heat? If so most heatsinks and fans can handle this so will be fine for cooling the TEC as most processors put out around 70W's of heat.

Craig

You presume wrong.

TEC are inherantly inefficent.

most people recomend doubling your heat scources wattage when you choose a pelt.

So for a 38watt heat scource you want a 70watt pelt. that means 100+ watts at the hot side. Not so nice...

if your in the UK www.greenweld.co.uk have some OK priced small pelts.

[lalPOOO]

I have great doubt that the northbridge creates much more then 38watts of heat though. So going for a 70watt pelt would probably mean he'd have to go through all the condensation proofing.

I know the princess auto near me carries the old socket 423 stock intel heatsinks for something like 8-15 bucks. With a little help from a saw (for fitting) that thing would probably be able to cool the hot side pretty well.

[Jabo]

TECs have certain design workig parameters:

Imax ->max current in amps they can draw

Vmax ->max operating voltage

Qmax ->max heat energy pumping capacity

dT -> max temp difference between cold and hot sides with no heat load applied.

Now, let us take 169 watt TEC

Imax = 16.1 Amps

Qmax = 169 Watts

Vmax = 16.4 Volts

Delta Tmax = 69 (C)

This element pumps 169watts of heat from 'cold' side to 'hot' side @ rated Imax and Vmax.

At 12 volts it works with 73% of it's rated efficiency being able to pum 123watts of heat energy.

What it means?
If it's working within rated environmental conditions (-50C to +200C) it'll cool to ambient (or coolant's) temperature a heat source producing 123 watts of energy. It it's used to cool a heatsource producing less heat (i.e. overvolted NB producing 60watts - my guess) it will cool it below ambient (coolant's) temperature using 'remaining 63 watts of rated heat pumping potencial to get it there.

You can only say that TECs are inefficient if you look at their electrical energy consumption, they are just power hungry beasts, that's all. It has nothing to do with their cooling efficiency.
Size = 50mm X 50mm X 3.4mm

[]JR[]

Originally posted by Jabo
This element pumps 169watts of heat from 'cold' side to 'hot' side @ rated Imax and Vmax.


Yes but at qmax dT is zero rendering the unit a waste of time and energy


Originally posted by Jabo
What it means?
If it's working within rated environmental conditions (-50C to +200C) it'll cool to ambient (or coolant's) temperature a heat source producing 123 watts of energy. It it's used to cool a heatsource producing less heat (i.e. overvolted NB producing 60watts - my guess) it will cool it below ambient (coolant's) temperature using 'remaining 63 watts of rated heat pumping potencial to get it there.


No, the efficency drops with hotside temperature dropping, it infact drops dramatically, if it pumps (say) 100watts of heat @ 20C hotside, it will only pump around 15watts @ -50C hotside (im guessing that figure in all reality its probably worse).

Running tecs @ imax/vmax also is its lowest efficency state, infact you can get efficencies > 1 when you use a big tec at low volts/amps, i.e. more heat pumped than energy input.


Originally posted by Holst
most people recomend doubling your heat scources wattage when you choose a pelt.


Thats because we want lowest temperature not maximum heat pumped, so we go for the biggest module possible irrespective of the heatload to get the best efficecny and dT.

]JR[

[CCW]

Originally posted by Holst
You presume wrong.

TEC are inherantly inefficent.

most people recomend doubling your heat scources wattage when you choose a pelt.

So for a 38watt heat scource you want a 70watt pelt. that means 100+ watts at the hot side. Not so nice...

if your in the UK www.greenweld.co.uk have some OK priced small pelts.

Thanks.

TBH, Intel chipsets dont really need pelts. They do fine with air. Due to their manufacturing process, Intel chips create far less heat than AMD chipsets. Not even mentioning the fact that some have a plastic coating which doesnt aid much to heat dissipation. :eek:

Craig

[Jabo]

[Q
Originally posted by ]JR[
Yes but at qmax dT is zero rendering the unit a waste of time and energy

Why? Temps of your heat source (CPU) will be equal with temp of your cooling media (air, water, chilled water) – heat source gets cooled anyway. It’s a matter of sizing your element as I wrote further down in my post – want to cool it sub ambient get more powerful module :)




No, the efficency drops with hotside temperature dropping, it infact drops dramatically, if it pumps (say) 100watts of heat @ 20C hotside, it will only pump around 15watts @ -50C hotside (im guessing that figure in all reality its probably worse).


From what you are saying it looks like we should heat ‘hot sides’ instead of cooling them to get more heat pumped ;)

How did you arrive at this numbers and where did you find any info on TECs efficiency dropping with environment (‘hot side’ @ lets say –20C and extremely good insulation) temps dropping? Post some links pleeez or write down some equations if you like. Dump our mobo with cpu cooled by swiftech’s air cooled 226watt TEC into a bath of fluorient @ -20C and measure CPU temps….

Visit any manufacturer website and check multistage assemblies and temps of cold side and of what elements are they made.



Running tecs @ imax/vmax also is its lowest efficency state, infact you can get efficencies > 1 when you use a big tec at low volts/amps, i.e. more heat pumped than energy input.


