TEC waterchiller

[Jorlen]

System is up and test running atm. Using the configuration described above with 9 TECs. No fans are running anywhere near the heatsinks. E4600 @ 3ghz, 1.4V core voltage setting. Ambient around 23C
Idle temps from Everest:

[systemviper]

NASA -we have lift-off

Sweet, fun stuff

[Jorlen]

First screenshot is during Prime95 that has been running for two hours and second screenshot is just a few seconds after i stopped Prime95.

Ambient temperature stayed the same, about 23 degrees, during the whole time. The CPU temp shoots up about 10-13 degrees C in a second when Prime95 starts. Then water temperature rises about 5-6 degrees over 10-15 minutes and then reaches equilibrium. When Prime95 stops cpu temps drop the same amount just as fast. And when I am writing this the water temperature is falling and I assume the cpu will reach same idle temps as before within a few minutes. As a comparison I had Arctic cooler Freezer 7 pro mounted prior to this setup.
http://www.arctic-cooling.com/cpu2.php?idx=79&disc=
And under same load it maxed out at about 45 degrees C, using a 7W 92mm fan at full blast.

I am very happy with the results. I am only using 25% of the what the 250W PSU can dish out. I not using any fans to cool the heatsinks. And at current humidity I am not condensing any water.
More pictures to come.

[FrogBite]

Nice results

So how much did you invest in total, 190usd?

What is the water temperature at load and idle?

[Jorlen]

Water temps I havent measured yet. But I would guess the Idle temps is about where its at. So 7-9C after long time Idle. And 16C at max load.

The cost has run away for sure. If you wanna count everything including the things I had lying around (ie. the PSU) it cost a lot more then $190 US.
Pump was $100, so was PSU 6 years ago.
Hoses will be almost $100 when include the NB and GPU. Silicone hose is freaking expensive, and I didnt get a good price. Only fortune is that I could build the reservoir and cpu waterblock for free at a friends workshop.
So more like $500 in total.
Normal watercooling is expensive. And when you change the radiator for 10 x (TECs + Waterblocks + Heatsinks) it doesnt get any cheaper.

[systemviper]

10 x (TECs + Waterblocks + Heatsinks)

now that is a monster TEC build,

I won't even go near that price thing, I know how the
cost just keeps going up, there is
always one thing that was forgotten or
had to be added, always,,,,

[n00b 0f l337]

I think the only reason your getting this is your only adding about 100W to the loop.

[Jorlen]

I wonder if the temperature difference of 5 degrees C between my two cores is because of uneven mounting of my waterblock.
I dont have any springs to even the load of the 4 bolts that hold the waterblock. And the hoses are thick and elastic, pushing the waterblock sideways as they are bent.

[Jorlen]

What do you mean NOL, only reason. Reason I get what I get is that I have what I have and load what I load.
Other load other temps. More fans other temps. More wattage to the Peltiers other temps. Gotta sleep now...

[Sam__]

nice job. have you tried upping and lowering the voltage to the TECs?

[FrogBite]

Ya, I wanna see something like this under a quad

[Jorlen]

Assuming my CPU puts out about 100W, I have a total load on 9 heatsinks of 163W + heatdump from pump, 170W.
With one more TEC and going 4V -> 5V, power will 109W to the TECs. System total will be 46W higher, 216W. Thats a 38% increase in cooling power and a 27% increase in heat output. With 11% more heatsinks.

I have just taken a regular inside/outside digital thermometer and plunged into the reservoir. And the temperature does not match what Everest shows. Water temperature lowest is 15.7C, 23C ambient. And it seem to stagnate at 21.5 under load. This is a cheap thermometer. But I doubt its scaling is off by much. So temp rise 5-6 degrees C is about right.
But the CPU temps are probably way off.

[scifikg]

Assuming my CPU puts out about 100W, I have a total load on 9 heatsinks of 163W + heatdump from pump, 170W.
With one more TEC and going 4V -> 5V, power will 109W to the TECs. System total will be 46W higher, 216W. Thats a 38% increase in cooling power and a 27% increase in heat output. With 11% more heatsinks.

You can always turn up your vcore to simulate hotter cpus and do your tests again. It would give you an idea of how it performs with various heat loads.

At 3GHz,

1.7v =150W
1.6v=133W
1.5v=117W
1.4v=102W
1.3=88W

[Scarlet Infidel]

Assuming my CPU puts out about 100W, I have a total load on 9 heatsinks of 163W + heatdump from pump, 170W.
With one more TEC and going 4V -> 5V, power will 109W to the TECs. System total will be 46W higher, 216W. Thats a 38% increase in cooling power and a 27% increase in heat output. With 11% more heatsinks.

I am currently building a little calculator program for this kind of application. So far its very rough, but I just ran some numbers through it and was pleased to see it seems to match what you are saying.

with 100w heat load, 9 heatsinks with peltiers at 4v I get power consumption of just of 63W.

with the same load and 10 heatisnks/peltiers at 5v I get power consumption of 109w (you said 63 + 46 = 109).

In the first case, to get these figures and water temp to match your 21.5c at ambient 23c I had to make assumptions about the c/w of the heatsinks. To get this to work I had to put in a value of 0.9C/W which is frankly, terrible. But then they were cheap and are passive.

However, this also showed that going to 10 units at 5v with heatsinks as they are, would give a water temperature more like 17.5c, does this come anywhere near your expectations?

EDIT: a quick look on frostytech suggests that the C/W of the heatsink with basically no airflow with a 27mm 50w heatsource is 0.94 C/W. This will improve with a lerger heatsource, modern thermal paste and get worse due to the lower heat load you are applying and less air speed, so 0.9 probably Isn't miles of. Also I add the thermal resistance of the thermal paste separately and they obviously include it. This probably brings the C/W of your hot side cooling dangerously close to 1 C/W.

