Peltier-based water-chilling experiment

Hi All,

This is the first stage of a longer term project I've got going. I picked up 2 x 172W 40x40mm peltiers today and setup a peltier-based water-chiller using my first Hydra prototype block (no - not the production prototype - just the first concept prototype I had made some months back).

Basically I've just conducted a non-heat-loaded experiment where I have an Eheim 1048 pump in a bucket, which pumps water to the Hydra block on the cold side of the pelt, and then back into the bucket again. The hot side ofthe peltier is cooled with a spare water-block I had kicking about which I just plugged into my main computer water-cooling loop so it would be cooled by the computer's main radiator.

The peltier was driven off the 12V rail from the PSU. Now this isn't exactly best for the peltier as it's rated for 24V, so the peltier is only running at half-voltage. As I bump the voltage there should be a good deal more to come from it.

Okay, so I got the system fired up with 4l of water in the bucket. The rate at which the water temperature was dropping is indicating that the peltier is removing heat at the rate of about 80W from the water. Not bad for a prototype block on the cold-side and for the hot side of the peltier not really being cooled that well, and the peltier at half power.

I got the water to ice solid in about 90 minutes, which if you think is crappy, then ask yourself how long does it take your freezer to get a room-temperature 2l bottle of coke to turn to ice?

I know you want to see them, so here's some piccies.

The Setup:









...and after the icing of the reservoir:



The water remains fluid until about -2.2C until the water ices up in the block. This causes the pump to stop, and then the currents stop in the reservoir, and over about 3-5 seconds the whole reservoir turns to ice and the temperature jumps up to -0.1C when the phase-change takes place. Very interesting to watch it happen.

This is just stage one of the experiment.

The end goal here is to power the peltiers properly (~18V) and hook them up in series with 4 blocks to provide around 3x the cooling power of this single peltier test. All up I'm aiming for 220-240W of cooling power to apply to the CPU and GPU water-blocks (which in total push out about 100W - being real watts - not "radiate" watts). Given that the 80W of cooling power of this test was able to bring the coolant down to -2.2C before it froze over, I'm hoping that with an alcohol/water mix and the surplus 120W of cooling power (220-100) of the final setup will allow me to achieve -10C (or maybe lower) coolant temperatures once I get the proper Hydra blocks onto the job.

ive olny flicked thru it, but thats lookin pretty sweet, nice figures 2. im suprised no ones replied yet, [praise]after all it is the god himself cathar[/praise], do i get a free WW now ??

I ain't no god. If I see further, it's only because I stand on the shoulders of giants who came before me.

LOL - a free WW! Tell you what, I can send you the raw bits and if you want to lap and polish it, put the barbs on yourself and assemble it (takes about 2 hours all up), you can have it for about 55US + shipping.

Hello,

I have been posting about doing something like that in this thread [ This Thread ] and there have been a lot of good ideas that have came up, take a look if you have time.

And when you finish, let us know what the temps are!

Hey, that is Newton's line!

Cathar said:

I ain't no god. If I see further, it's only because I stand on the shoulders of giants who came before me.

Click to expand...

now, as for the main subject, why dont you applied the peltier directly on the cpu and gpu?? It should be more efficient, no??

Why not do both, chill the water with a peltier and use a peltier on the CPU?

Prandtl said:

Hey, that is Newton's line!



now, as for the main subject, why dont you applied the peltier directly on the cpu and gpu?? It should be more efficient, no??

Click to expand...

Compared to Newton, I am an intellectual insect burrowing in the dirt. That guy was a true genius, and a humble one at that.

As for applying the peltier directly - well here's my take on it. It probably is more efficient when one simplistically thinks about it. A copper cold-plate on a CPU offers its own spreading thermal resistance. In order for a ~130mm^2 heat source (which is what a P4/AMD/ATI/nVidia silicon die looks like) to adequately spread its heat over an area the size of a peltier. Research seems to suggest about an 8-10mm thick copper plate is required. Any less and the heat isn't spread wide enough. Any more, and the copper starts to act an an insulator compared to the thinner thicknesses (albeit an extremely poor one). For more information on what I'm talking about here, refer to this article:

http://www.overclockers.com/articles699/

We can see this with the White Water CPU block next to thick copper plate designs like the Swiftechs. Now a Swiftech is actually better than an evenly cooled copper slab because it has dimpled and the water is focussed on the hotter middle section of the block, whereas a peltier cold-plate is cooled evenly by the peltier. Conceptually speaking, the total thermal resistance offered by a peltier cold-plate over something like a White Water CPU block can mean up to 10-15C difference for the CPU. At least that's the figure that I'm placing on it.

