Overclock.net › Forums › Cooling › Cooling Experiments › Water Chiller: Peltier Cooling below ambient but above dew point, using a custom dew point controller
New Posts  All Forums:Forum Nav:

Water Chiller: Peltier Cooling below ambient but above dew point, using a custom dew point controller

post #1 of 269
Thread Starter 
(If all you are interested in is the controller system it's the second post, it's not completed and I still have quite a bit of programming to go, but the concept is there at least!)

Hey all!

I decided to build a custom cooling system for my processor. It's a work in progress but we're getting there. This is in the nature of an experiment. I got my inspiration from Ryan's buildfrom over at Tom's Hardware, customizing it to make it my own, and learning a ton as I go. thumb.gif

I thought about evaporative cooling, but I wanted something maintenance free once it's built.

I've got most of the parts on order and still have quite a bit that I'm going to have to fabricate up, but we're moving along faster than I originally thought!

Figured I'd share what I'm doing with you guys and you can follow along as I either crash and burn, or successfully build myself a neat little project! biggrin.gif

So I'm building a three TEC system, I'm going to be using air cooled heat sinks to cool the hot side of the peltier. I know, I know, they WON'T work a bunch of people say. Either way, we'll learn together if the nay sayers are correct, or if the 200 watts of cooling these things are rated at will be sufficient.

My largest worry at this point is that the TEC's are not going to be large enough and I may either need a fourth or to step up to the next size TEC. I currently have three 12715's on the way, running at 12v they should be running a Qmax of around 120w, so all together that should give me 360-ish watts, kinda. It's the efficiency loss to --> copper block --> water that has me worried..... Oh well, nothing to do but try! wink.gif

So when drawing up the design (in my head), I had it all planned out. Everything was perfect. So I started buying parts, TEC's, coolers, insulation, fittings, PVC pipe (following in Ryan's foot steps there), water blocks ect.

Then I found that my internal tape measure was off.... the PVC wouldn't fit the case I was planning to build. Well crap. So I started looking online for a cheap solution for a reservoir. Nothing, nada, zip, zilch. People are PROUD when they build one of those things. We're talking around $80++ for one the size I want. So I toddled off to my local plastics shop to see how much they would charge. Ends up the material is $10, for them to cut it is another $8 or so, then $80 for them to put it together!!! But that bottle of weld on 4 over there is only $10.... So, here I am assembling my own reservoir! (I can't thank the guys down at SSD Plastics enough, super helpful, even threw in some practice pieces for me!)

Drilling and tapping the 1/4" NPT holes wasn't too bad. The 3/4" NPT fill port got me though, I think I know how to fix it next time, but I had some minor chipping around the edges, but since this isn't a show piece... who cares? It still sealed! Here are a couple of pics of the build and completed res.

Drilling and tapping, making a mess.


The 1/4" NPT hole


Test fitting the barb


Yup, just about right


Leak testing, Look Mom! I built this!


Final assembly



So, the res is out of the way now. (well crap. You know what I just realized???? I COMPLETELY FORGOT TO PUT A HOLE FOR MY DAMN TEMP PROBE!!!!!!!) *sigh* Let's see if I can drill a hole without cracking now that it's installed... and then get any trash back out....

Continued.... next post!
Edited by lmarklar - 4/27/15 at 9:58pm
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #2 of 269
Thread Starter 
Updated 06/12/2015
Latest post on the controller:
http://www.overclock.net/t/1549736/water-chiller-cooling-below-ambient-but-above-dew-point-using-a-custom-dew-point-controller/120#post_24031012

This post has also been updated.

First Draft:


I started researching dew point controllers. Now maybe I just didn't look hard enough, but man o man those things are pricey!!! Like $300 pricey, for a freaking controller!!!!

So again, I started doing more research, which led me to programmable micro-controllers.... which got me to thinking....

So, I bought a bunch of stuff, none of it overly expensive, I think all told I probably have $40 or 50 into it. $35 for the Arduino and touch screen, $3 for the box, $3 for the interface board, $3 in components for the voltage dividers ect and a bit more in wire, loom, and paint.

Here is the gist of the controller:
  • Reads temp and humidity from the DHT.22 sensor, which I will mount inside the computer case
  • Calculates the dew point based on the above readings
  • Has a programmed set point for the water temperature, by default 2C° above dew point, can adjust on the screen with buttons
  • Reads the temp of the water in the reservoir using a thermistor and voltage divider
  • Adjusts the peltier PWM based off of the reservoir temp vs set point (2C° over dew point or user defined)
  • Reads voltage from each peltier by using dual voltage dividers one on each of the + and - legs, then gets the difference between them (was unable to just measure the + due to the FET's being on the ground leg)
  • Reads the temp of one of the heat sinks on the peltier using a thermistor and voltage divider
  • Controls the peltier heat sink fans via PWM based off of the heat sink temperature (Currently not adjustable except in the program)
  • I think it does something else... but I can't remember...