Have you considered some low speed fans to give the system a boost? I think it will make the world of difference.

[Jorlen]

I have considered fans. But I need to point out the keeping noise down is more important.
There is a rather large copper heatplate between the heatlanes on the heatsink and the hotside of the TEC. The plate is warmer then the heatsink by a noticable margin. Which to me suggests that thermalresistance between heatsink and TECs is far from optimal.
I knew this when I decided to include the hotplate. It comes with the heatsink. It made things a lot easier when coming up with a way to assemble everything. And the peltier is 40mm while the heatsink is about 60mm, which mean I would use only 2/3 of the small heatpipes in the heatlane without at heatplate in between.

Thermal resistance of heatsinks using passive convection goes down as they get warmer. I dont know if its obvious, but at constant airflow, energy flow between air and heatsink is in linear relation to temperature difference of the air and heatsink. But since the airflow increases with the higher temperature of the heatsink the C/W goes down. With this in mind I am still hoping to cool well without fans.
There are so many materials for the heat to transfer through in my setup:
CPU, thermalcompound, heatspreader, TC, Waterblock, water, waterblock, TC, peltier, TC, heatplate, TC, heatsink(which is complex aswell, since it heatpipe-based), air.
Something about this is not optimal.

[Scarlet Infidel]

I think you are right to use the plate for mounting. Its very very important to cover the whole peltier and finding heat sinks with a large enough base area isn't easy.

As I'm sure you know; things to check are that the plate top and heatsink base are both flat, that you have a thin but even layer of good quality thermal paste and that you have the two clamped together as evenly and tightly as you can manage.

It might also be possible to permanently bond the plate to the heat sink for better thermal conductivity.

Thermal resistance of heatsinks using passive convection goes down as they get warmer.

I know this very well. On frostytechs review then got 0.94 C/W with a 50w load and 0.64 C/W with a 100w load. But you have less than 20w load, so you are hardly pushing your heatsinks. Hmm, I need to think.

[coo-coo-clocker]

Great work on this Jorlen -
thoughtful design and a cool idea. Not exactly a price performer, but hey, when is anything extreme also affordable?!

[Jorlen]

Ok, the system is running with 10 TECs now at 6V per TEC. I have connected the GPU and northbridge to the loop. It was a pain to make a mounting system for the koolance waterblocks, to mount them on the GPU and NB. But it turned out ok in the end. Since I didnt know if the wiring would be final I didnt bother to make it pretty. Have that in mind when I post pictures later.
If I get my hands on a high precision thermometer I can find out if all the TECs get the same amount of hot water. They feel about the same temperature, but that tells me close to nothing.
Since I hooked up the waterblocks in parallell and GPU CPU NB in parallell I can choke the flow to each part independantly. I doesnt seem necessary with my current temps and ambitions. But I have that option atleast.

Regarding the temperatures I am pushing the limits of the passive cooling capability of my 10 Scythe Heatlane. Its barely worth giving more voltage to the TECs the C/W is so bad that the increase in performance is barely measurable.
So with double the energy expended in running the TECs all i get is a water temperature that is 1C lower at max load.

My system is down to 3 moving parts now. A reasonably quiet 500gb Samsung HD, the 10W DDC pump, Corsair 450W, 120mm PSU-fan.
The pump is hanging from the hosing outside out in the open. Its fairly quiet. The HD is suspended in rubber. And the PSU is about as noisy as the pump.

I have an old PSU that will come under the knife next, I plan to lengthen the legs of the power transistors in the PSU. And move them outside the PSU box, possibly outside the computer case aswell. And with bigger heatsinks on them, their temperature should stay within operational range, using passive convection only.
I dont know at which frequency the transistors work. And I dont know the inductance the extra length of the legs will add. Or how that might have an effect on the operation of the PSU. If the PSU mod works and it doesnt get too warm under these new conditions, I will perform the same modifications on the Corsair.

The final task will be to build am sound proofed box to house the pump and HD.

[Jorlen]

E4600 @ 3350mhz and 1.55V setting (shows as 1.5v in CPUID)
10 TECs @ 6v
Prime95 for 1h 30min
Ambient 23.5C
Watertemp 22.3C

[Scarlet Infidel]

Well done, those are good water temperatures for a passive cooler (obviously).

Whats the power consumption of the CPU at those settings? Around 150w?

Did you measure the current draw of the TECs running like this?

Do you know how this compares to running the TECs at 5v?

[Jorlen]

I have not connected the Ampere-meter, although I have an old instrument that will sit on top showing current draw of the TECs.
I have not run the tecs at 5V under this load. Will get back to you on that.

I dont know the exact load, the GPU and NB add a few watts. I suppose its between 120W and 145W all together.
Comparing external thermometer in the waterloop and CPU temp in Everest. I find it strange that the water temperature ranges about 7 deg C from idle to max out temp, and CPU temp ranges 33 deg C during the same test.
One explanation is that the thermal diod on my CPU is somewhat broken. I already know its off measuring the individual cores. Since the measure below water temps. What else is wrong I dont know.

[Scarlet Infidel]

I wouldn't use Everest at all for reading CPU temperatures. Try something like Coretemp or TAT.

Either way, I think the water temperatures are the most important measure of a chillers performance, assuming you have the CPU well enough connected so that most of its heat actually reaches the water in the first place.

[Exal]

everest uses the same readings as core temp

[Scarlet Infidel]

Oh, ok. I'm a bit behind with temperature reading software. Have you tried Real-Temp?

[Fusion2002]

Very cool idea. I really like your concept and setup. Great job!