eXtreme Liquid Cooling - Build Log + Overclocking

thinking about it using the 2nd method around the middle of the sandwich I could insulate with some 3mm silicon rubber and arm flex tape, if I am not going to enclose the unit then I could consider sticking some copper finned heatsinks on the outside of the HX plates and having some airflow over the unit... the hot HX plates are never going to be sub-ambient so it could only help to remove the heat load..

I was planning on mounting the HX unit inside the case, but I think I may well have to look at outside mounting now as if its radiating I really don't want all that hot air inside. The idea of placing a number of my radiators, the HX and power units in a 'radbox' type setup is looking more appealing..
 
Radbox is almost certainly the way to go, otherwise the computer will have to be enormous.

After spending some time trying to find 20mm thick copper plate online and failing, I've put an alternative arrangement together. No equations done on this and the channelling is crude. The thick black bits are acetal which is a reasonable insulator. Cold plates are 3mm copper recessed into the acetal, structure bolted together through the acetal instead of the copper (might not be viable, I like it as far as insulating cold plate from hot but the acetal may shear).

Full size
2eewjlc.jpg


Side plates linking the top and bottom hot plates together seems less important if the structure is mainly acetal so I haven't drawn any. I suppose this is rev2.
 
JonJ678 said:
but the acetal may shear
Click to expand...

If you mean for tapped holes in acetal I guess that might be a problem. TEC clamping pressure is ideally around 250 psi. For my chiller where the surface area of the plates (copper) is ca 12 sq in, with 6 M4 bolts this equates to a torque for each bolt of 15 in lbs. I have no idea of the shear strength of an M4 thread in acetal, but I'm sure you can work it out;) Of course I guess one could just use washers and nuts..

:)
 
Last edited:
Pneumonic said:
thinking about it using the 2nd method around the middle of the sandwich I could insulate with some 3mm silicon rubber and arm flex tape, if I am not going to enclose the unit then I could consider sticking some copper finned heatsinks on the outside of the HX plates and having some airflow over the unit... the hot HX plates are never going to be sub-ambient so it could only help to remove the heat load..

I was planning on mounting the HX unit inside the case, but I think I may well have to look at outside mounting now as if its radiating I really don't want all that hot air inside. The idea of placing a number of my radiators, the HX and power units in a 'radbox' type setup is looking more appealing..
Click to expand...

Yes definately I like the sound of things here...sure if you could fin the back of the hot HX and get some air on it that would be great. I don't think the HX alone will get rid of all the TEC heat once your PC starts working in earnest after all 2/3rds of the TEC heat is going nowhere near your coolant but instead straight into the aluminium block. ( 40mm TEC on 15mmm tube.)
 
10mm deep holes for m4 into acrylic can take the the required pressure

the above design is pretty much what im intending on making er .. though with a proper copper milled design
 
Milled copper would really help the centre piece, don't think it would matter as much for the hot plates. It's difficult to source in thicker sizes than 12mm, which makes the design interesting. Drilling & tapping G1/4" threads in the side won't work so well for one thing. Milling acetal is trivial relative to copper for another.

Not very worried about threads shearing. The acetal block is under shear from the bolts around the edge vs the pressure from the pelts in the middle, I haven't checked how deflection/shear fracture works out at the expected temperatures. Much less of an issue if it was 20mm thick copper, but I haven't managed to find copper in greater than 12mm yet. Occurs that a 3mm steel plate on both ends would reduce the odds of deflection, and that bolts nearer the middle would be useful.

Mainly posting a suggestion for avoiding condensation on tubing. 11/8mm tubing, with 6mm closed cell foam around it. Then wrap lacquered nichrome wire clockwise along the tube from one end, turn around and wrap the same length clockwise back down. The result is two wires going into one end, and two opposing spirals whose magnetic fields will pretty much cancel out. Put a layer of braid over it to help hold it together, then feed the whole lot down some 1" ID clear pvc. End result is 32mm in diameter and won't turn terribly sharply. The idea is pass a current down the wire and it heats up, some heat goes inwards through the insulation but most goes into the pvc tubing as resistance is much lower (about 10x lower) in that direction. The objective is to hold the 1" pvc at ambient temperatures even when the core of the tubing is at -40 or so (relative to ambient).