Now with a water-exchanger, many people have also been using blocks like Maze 2's, which in my mind are remarkably inefficient metal-water heat exchangers. I intend to use some very finely detailed micro-channel blocks for the cold-side of the peltiers (also on the hot sides too) which in my mind will lower the thermal gradient difference between the peltier and the water coming into contact with the waterblock's surfaces. In essence, the hot side of the peltier will be cooler, and the cold-side of the peltier warmer.

Now the cold-side of the peltier warmer may seem counter-intuitive to some, but what it means that is that water that touches the cold-side block will more closely match the peltier's cold-side temperature, which as we know can be up to 65C below the hot-side temperature.

Why is this important? Because for a peltier, the less the difference between the hot/cold side, the more efficient the peltier given the same power source. I refer you to this article as research-based evidence:

http://www.tedist.com/publications/pubs/ICT94MJN.pdf

So now we have a more efficient peltier-water heat exchange which boosts the efficiency of the peltier, and a more efficient CPU->water heat exchange due to the use of a White Water CPU block.

In the past, when using say a Maze2 for the CPU, and for the peltiers, the total heat exchange efficiency at all points is much lower. In this instance, I'm not surprised that people commonly think that applying a peltier directly to the CPU with a cold-plate is more efficient. Problem is though is that after adding one peltier (a 226W 50x50mm peltier is about the biggest) you gain very little by adding more, and it's hard to find room for them.

In this instance, as many pelts as are desired can be added to the loop, all providing in my mind a greater net efficiency boost to the cooling.

The whole thing hinges on using appropriately designed heat-exchange blocks for the critical elements in the system. Of course, the extra hassle here is that the cold-loop needs to be insulated, but that's easily achievable with rubber insulation hose wraps like what's used in cars, and then the standard peltier insulation techniques that are already used.

Well, that's my theory, and that's what I'll be doing. Happy for anyone to try and shoot it down, but be warned, don't be referring to some old Maze 2 experiment 'cos the rules have changed.

Cathar, I don't know why you had to post such a long answer, when the answer is simply:

The CPU can't handle the clamping pressure that a TEC requires, to run properly!

Article by OnDaEdg here

Cheers!

bigben2k said:

Cathar, I don't know why you had to post such a long answer, when the answer is simply:

The CPU can't handle the clamping pressure that a TEC requires, to run properly!

Article by OnDaEdg here

Cheers!

Click to expand...

Who would ever mount a peltier to a block in such a fashion that the CPU is used to push the peltier against the block/cold-plate? That's just silly. A peltier + cold-plate simply MUST be clamped independently of the mounting onto the CPU. My above narration of the process assumed at least that to being with.

Wow, that is a hell of an answer and back up with very good arguments and facts. I have to admit, i had not think it throughout before posting (espicially in regards of the pelt surface vs the die surface)

Also, one of the reason that will had me think it would be less efficient was that:

Cathar said:

Of course, the extra hassle here is that the cold-loop needs to be insulated

Click to expand...

but it looks like you can handle without any major loss, so i guess you are all set up.

but that's easily achievable with rubber insulation hose wraps like what's used in cars, and then the standard peltier insulation techniques that are already used

Click to expand...

All that make me wonder what your educational background is, did you self educate yourself on that matter, or did you get a formation on heat transfer (mechanical engineering maybe?)

I have nothing really special to say here, so you can skip reading this if you want.

Cathar: Thanks for doing this. I've been building a theoretical water cooled system that was designed around a future peltier upgrade. When reading about peltiers this concept was the first thing that crossed my mind, but I've read numerous posts (on this board no less) that the idea was completely unfeasible. The posters who refuted it backed the idea with some mathematical and graphics wizardy that confuses me eh? but seemed sound.

Keep up the good work. I'd like to see this idea as a standalone upgrade to any water cooled kit, so someone could 'plug-in' low temp water to their system.

.:: epoch of entropy

@Cathar: On coldplate thickness and other relevant peltier information: http://www.overclockers.com/articles305/

I'm not sure if this coincides or not with the mm measurements that you noted cathar, but if you aren't aware of this article it is a good one and applies well to what is being discussed.

It's exciting to me that you have taken interest in this... i have previously stated, with much less knowledge on the subject, that i thought pelt chillers could be much better than direct applications of a pelt on the cpu. The more i have learned around here, the more convinced of this i am. You have some good points, and although as you said some see counter-intuitive, when read closely enough they can make perfect sense. my next post will concern places where you may be stearing off course a bit AFA my knowledge covers the subject.

