[teatchme]

So since my previous thread/build i was tossing the idea on how to improve my chiller for quiete a while and during my research came over a very interresting build on coolitsystems forums which was a dual boreas cascade chiller (but the boreas heatsink was not used for hot side cooling but a radiator, since the mod was liquid to liquid type). And the results seemed quiete good actually.
Please refrain from comments regarding TEC inefficiency especially in cascade config as i am well aware of those.
Today I decided to put my idea down on paper and downloaded google sketchup for that purpose (lovely piece of easy to use software). Since today is my first attempt at using this software some parts of my design are missing and are inaccurate, but for the most part the dimensions are correct (except for fittings size, tubing, and 5 1/4 inch bay is larger then usual and has a transparent window on purpose).

So firts i'll upload some pics, then the explanation:

 

[teatchme]

More pics of the assembly in the case with a meanwell psu (the tubing is wrong) and 1 480 and 1 240 rad.

 

[teatchme]

Now to the fun part

the explanation:
N.B: i did not create everything from scratch and used sketchup database alot so i do not claim everything u see as my own work.

The idea is to have an assembly of 3 heat exchangers and 4 series of 4 TEC cascades in parallel.

it goes like this:
hot side WB
cascade tecs
cold side wb
cascade tecs
hot side wb

sealed with neoprene

the tecs used are cascades of 40mm tecs (and btw all the dimensions are quiete accurate in my sketches) around 5 mm tall (10mm total), i should include a few extra mm for arctic ceramic though. The WB design is NOT perfect but should be along those lines. the cold side heat exchanger also needs more work since it is not optimised for heat exchange on both sides.

the dimensions of the unit are 56.7mm tall, 214.2 cm long (no fittings), 100.2 mm large (corners) 125.3 large at the widest part.
basically very similar to a meanwell psu and will fit in 2 or 3 drive bays depending on the thickness of copper chosen.

The way it works is hot side heat exchangers are in the loop with one d5 pump, one 240 and 1 480 rad. and the cold side HE goes to the cpu, the second d5 and other cooling needs (the reservoirs are missing for the moment).

The mounting bracket will make the unit fit in the drive bay but i did not include it in the illustration.

Now to tecs. something i have not thought through completely yet, but i can say a few things:
the tecs in cascade will run at same V to use the same psu, and the upper tec will be a slightly more powerful one. i plan to run those at around 50% but am not sure yet.

this is just the first draft of the design so is a little rough around the edges, WB design is yet to be improved.

 

[DuckieHo]

Cascading TECs is pretty inefficient. Why not just make a single larger block with more TEC in parallel? Have this thinner block be the length of the bottom of your case.

 

[teatchme]

mmmm, i kinda said i KNOW it is not very efficient. the idea is to get a bigger dT. why? because hot side cooling gets better the higher the dT of hot side is from ambient. this was one of the probs i had with my previous build, no matter how many rads i threw at id the cooling of hot side would improve only marginally.

I know i can get better dT by upping the V but not as much as cascade tecs

 

[Random\/]

Can't really comment with any technical advice but seems like a very interesting setup, also very nice sketches! Hope it works out

 

[Scarlet Infidel]

I would urge you to do some real calculations here.

Stacking peltiers is a reasonable way to get great dT when you come close to the ~70c dT possible with a single module. For this application I do not think you will improve temperatures by doing this and would be better of with a single layer at higher voltages.

When I get home from work I will run some rough numbers. I'd be interested if you can give some approximate numbers too so we can see if we are on the same lines.

 

[teatchme]

I was about to do some number crunching later as well.

p.s: Check out the boreas cascade boreas build at coolit forums, really interresting stuff.
Thanks for the encouraging comment RandomV

edit: a slight improvement pic

 

[Scarlet Infidel]

Ignoring the stacking issue for now. This doesnt look like a bad start. The copper and TECs are going to cost a lot.

The middle section is going to be cold so it would be wise to extend the insulation around a smuch of it as possible (but not he hot sections).