The idea is sound, it's extensively employed in industry though in less diy fashion. The other approach is jacketing the tubing with ambient temperature water, I'm pretty sure this is much more difficult to achieve than the above. The idea of course is that a 3 plate HX can be organised such that the entire outer shell is hotter than ambient, with only the two cold tubes exposed for condensation. If the cold tubes right up to and a short distance into the HX enclosure are warmed, that only leaves the socket still to go. Heated matt below it is simple enough, no ideas for above the socket yet.

On a vaguely related note, regarding stacked peltiers, based on published specs. A 12711-5M31-15CQ running at 6V will hold a delta of 20 degrees while moving 43W. It'll charge 33W for the privilege, so this pelt will output 76W total on its hot side. A 19911-5L31-15CQ running at 12V will hold a delta of 20 degrees while moving 85W, at a cost of 78W. With a 10W margin for error, this suggests that the bigger one at 12V can cope with the heat dump & transferred heat from the smaller one at 6V. This implies that these two stacked directly will move 40W through a delta of 40 degrees at a cost of (33+78)=110W or a cop of 0.35, which is pretty good for a 40 degree delta really. This is remarkably similar to the larger one running by itself at 15V. So while it's fair to say stacking them makes life considerably more difficult, I think it's misleading to write it off as impossible to use in computing. At least, if sticking to my conviction that two psu's for one computer is too untidy, then one at 12V and one at 6V avoids needing to use 15v.
 
Last edited:
JonJ678 said:
On a vaguely related note, regarding stacked peltiers, based on published specs. A 12711-5M31-15CQ running at 6V will hold a delta of 20 degrees while moving 43W. It'll charge 33W for the privilege, so this pelt will output 76W total on its hot side. A 19911-5L31-15CQ running at 12V will hold a delta of 20 degrees while moving 85W, at a cost of 78W. With a 10W margin for error, this suggests that the bigger one at 12V can cope with the heat dump & transferred heat from the smaller one at 6V. This implies that these two stacked directly will move 40W through a delta of 40 degrees at a cost of (33+78)=110W or a cop of 0.35, which is pretty good for a 40 degree delta really. This is remarkably similar to the larger one running by itself at 15V. So while it's fair to say stacking them makes life considerably more difficult, I think it's misleading to write it off as impossible to use in computing. At least, if sticking to my conviction that two psu's for one computer is too untidy, then one at 12V and one at 6V avoids needing to use 15v.
Click to expand...

So you think 40w Qmax is a useful size for computers ?????
What you cooling....a capacitor ?

And as I understood things you don't add Dt's together the specs of the first TEC only stand because this is the TEC that is actually pumping.
 
zipdogso said:
So you think 40w Qmax is a useful size for computers ?????
What you cooling....a capacitor ?

And as I understood things you don't add Dt's together the specs of the first TEC only stand because this is the TEC that is actually pumping.
Click to expand...

I think JonJ is referring to his liquid chiller rather than direct die :)
 
Not 40W qmax. 40W moved through a delta of 40 degrees. Both tecs are considerably undervolted, so it's blatantly not a qmax value. Both tecs are moving heat as they both have a voltage applied.

The problem with stacking tecs is simple. The one in direct contact with the chilled water works as normal, moving some wattage dependent on voltage and the temperature of either side. However the wattage dumped into the hot side cooling is the heat transferred plus the electricity wasted in the process. The second one, on the side of the cooling loop, has to move the original wattage, plus the waste from the first, and dump it into a cooling loop which has to deal with (cpu wattage) + (waste from first tec) + (waste from second). So for stacking to work, you need a small, efficient tec dumping heat into a big, preferably efficient tec. So a 127 @ 6V dumping into a 199 @ 12V balances out fairly well. Efficiency sucks though, and a single large tec at high voltage achieves about the dame delta with the same efficiency.

Adding the dts will be absolutely fine as long as you're working with the right wattage for the bigger tec. If you take the dt from both above on the assumption that they're both moving 40W then no, it wont work. I think it would be wise to put a copper sheet between the two tecs in the stack though.
 
JonJ678 said:
Not 40W qmax. 40W moved through a delta of 40 degrees. Both tecs are considerably undervolted, so it's blatantly not a qmax value. Both tecs are moving heat as they both have a voltage applied.

The problem with stacking tecs is simple. The one in direct contact with the chilled water works as normal, moving some wattage dependent on voltage and the temperature of either side. However the wattage dumped into the hot side cooling is the heat transferred plus the electricity wasted in the process. The second one, on the side of the cooling loop, has to move the original wattage, plus the waste from the first, and dump it into a cooling loop which has to deal with (cpu wattage) + (waste from first tec) + (waste from second). So for stacking to work, you need a small, efficient tec dumping heat into a big, preferably efficient tec. So a 127 @ 6V dumping into a 199 @ 12V balances out fairly well. Efficiency sucks though, and a single large tec at high voltage achieves about the dame delta with the same efficiency.