That’s true but at max volts it pumps the most heat, irrespective of electrical efficiency, we want it cool not energy efficient :). If was to be energy efficient I’d use photovoltaic cells to power my machine :)



Thats because we want lowest temperature not maximum heat pumped, so we go for the biggest module possible irrespective of the heatload to get the best efficecny and dT.

]JR[

:rolleyes: The more heat you pump the lower the temp you’ll get. 226watt TEC @ 12v will give you –7C (my case) the same TEC @ it’s rated volts would give you ~-9C.All with sttraight water cooling
Imagine now I am using chiiled water and getting –25C CP temp @12v. Upping voltage to rated 15 I am getting now ~–31C.
All of the above is assuming that ‘hot side’ (coolant) is at the same temp with 12v and 15v TEC, meaning very efficient cooling system :).

Efficiency and dT are inversly proportional I‘m afraid. The bigger dT the worse eficiency of your module since it needs more power = produces more ‘own’ heat to achieve it’s max dT. At 12v max dT without any load (no heat pumped) is 80% of rated 62 C (at 15v) and equals 49C. Add volts and dT will increase but energy efficiency of the whole cooling unit will drop dramatically….

[]JR[]

By sheer definition of every tec manafacturer qmax is the amount of heat pumped to make dT = 0, conversly dT is the value given at where q = 0, thus the more pumped heat the lower the dT.

http://www.phjrw.34sp.com/172per.jpg

The graphs show the drop in dT w.r.t Q, for a 172w 24v module.

http://www.phjrw.34sp.com/172per2.jpg

That graph shows that the current draw is dependant (and not linearly) on the hotside and cold side temps, which is the beginning of my argument towards the losing efficency and q pumped ;)

Next take kryothems (http://www.kryotherm.ru/soft.zip) software to simulate some tec situations, kryotherm are the industry leaders in tecs...

Here is some screenies of the simulations...

http://uffish-thought.net/wc-gifs/hi-cop.gif

This shows our favorite the drift 0.8 (24.6vmax 172w) powered @ 3v and 1.51amps pumping 25watts of heat, oh the in-efficency.

Once youve downloaded the software go thru and setup a system calculation, 1 drift 0.8, 18v, wob 50, 0.1 for hotside and coldside parameters and 1k/w for the insualtion. All of which are realistic for an overclocker cooling a 50w gpu.

With an amb of 50c the cop comes out at 0.462, 20c 0.413, -20c 0.348, -50 0.299. cop is q / (v * i) youve already seen that i varies with the hotside and coldside temps.

Now any simulation you do with a drift 0.8 is over 25% better than any other overclocking sized module.

But have a play then come back to me with an argument.

]JR[

[Jabo]

Cheers mate, great stuff, will have lots of fun with it, serioiudly, thanks man :)

I love when sb can come back to me with VERY solid argument, the trashing afterwards is so much sweeter ;)
Just joking :)

I love to be PROVED wrong, have no probs with asimilating new knowledge which is in contradiction to my current state of info :)

Off to have a play with your Kryotherm soft :)

See ya soon :)

and thanks again :)

[Jabo]

There's sth wrong with this software or I have not yet got deep enough to get reliable numbers. It shows some obviously wrong results with basic default input. Gotta work on it a bit to check it out properly.
Russian friends might just got sth wrong in terms of non standard applications ;) :)

[Jabo]

As far as graohs are concerned...

I worte in some other threads that TEC working @ lower voltages draw lower amps and therefore despite very high Imax values they can be run be regular PSUs (second graph).

First graph is also perfectly true.
It all boils down to data interpretation.
dT is max when heat pumped equlas zero.
What it actually means is that the biggest temp difference between hot and cold sides is when there's no heat applied which is of course obvious :).
Make no mistake here because 'heat' is still being pumped between 'cold' and 'hot' sides which manifests itself in dT 'maxed out'.
Work is being done and 'heat is being pumped ('cold' side gets cold, doesn't it?)

Now apply heat to 'cold' side.
The amount of 'pumped' heat is still the same, nothing changes here. To make it work (physics equations) 'cold's' side temp has to rise to keep system balanced.
If heat input to 'cold' side does not exceed 'heat pumping' capacity of your module cold side goes below coolants temp.
Lowering colants temp will lower cold side temps proportionally. There's a cut off point described by working temp brackets of particular module.
Another issue is loss of TEC efficiency with below ambient temps (my water is @ 30C and I get my CPU ful load temps @ ~10C; getting water to lets say+10C should drop my CPU temps by 20 degrees to -10C, right? Getting pelts 'hot' side to 0C should get my CPU to -20C full load and -36C idle...

[]JR[]

The software is right, trust me ;) Its the benchmark, industry standard software infact.

Also run the simple simulation I described and you will find that as the hotside temp drops, so does the current drawn, which means we are giving the tec less power (it doesnt behave like a resistor) which lowers our qmax intuitevly. Again graph 2 shows the current draw for the max and min scenario q = 0, and dt = 0. The lines are a) not y = x and b) not = to each other. This alone can account for the drop in power at lower hotside temps.