As always with chiller designs, good luck, I genuinely want it to work well.

 

[teatchme]

new pics:
made a psu combo in a mounting bracket

Thanks scarlett. i know the price is going to be just CRAZY, but for the sake of the discussion well leave it out of the picture for the moment.

the dimensions are around 115mm (without controller) high and 241 long

edit: added a controller (its a schyte but its for illustration purposes only, should be a pid controller)

edit 2: uploaded with pump rez combo just for fun, the height becomes an issue as is about 193 mm, should fit in a fulltower nonetheless.

 

[teatchme]

more pics of the assembly

pump, rez, pid controller, heat exchanger and psu all together in one unit.

p.s: tubing is not complete but rather indicative of how it should look like.
also tec wiring is not done, but thats too much work for nothing. just use your imagination.

edid 2 more pics of the unit installed inside the case

 

[zipdogso]

Hi teachme...nice to see you again...

I appreciate what your doing but there are 1 or 2 concerns...

Until one can get the hotside down pretty low cascading is not actually going to make that much difference. Cascading is generally used when Dt is very low your not going to get to that point...you need low voltage (high undervolting.) and a very chilly hotside.

Cascading generally only increases the Dt MAX by about 10ºC per layer under ideal conditions. so basically instead of the average 69º to stack one TEC on top of another it becomes 79º but you see that increased Dt is spread proportionally same as the standard Dt max so if you ran a 15.6v TEC (127 couple.) at 12v you would not get 10º increase in Dt...more like 7º or 6º even...you see the problem ?

Most cascades used in industry are 3 or 4 layer...now your talking about 30-40º increase in Dt max but the Qmax of these units is ridiculously small, nothing usable to us in computers. It is small because it has to be for the fourth layer to cope with everything under it, plus they are generally used on a specific small electrical component.

Also don't forget the Qmax of the whole 2 TEC unit is only equal to the Qmax of the bottom TEC i.e. you dont add the cooling power of both TEC's together the cooling power is only equal to that of the bottom TEC, the one next to the heat source. So the Qmax of your 16 TEC's will only be the equivalent of 8 TEC's. (And hence the inefficiency of cascades.)

 

[teatchme]

Hi zipdogso, i appreciate your concern, indeed i have looked up cascade tecs and it's as u say: the dT is not much bigger and qmax is small. but i have been tossing some numbers around and efficiency seems pretty decent as well as cold side temp under a fully loaded cpu, 2 gpus.
i have a feeling the general limitation to cascade tecs is that way below zero dT exponentially struggles, and every extra C is very hard earned, so if i got it right untill i gow under 0C the dT does not take a serious hit. try running some numbers, cascade low V vs single high V. the efficiency is not that bad really.

 

[zipdogso]

Quote:

Originally Posted by teatchme Hi zipdogso, i appreciate your concern, indeed i have looked up cascade tecs and it's as u say: the dT is not much bigger and qmax is small. but i have been tossing some numbers around and efficiency seems pretty decent as well as cold side temp under a fully loaded cpu, 2 gpus. i have a feeling the general limitation to cascade tecs is that way below zero dT exponentially struggles, and every extra C is very hard earned, so if i got it right untill i gow under 0C the dT does not take a serious hit. try running some numbers, cascade low V vs single high V. the efficiency is not that bad really.

There is another problem you will be better to put the 2 sets of TEC's side by side rather than one on top of another the hotside of the bottom set will be warming the coldside of the top set.

EDITED DUE TO AMBIGUITY :-

Your setup is WB - cascade TEC - WB - cascade TEC - WB

the middle WB is both cooled by the TOP set of TEC's and heated by the bottom set of TEC's.
Irrespective of what qmax the TEC's are because you are using cascades the heat from the bottom sets hotside will far exceed the cooling of the top sets coldside and you may well find the coolant in the middle WB actually gets quite a bit warmer NOT colder as planned.