Adding the dts will be absolutely fine as long as you're working with the right wattage for the bigger tec. If you take the dt from both above on the assumption that they're both moving 40W then no, it wont work. I think it would be wise to put a copper sheet between the two tecs in the stack though.
Click to expand...


As you say 2 Stacked TEC's never seems to work out ..especially with out using a billion watts of electricity when comparing it to a high powered single stage setup.

However the calculations i have performed suggest you can achieve the holly grain of a getter loaded Deltas while using a reasonable amount of electricity IF you use 3 tec's in the stack. the problem with this is it would cost a bomb to implement. It would require a huge number of TECs and a massive chillers or two .. so it's not very practical.
 
I'm very interested. I'd have expected three stacked to have the same problem with cumulative heat load as two. Which three do you have in mind, and operating at what sort of voltage?

I think it'll involve increasing the dimensions of the tec as well as the voltage as you move up the stack, which makes the delta vs cop more attractive but the hx design more complicated. Definitely interested in which ones you have in mind :)
 
With 3 you can run the TEC's at a low enough input to have a good COP for each and the 3 stages will enable you to keep the loaded delta up.

i worked it all out based on using 62mm 360s. It was gonna take 27 of them or something like that
 
interesting. The big ones don't work very well at low voltage, cop and wattage moved both get worse relative to the smaller ones at the same voltage. Cost is probably prohibitive, but it may be possible to optimise beyond using 27 identical ones.

If the hot-cold-hot sandwich is sufficiently carefully designed it may be possible to populate it with one set of pelts and use it as such. Then at some point in the future add a second layer of them for the cost of the tecs and a couple of copper plates, allows time to save for the tecs, decide if another 20 degrees is worth it, and buy some more radiators.
 
JonJ678 said:
interesting. The big ones don't work very well at low voltage, cop and wattage moved both get worse relative to the smaller ones at the same voltage. Cost is probably prohibitive, but it may be possible to optimise beyond using 27 identical ones.

If the hot-cold-hot sandwich is sufficiently carefully designed it may be possible to populate it with one set of pelts and use it as such. Then at some point in the future add a second layer of them for the cost of the tecs and a couple of copper plates, allows time to save for the tecs, decide if another 20 degrees is worth it, and buy some more radiators.
Click to expand...

oh yeah it's something id love to give a go though ... that would probably be in 5 years time.. and it's only point would be to prove that it's possible
 
JonJ678 said:
interesting. The big ones don't work very well at low voltage, cop and wattage moved both get worse relative to the smaller ones at the same voltage. Cost is probably prohibitive, but it may be possible to optimise beyond using 27 identical ones.

If the hot-cold-hot sandwich is sufficiently carefully designed it may be possible to populate it with one set of pelts and use it as such. Then at some point in the future add a second layer of them for the cost of the tecs and a couple of copper plates, allows time to save for the tecs, decide if another 20 degrees is worth it, and buy some more radiators.
Click to expand...


The idea to use identical TECs is to keep costs down. if there not identical then you have to use a heat spreader (like you said) which increases the cost decrease the total performance to be had
 
Do they have to be identical, or just the same dimensions? The 12711 and 19911 I was discussing above are both 40mm square, but one is a 15.2V 141.3W Qmax pelt, and the other a 24V 225 Qmax one.

The heatspreader might actually improve performance, I'll need to think about that. Conductivity of copper is ludicrously high, a mm or so of it just doesn't matter. Would need to check the coefficients for lapped ceramic to lapped copper & ceramic to ceramic but I think it would help. The temperature across the surface of the pelt won't be uniform as it's a finite number of junctions, putting the two directly against each other will lead to some junctions lining up but most not, meaning conduction laterally through ceramic is needed to transfer heat, leading to poor performance / dead peltiers. Not 100% on this, but I think it'll be the case. Otherwise it wouldn't matter so much that the peltier and cold plate are both lapped in normal operation.
 
definitely exactly the same tec with the same number of couples otherwise they wouldn't line up

it's not just the copper your adding in your also adding in extra layers of TIM whihc is a rubbish conductor

id like to try with the spreader and without for interest sake
 
You must log in or register to reply here.
Back
Top Bottom