Here are a couple shots:

Voltage dividers on the interface board:


Wiring Harness:


Completed Controller:

Edited by lmarklar - 6/12/15 at 5:46pm
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #3 of 269
Thread Starter 
Ok, the build is close to done, close enough to update this final build post.

So the cooler works. Just not so great with a 5820k due to the massive amount of heat that these chips put out when OC'd. Sitting at desktop with a 4.4Ghz OC I can hold the reservoir temp to 10c° and my CPU temps at about 15-16c°. With a 15 minute stress test I reach 48c°. This OC has a pretty substantial system agent voltage increase for memory also that adds quite a bit of heat. I actually get better results with 1.38 vcore and no memory OC than I do with the above.

If I drop the CPU to stock I can hold 9.6c° with one TEC off. As a reference the thermal output on this chip at stock is pretty close to a 3700k at 5Ghz.

So the final build ended up being three 200W Qmax TEC's running at 12v, and one 145W Qmax TEC running at 5v. The voltage controller was a bust, it just generated too much heat and caused all of the sensors to go nuts with ground switching. But with straight voltage from the PSU the system works great.

If I had an i5 setup this would be an ideal cooler. I think that for an i7 or haswell / haswell-e that the TDP of the chips is just to great for this system to completely overcome, at least with the TEC's and heat sinks I'm running.

I still have a bit of testing to do, I want to find the OC point where the cooler just can't keep up and the chip reaches 80 or so under a 30 minute stress test. If I can get the chip to 5Ghz (which I'm doubting) then I'll keep the setup. If not.. well then the extra power bill and cooling bill probably aren't worth running this 24/7 for me.

I don't have any regrets from the project, it was expensive for the R&D, I've spent a ton of time on it. But I've met some great people, had some interesting conversations and learned a ton. I'll update this post with my final results as they come in, and with my decision on if I'm keeping it or not. As I said, if I had an i5 CPU this thing would rock!

EDIT:
The chip just doesn't have to OC potential to make running this cooler worth it. The chips wall is 4.4, I can't even reach 4.6 stable at 1.38 volts. The chiller works great, even with all TEC's set to half voltage I can pull my idle temps down to 12c° or so, and the system is able to hold the temps to safe levels even under a 100% stress test.
I ran a stress test when very first starting the computer with everything at half voltage, the res temp raised by 2c° and the CPU never went over 52c° with my 4.4 freq, 4.0 cache, plus 1.25 on sys agent voltage for memory OC.

The chiller is a definite success. What holds me back now is my CPU.
Edited by lmarklar - 7/5/15 at 12:24pm
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #4 of 269
I tried this with 3 12V peltiers on copper blocks once upon a time and it couldn't keep an Athlon XP below ambient under load. My setup wasn't great, but the performance wasn't even worth keeping the TEC blocks. I got the same performance by removing the TECs and putting the rads in the CPU loop.

To cool a modern CPU better than a high end HSF, I predict you'll need a lot of TECs running on proper 24V PSUs. I'm excited to see your results because I would love my suspicions to proven incorrect. The most fun I ever had tuning was using 2 TECs on a Pentium 3 Slot copper block/cold plate. That thing idled at -20C 24/7.
Overkill
(19 items)
 
  
CPUMotherboardGraphicsRAM
i7 5820k @ 4.625Ghz ASUS Rampage V Extreme GTX 1080 Ti 32GB G.Skill RIPJAWS 3200Mhz CL14 
Hard DriveHard DriveHard DriveHard Drive
Intel 750 400GB Samsung 850 Pro 512GB Samsung 850 EVO 500GB WD Black 4TB 7200RPM 
Hard DriveCoolingOSMonitor
Hitachi 2TB 7200RPM Corsair H80i GT Windows 10 Professional Acer Predator X34 3440x1440 100hz IPS 
KeyboardPowerCaseMouse
Razer BlackWidow Chroma Corsair RM1000i SilverStone RV02 Razer DeathAdder 3G 
AudioAudioAudio
TEAC HA-501 HRT Music Streamer II+ DAC Sennheiser HD650 
  hide details  
Reply
Overkill
(19 items)
 