You are likely familiar with wymjym's "not too extreme, water chiller" project... i dont remember if i have referenced it before around you, but i have referenced it to people considering this application many times, it is some good information on the matter. here is a link: http://www.overclockers.com/articles305/

For reference on any peltier calculations you may be concerned with while working on this project, doc is the most helpful and knowledgeable member of the forums. a forum search using his name will also turn up more information on peltier applications and calculations. virtually all threads he has participated in have been TEC threads i believe.

other people who have participated in informative threads i have read on peltiers are paxmax and since87... one thread they participated in explained the difficulties of calculations and the problems peltiers face below zero and why. forum searchs using these names and certain keywords may also help if you are looking to do any further research.

possible errors as i see them

Cathar said:

1) Now with a water-exchanger, many people have also been using blocks like Maze 2's, which in my mind are remarkably inefficient metal-water heat exchangers. I intend to use some very finely detailed micro-channel blocks for the cold-side of the peltiers (also on the hot sides too) which in my mind will lower the thermal gradient difference between the peltier and the water coming into contact with the waterblock's surfaces. In essence, the hot side of the peltier will be cooler, and the cold-side of the peltier warmer.


2) In this instance, as many pelts as are desired can be added to the loop, all providing in my mind a greater net efficiency boost to the cooling.

Click to expand...

1) To avoid error for others reading this, the keywords here are "in essence". He is saying that the water in the cpu cooling loop will be passing its heat to the cold side of the pelt better (due to improved block design) and hence the coldside will be taking more heat. He is also saying that the water in the pelt cooling loop will be taking more heat from the hot side of the pelt better and hence the hotside will be cooler. He is absolutely correct here. But actually what would be observed, if this works according to the theory expressed, the hotside of the pelt will be cooler than before and although the cold-side of the pelt will be warmer, the water will still be cooler than before.

2) This, in my mind, may be wrong, but as you stated it is how it would work, in your mind. Using multiple pelts of the same power will not improve cooling, this can be seen in the application of peltier stacking. When peltier stacking is used (not in computer cooling, the instances in which it has been used around here are COMPLETELY flawed), peltiers are stacked with differing and carefully chosen power rateings and this can contribute to better cooling. However, if you stack multiple pelts of the same power rateings this will only detriment your cooling capability - the pelts used have to be carefully chosen in order to increase efficiency and cooling capability. In a water cooling loop, even if the pelts were not directly stacked and were relatively remote to eachother, the same principle would accurately apply. I have read some pelt industry/educational research material on peltier stacking that i found through a simple search on google, and i'm sorry i dont have any direct links to give you on this information, but i'm confident it can be rather easily found from a google search. edit: this may explain sufficiently, but i have read more thorough accounts: http://www.overclockers.com/tips313/

Hope that helps.

This is a very interesting idea, I like it. I am building a water chilled system soon and have seen many good ideas thus far in the above posts. I have one question, a pelt will work better in this setup then directly placed on the CPU? Also, I have talked to IMOG about something kind of like this, though I am going to use a A/C compressor to cool the water, then i am going to use a huge radiator/fan setup to cool the hot side of the coils coming from the compressor in a second res. Cathar, keep us posted. Also would you get better temps with this pelt idea or an A/C compressor setup? I like how much more compact yours would be.

skahtul said:

1) I have one question, a pelt will work better in this setup then directly placed on the CPU?

2) Also would you get better temps with this pelt idea or an A/C compressor setup? I like how much more compact yours would be.

Click to expand...

1) The theory Cathar is operating upon is that this setup can work better, if done right, than a TEC directly place on the cpu. His justifications for thinking this way are solid.

2) With a compressor setup done correctly the temps that can be achieved are much better... All the way down to ~40C is possible and realistic if employed correctly. With a peltier chilling setup done correctly, the most optimistic of possibilities would be -20C, but this is not realistic at all. However the ease with which the peltier chilling can be employed in a setup is far greater and requires far less knowledge, tools, time, and work.

I have built a water chiller as well. I made a copper plate 4 inches wide and 8 inches long. The plate is hollow with an internal height of 3/8 inch. Attached to one side of this plate are three pelts. The pelts at sandwiched between this copper plate and 3 K7 HSF. As the water flows through the copper plate is it cooled. The K7 HSF get rid of the heat from the hot side of the pelts. Currently I am testing the pelts at 9.6 volts. I have programed an Allen Bradley Control (PLC) that monitors water temp at the copper plate, CPU, case temp. The PLC then turns on as many of the pelts as I need to keep my CPU temp 2 degrees about the dew point. I don't have to worry about condensation with this method. This system is very very quiet and can remove a lot of heat. I used my test water block clamped to a piece of 3/8 steel and heated the steel with a torch .....Water coming out of the water block was very warm but was very cold going back into the WB after running through the chiller plate. Currently I am working on a decent way to mount this gear inside my Lan box.