If you need anymore convincing consider this :-

Lets say for instance the bottom TEC of each cascade has a Qmax of 50w, the max heat output of the hotside will be about 150w so the top TEC needs a qmax of at least 150w which will have a max heat output of about 400w.
So you have 4 of these cascades in the top row and 4 in the bottom row.
The max cooling power is = qmax of the bottom cascade TEC =50w times 4 cascades = 200w max cooling power.
the max heat coming off the hotsides of the bottom cascades = 4 times 400w = 1600w max heat coming of hotsides.

In the battle for domination of the middle WB who will win 200w cooling or 1600w heating...hmmm

 

[flak-spammer]

hmm I'd think it would be more efficient to have say 8 high power TECs running at a low voltage (high thermal efficiency) and say 4 medium power TECs running at a high voltage to gain your dt back. Otherwise you'll end up with a massive heatload because you end up dealing with the heatload of each TEC + the TEC stacked on top.

 

[Ultrasonic2 (muffy)]

Stacking does work ..as long as you've selected the right TEC's to do the job and apply the correct voltage. It will just draw ALOT of power though..

I've always wanted to give it a go myself and probably will with my next block

 

[zipdogso]

Quote:

Originally Posted by teatchme Hi zipdogso, i appreciate your concern, indeed i have looked up cascade tecs and it's as u say: the dT is not much bigger and qmax is small. but i have been tossing some numbers around and efficiency seems pretty decent as well as cold side temp under a fully loaded cpu, 2 gpus. i have a feeling the general limitation to cascade tecs is that way below zero dT exponentially struggles, and every extra C is very hard earned, so if i got it right untill i gow under 0C the dT does not take a serious hit. try running some numbers, cascade low V vs single high V. the efficiency is not that bad really.

I would be interested to see these numbers your tossing around....cascades cannot possibly be efficient. COP is the relationship of cooling power to input power in your cascades you have 2 x input power for 1 times cooling power cascades can never be efficient....ever. I find it amazing your saying the efficiency "seems pretty decent". What TEC's are you planning to use and at what voltage.

I have edited my last post as I felt it a bit ambiguous.

 

[Ultrasonic2 (muffy)]

i guess that depends on what he means by "Efficient"

but your right we really need to see the numbers he's talking about. so we can check the theory

 

[teatchme]

zipdogso, i dont think you got it right.

the tecs on both sides are facing towards each other. the middle block is cooled ny cascades on both sides..

just woke up, so will try to give some numbers soon...

to gain more efficiency tecs will be used at around 20% so the electricity consumption should be ok as well.

ok here is an example: 16 tec cascades

top 226 Umax 15
bottom 172 umax 24.6
pc load at 600W

so the 172 tecs at 20% run at 5V power draw around 200w (max load around 1000)
226 runs 5V at 32 percent. (same psu) around 600w (max load 2000)

now the load is 600 so the total for the 226 tecs to cool is around 800, at 35 hot side coolind temp i get a dT of 25 under load, which gives 10C. so the bottom layer of tecs at around 10C ambient and under 600W load gives a dT of 9C. so we're down to 1C at full load.
and 1500W to be cooled by the radiators.
total heat to be dissipated by rads is around 1500w (2 480 rads should do ok)

i realize there are some losses of efficiency here and there, but this is just really brute estimate. more tuning should be done to identify what V and tecs to use, WB design, etc...

EDIT: i am trying to tune my build for 1000W of heat and use less than 1000 to power the tecs. the goal is to be at around 10C at full load. i've been trying to estimate what result a non cascade setup would yield and it seems i would need around 20 tecs (and a bit longer WBs) to be comfy.

 

[zipdogso]

Quote:

Originally Posted by teatchme zipdogso, i dont think you got it right.

LOL...trust me to be so blinkered....yeah ok.

Not that I, personally, would like that setup, I think that bearing in mind heat rises and cold falls you will need to ensure that the method of heat removal works well for the bottom set.