  
CPUMotherboardGraphicsRAM
i7 5820k @ 4.625Ghz ASUS Rampage V Extreme GTX 1080 Ti 32GB G.Skill RIPJAWS 3200Mhz CL14 
Hard DriveHard DriveHard DriveHard Drive
Intel 750 400GB Samsung 850 Pro 512GB Samsung 850 EVO 500GB WD Black 4TB 7200RPM 
Hard DriveCoolingOSMonitor
Hitachi 2TB 7200RPM Corsair H80i GT Windows 10 Professional Acer Predator X34 3440x1440 100hz IPS 
KeyboardPowerCaseMouse
Razer BlackWidow Chroma Corsair RM1000i SilverStone RV02 Razer DeathAdder 3G 
AudioAudioAudio
TEAC HA-501 HRT Music Streamer II+ DAC Sennheiser HD650 
  hide details  
Reply
post #5 of 269
Thread Starter 
Quote:
Originally Posted by littledonny View Post

I tried this with 3 12V peltiers on copper blocks once upon a time and it couldn't keep an Athlon XP below ambient under load. My setup wasn't great, but the performance wasn't even worth keeping the TEC blocks. I got the same performance by removing the TECs and putting the rads in the CPU loop.

To cool a modern CPU better than a high end HSF, I predict you'll need a lot of TECs running on proper 24V PSUs. I'm excited to see your results because I would love my suspicions to proven incorrect. The most fun I ever had tuning was using 2 TECs on a Pentium 3 Slot copper block/cold plate. That thing idled at -20C 24/7.

Ryan over at Tom's Hardware has a setup using three TEC's that idles at around 12c or so, think his load temps are around 60, but that's with a 3770k at 5Ghz and it's not delidded. He is using TEC's that are the next size up from mine however. But he swears the trick is to use insulated reservoirs, so we'll give it a shot! Like I said, should at least be a fun experiment if nothing else, and I'm building stuff that can be useful to other people in different applications (custom controllers for just about anything voltage or PWM controlled).

Ryan's cooling experiment page 8, where the temps are linked under intel stress test with the above processor
Edited by lmarklar - 4/6/15 at 12:11am
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #6 of 269
Thread Starter 
So I was fiddling around with the controller tonight, experimenting with using it as PWM controller for the fans that will be in the cooling box.

I was running into issues. The specifications for the fans says that the PWM needs to be 21-28Khz, 25Khz being optimal. The controller I have is capable of that, on one channel out of six available. (This was actually incorrect and I've since learned a bit more about the PWM freq, here is a link to the guide I used http://playground.arduino.cc/Main/TimerPWMCheatsheet) If I adjust the other channels to that frequency then I start screwing the internal timing on the chip and any time based functions or interrupts run on the new clock speed.....

So me being the 'eh, what the hell!' kinda guy, I just threw the base 500hz PWM onto the fan using a simple -analogWrite(3, 120); .... yeah, five HUNDRED hertz, not 25,000..... Fan works just fine.... thinking.gif

So apparently it's just duty cycle that's important?

So, I am assuming that the reason for the 21-28Khz PWM range is due to the audible factor? Motors that are modulated under 20Khz can have an audible sound to the human ear? Or is there some other reason that I am missing? I put the fan up to my ear and the oscilloscope fan sitting three feet away was louder.... No noticeable motor noise being induced at the lower frequency. Can anyone shed any light on the subject?


As a side note, I did notice that there was a 1 volt ripple (p-p) 200nS long, as soon as the fan was hooked up PWM, the switching on that thing was nuts. So I used my 'Call a friend' and we were bouncing ideas (I had already tried a decoupling cap), he suggested a inductor in front of the power on the fan, so I was scouring around trying to find one, nope, not gonna happen, then he says maybe a ferrite bead? Hey!!! I have a couple of burned out PSU's floating around!!! Threw a ferrite bead on the 12v side of the fan, no dice, the noise was still there, so I put it on the PWM line and it reduced the noise... 5 wraps later the noise is under 100mV, nice!!!! I'll throw a cap in there after a bit more experimenting to finish up the smoothing, but looking good so far!

EDIT: The ripple that I couldn't smooth out with a decoupling cap? Only with an old corsair 4 pin. I put a newer coolermaster 4 pin on there, decoupling cap, relatively clean signal with minimal effort. I may still end up throwing a tiny inductor on there if I see any adverse effects, but I think 100mV might be acceptable to this circuit.

I'm now working on a PWM control method for the peltiers, working out the hardware first, then I'll get into programming the controls based on temps. The hardware is kicking my butt though, so might be a bit buttkick.gif
Edited by lmarklar - 4/17/15 at 10:19am
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #7 of 269
Great post smile.gif

This is OCN, we love this!
post #8 of 269
That also depends on which type of pwm, if they are 2 wire fans, or if they are 4 wire fans, I've been thinking about what sort of difficulties could happen with 4 wire.
post #9 of 269
Thread Starter 
Quote:
Originally Posted by Serifus View Post

That also depends on which type of pwm, if they are 2 wire fans, or if they are 4 wire fans, I've been thinking about what sort of difficulties could happen with 4 wire.