Specs in my signature are for my main computer and not for my Lan box as described above.

IMOG
1) The cold-side will be warmer at equal Voltage, giving higher current but higher Cooling Capacacity (Watts) at a higher(better) COP.
2) The number of Peltiers is an essential design element. The COP of a Peltier decreases dramatically with Voltage. Basically, many at low Voltage is superior to few at high voltage.

See Kryotherm's "Kryo TESC" for (1) and "Choice Module" for (2) - downloadable from http://www.kryotherm.ru/soft.htm.

Cathar
I think you "figure" is a way out.

Cathar said:

........................Conceptually speaking, the total thermal resistance offered by a peltier cold-plate over something like a White Water CPU block can mean up to 10-15C difference for the CPU. At least that's the figure that I'm placing on it.
.............

Click to expand...

The total thermal resistance offered by a 10mm thick peltier cold-plate(40x40mm) over something like a White Water CPU block will theoretically* mean up to 2.5C for the CPU(real 100W load).

* Calculated for the "example" 130sq mm CPU with Waterlloo ( http://www.mhtl.uwaterloo.ca/cgi-bin/strip ) for h=120kW/m*m*c. 120KW/m*m*c is considered to be the possible value in high flow conditions(12 LPM) for a WW type wb. This difference will drop to ~ 0.5C at low flow (2LPM0

I agree with your conclusions about increasing the voltage to the pelts.....With the pelts I have anything over 14 volts just produces more heat on the hot side while not really cooling the cold side any more......I cooked a few power supplies playing with different voltages and ended up making my own varible output power supply. At 9.6 volts my pelt will freeze a drop of water in 3 seconds.

Les56 said:

IMOG
1) The cold-side will be warmer at equal Voltage, giving higher current but higher Cooling Capacacity (Watts) at a higher(better) COP.
2) The number of Peltiers is an essential design element. The COP of a Peltier decreases dramatically with Voltage. Basically, many at low Voltage is superior to few at high voltage.

See Kryotherm's "Kryo TESC" for (1) and "Choice Module" for (2) - downloadable from http://www.kryotherm.ru/soft.htm.

Cathar
I think you "figure" is a way out.


The total thermal resistance offered by a 10mm thick peltier cold-plate(40x40mm) over something like a White Water CPU block will theoretically* mean up to 2.5C for the CPU(real 100W load).

* Calculated for the "example" 130sq mm CPU with Waterlloo ( http://www.mhtl.uwaterloo.ca/cgi-bin/strip ) for h=120kW/m*m*c. 120KW/m*m*c is considered to be the possible value in high flow conditions(12 LPM) for a WW type wb. This difference will drop to ~ 0.5C at low flow (2LPM0

Click to expand...

1)corrected my previous post.

2)i think i misread, i was thinking of additional pelts in additional loops so that it would be equivalent to stacking... sorry. more pelts at lower voltages side by side would obviously be better.

thanks, i need to quit posting with lack of sleep.

ve6jhc said:

This system is very very quiet and can remove a lot of heat. I used my test water block clamped to a piece of 3/8 steel and heated the steel with a torch .....Water coming out of the water block was very warm but was very cold going back into the WB after running through the chiller plate.

Click to expand...

i do not know if i could believe this without witnessing it myself. pelts have been shown to be unsuccessful when the hotside is air cooled numerous times.

disregarding that, your second statement (after the "...") is almost completely unbelievable. Unless the flow of your system is terribly slow, it is impossible for the water going into the chiller to be VERY warm going in and VERY cold going out. The difference between water temperature within a loop has been repeatedly shown to be around .5C in scale, and that is with pelt use... the effects of pelts are by no means instantaneous, their temperature at the cold side is no where near low enough to instantly remove that much heat - so it is the recirculation of the same water that makes the overall average water temperature lower. maybe your water flow was very very slow, i still dont think that is a legitimate account. i wouldnt say anything if it werent for the fact that it could mislead those who dont understand how pelts work. SORRY if i am wrong on this.

i wish i could see how you were doing it... i know most people dont use a pelt chiller block the size you are using, or pelts at the settings you are using them... it just seems incredible to me that this could be working the way you say. are you sure you are getting your temp readings correctly so you know that there is a large difference between water going out of the chiller and water leaving the wb?