But to add to that the efficency drops also with the hotside temp, even if you were to magically make the tec draw the same i-v from the psu at 20c hotside and 0c hotside (although ive already proved you cant) given everything the same apart from the hotside temp drop it would still be worse in qmax and dt than at 20c.

Couple these effects together and you see large drops in cooling potential at low hotside temps. In your example 30c Thot 10c Tcold, drop Thot to 10c and Tcold wont be -10c it would be -6c or alike with a qmax of 140w rather than 172w (im guessing here, i dont have the simulation software at work).

Now also with the simulation software its quite easy infact, to setup a case where you Tcold is hotter than Thot running a tec at full power with a heatload approaching 60% of qmax. Now this is the point which OPP and the other guys reached with there uberchillers (before abandoning them for direct die) where they get better temps with chilled water and no tec than with a tec.

]JR[

[Jabo]

OK, finally I am getting somwhere here :)

So efficieny drops because at lower temps thermoelectric module draws less power.
That seems to be the bottom line here.

The only thing left for me to understand here is why on earth do they draw less power.

Is it because with dropping temps each condustor/semi-conductor looses resistance and becomes more efficient and doesn't 'need' as many amps?

Damm, I HAVE TO UNDERSTAND WHY is it happening, what's the physics behind it??!!!

[]JR[]

No but theyres also a secondary loss in efficency also which is why going below -50c with tecs is pretty much impossible irrespective of how big your tec stack is (or hotside coolant). And the efficecny loss is quadratic iirc.

The physics behind it is due to back emf i believe since in essence a tec is a big a$$ thermocouple, i.e. they generate a voltage given a temp difference. Were creating a temp difference with water and cpu, and then by dropping the temp were dropping the back emf or the resistance to feeding it current.

If you disconnect your tec from the psu and then turn the pc and watercooling on, then voltmeter the tec itll have a fairly big voltage and a reasonable current! ;)

]JR[

[Jabo]

ad a play time with Kryo soft.
Got it cracked and it's not really suitable for PC cooling purposes (btw, it works fine an there's nothing wrong with it ;) )

My goal was to test TEM's performance under below zero temps and how it's changing with lowering temps to some ridiculous levels (theoretically).

Kryo's soft is based on both cold and hot sides having radiators/heat sinks of dif fescriptions on both sides. I wanted to judge Thermoelectric Module's performance without any consideration towards thermal properties of the rest of the system.

I set hot side to be water cooled (gives me 100% control over hot side temp) and cold side to convection (natural) cold dissipation and both hot and cold heat exchangers' thermal resistance to 0.001K/W (to minimise their influence on my results). Cold's side insulation was also made as big as possible (lowest heat transfer ratio)with simultanious compartment size reduction.

What I got in result was a system where cold side thermal losses/gains were reduced to almost zero and heat resistance was also close to zero.

'Hot' side's parameter which mattered was water temp.
Ambient temp was irrelevant in my setup which is what I wanted.

Here are results for –10C water

http://www.hpphotos.com/servlet/LinkPhoto?GUID=2fe0192b-ebed-1a95-584a-5c7f32cb290c&size=

And –15C water temps

http://www.hpphotos.com/servlet/LinkPhoto?GUID=ec7b579b-3d94-253f-8ec7-6c1f5c1875aa&size=

Load was 50watts in both cases

Changing water temps produced results to contrary with what JR said, at least in some respects ;) :)
Amperage drawn by module increases instead of decreasing with lowering water temps. The lowest I could get was -110C with water @ -110 LOL - going any lower produced increase of Tob. At -35C Tob was -53.8C. :)
-15C temp of water seemed to be a sweet spot :).
At +29C water temp (That’s what my current system is producing at the mo. Tob. was –2.1C
Performance of this module decreases with lowering temps much below zero but within easily attainable range of self made water chillers is very good :)
-37.5C at –15C water temps is imho a fantastic achievement :). If you make it more realistic and account for thermal losses along the way (will get it done over the weekend and ‘real life’ calcs will be ready for Monday) you should be getting sth along –27C.

[]JR[]

Now your beginning to realise how many variables are involved in a tec.

And dont forget a best fit of two points is always a perfect straight line it takes 3 to prove your point ;) Youve found another oddity case there where the current goes up when the temp goes down, the other one is the example of a cop > 1 like i showed.

The final point being, your operating in a dream world with your parameters ;)

But welcome to the world of tecs where you understand theres more to it than just vmax/imax and qmax...

]JR[

[Jabo]

Cheers mate :)
Thanks for showing the way ;) :)

Yea, what I showed above was only to test this module by 'itself' :)

As I said I'll try to make real life calcs interpolating design parameters from my system :).

Now, I know how it works exactly I wonder what's more effective:

1. Water chiller which keeps water @ -35C or TECs with -15C chilled water.
I am talking here money, and availability (I am not going to search the Commonwealth for some obscure medical/lab chiller selling for penuts ;) )