Four wire fans are what I'm running my tests with. Straight PWM from an Arduino would not work on a standard three wire computer fan (or two if you ditch the RPM line). You would have to have a separate 12v power source and then run the PWM pulse through a MOSFET or something, hell you might even be able to get away with using a MOSFET driver instead, faster response time, harder switching, you're not putting much of a load with a single fan.

Not sure what the noise effect would be using 12v PWM (in essence...), probably not the best though at a lower frequency. As an alternative you could build a simple filtering circuit, basically just a resistor and a cap, to filter the 12v power until you got a DC voltage instead of pulses, you would have to calculate the ripple, but it's a simple circuit, matter of fact... I wonder if that's what they do in four wire fans.... I just don't have any that I want to donate to science ATM or I'd tear one apart and see if they just put a filter in there to smooth the voltage from PWM control.....
Edited by lmarklar - 4/11/15 at 12:53am
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
post #10 of 269
Thread Starter 
So, on the subject of PWM control...

I've been tearing my hair out over building a control system for the Peltiers. My initial plan was really really simple. On until at temp, then Off until XX degrees over temp. But I got to doing some reading, and apparently TEC's are pretty susceptible to thermal cycling and it tears them up pretty quick. With a overclocked CPU it won't take long to heat up half a gallon so the TEC's would constantly be cycling on and off... maybe even multiple times a minute depending on how hot the CPU is. Crap. Not gonna work.

No biggie, just do exactly what I mentioned in the post above, run a PWM circuit to a 30amp MOSFET (using a driver to ensure hard switching as an Arduino only provides 40mA. Most large FETs can draw upwards of 175mA at max temp and soft switching{not providing enough amperage} causes heat... vicious cycle) and Voila! Instant TEC control!
(for proof of the concept about soft switching above see post 48 in this thread. Oops)

The problem with this is efficiency. I'm a cheap bastard and got the $5 TEC's. I NEED them to run as efficiently as possible or I'm going to have to upgrade..... $$$$$$$. (Which I'm probably going to have to do anyway, but hey design is fun!) The problem with PWM and TEC's is that a TEC's cooling is a direct effect of I*R, but the heat produced is I²*R.... so the more you crank em up, the more heat they produce and less efficient they become. So a PWM controller would (as it sits) pulse them full on, and full off, the net effect at say 50% duty cycle is that your power *seems* to be roughly 6 volts and 6 amps.... but that's not how the peltier see's it, it would see full voltage and amperage being turned on and off (granted thousands of times a second... but still full ON/OFF) and be running at something like 40% efficiency loss by the ON/OFF pulses instead of a real 6v.

Simple! Do the above filter circuit and smooth the voltage and give it a smoother source! Yeah...... NO. Have you ever priced out a 250-300w resistor??? Holy Freaking Wallet on Fire Batman!!!!

I'm no engineer, so I've been scouring the net looking for circuit designs and reading up on my electronics again. I can fix stuff, but making it is a whole nother ball game.... Especially something that will handle roughly 20-25amps for EACH line. (no point in building all this twice.... I'm going to build it to be able to handle the next step up in Peltiers just in case I upgrade)

I have some parts at work that I'm going to play with this weekend, I've been tearing apart bad PSU's and scavenging the inductors and caps from them, I have plenty of normal resistors, fets, ect. I'm going to try two methods I think, putting the FET on the ground side with an inductor on the positive (with schottky diode across the inductor) and a couple of caps. See how smooth the power applied is. Then I'll try moving the FET to the + rail instead using the same circuit as above. I may have to add a big ass cap to protect the PSU though.... hammering the 12v line with a FET and a huge load might not be such a good idea.....

I won't be able to load them down as the TEC's haven't shown up yet and the most I can load the circuit is around 2 amps (see above about high wattage resistors...) I have no idea if the design is actually going to work or if I'm going to blow stuff up, but figure there's only one way to see!

Any electronics guru's around want to chime in with advice or ideas?
Edited by lmarklar - 4/27/15 at 10:06pm
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
X99 Reboot
(9 items)
 
CPUMotherboardGraphicsRAM
Intel 5820k Asus X99 Pro 2 x GTX 780 Mushkin PC22400 
Hard DriveHard DriveHard DriveCooling
Mushkin PCIe Scorpion Intel SSD Seagate 7200rpm Alpha Cool 360 and 240 XT45 
OS
Win7 64 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Cooling Experiments
Overclock.net › Forums › Cooling › Cooling Experiments › Water Chiller: Peltier Cooling below ambient but above dew point, using a custom dew point controller