Sub Ambiant Water Cooling System

[DieselTwitch]

I debated to posting between Extream cooling and just water cooling. However, because this is a below ambient cooling idea I figured it would be better here.

So first my system the important parts as it is currently: Air-cooled (ND15) i9-7920X OC'd to 4Ghz, and 2x 1080ti's in SLi. This machine is not a gaming rig and is only used to process data. It runs at 100% 6 days out of the week.

Right now its stable at 75-80c with 1.1V and 4GHz. I haven't yet gotten into the memory. I'm a little new to OC so I'm doing things one at a time

One thing I've been playing with is the idea of using the outside air (in the winter its well below -20c most of the winter) I live in 3rd - 4th coldest place in the US. I know i can't dirreclty dump heat out side as it would over cool and drop the system below the dewpoint. My plan is to build a sealed case (mostly) and drop the RH inside the case to as low as posible . If i can keep the RH level inside the case to 10% or less and drop the internal case tep to 10c I can run the loop at -18c. Im thinking of only going to with in 2c or 3c above 0c. Todo this I need to build a two system that controls the rejection of heat to the outside so that the primary system loop doesnt drop below the dew point. at the same time I need to build a small dehumdifer and an air radiator system to cool the case. I was thinking of using one of the large dual custom plexglass case with two chambers. One for the actuall MB,VC... and one that holds nothing but water cooling equiment.

I know I will need to build an HX between the case loops and the exterior loop as I'll need to keep the exterior loop from freezing. in the summer im planning on building in some type of heat pump system to keep it cooling even at outside temps up to about 30c

For control of all this, I'm thinking of using a Pi to monitor and calculate the dew point, control all fans and pumps for the exterior system. The MB will still handle all the pumps, and even some internal case fans to keep the other parts cool. I just wont be rejecting heat to the room.

I'm working on a diagram.

One thing I'm questioning is dropping the RH level too low. will there be a problem with a static charge

What are your thoughts? Stupid?, Worth it?, Fun? One of the things i hope to do is add more Systems to this one loop

 

[Tír na nÓg]

Your last line of questions is the most interesting:
Stupid?, Worth it?, Fun?

No, no, yes!

You can get more out of this sytem by de-lidding it: 80c with 1.1v seems horrendously high. You need at least 2*120 (cpu) + 6*120 (GPU's) worth of rad to cool it efficiently. 6*140 would do

Add rads (external?), and you should be golden without all the work above. In any case, lowering the temps by 30c will only alow for a mere 100MHz extra.

I really think your issue is the lack of rads.
A 280 AIO should be able to take it to 4.8GHz+ after delid, for the CPU only.

See this post:
http://www.overclockers.com/forums/showthread.php/785276-photos-of-delidding-the-i9-7920X

 

[Blaylock]

This sounds like a really fun project. I would recommend finding out what kind of gains you would see by going sub ambient first. You're looking at a custom loop either way, just buy additional tubing so you can run the radiator out a window and see what your OC will be. There are a few problems with this config though.

1. As you mention, the dew point. You will need to address condensation. We do this by applying Liquid Electrical Tape, Soft Art Eraser, or some other medium to prevent the cold board from contacting the warm(ambient) air.

2. As the temps outside fluctuate so will your OC stability. This will likely be your biggest hurdle. When we hang our radiators out in the cold its for a quick OC'd benchmark, not a 24/7 data cruncher. Just a thought but if you know anyone that's good with programming (I'm not) then maybe write a program to run a variable OC off of the CPU's TJ Max sensor.

3. I haven't seen or read anything about altering RH levels in this Forum. This maybe uncharted territory here. You may need to do a bit of researching other industries to get an answer to this, or maybe there's someone here with said knowledge.

Please do create a build log if this does become a reality. I'd love to see as I'm sure others would as well.

 

[DieselTwitch]

Your last line of questions is the most interesting:
You need at least 2*120 (cpu) + 6*120 (GPU's) worth of rad to cool it efficiently. 6*140 would do

Not sure what you mean by this? what exactly are you referring to? Remeber I'm new to OCing

2. As the temps outside fluctuate so will your OC stability. This will likely be your biggest hurdle. When we hang our radiators out in the cold its for a quick OC'd benchmark, not a 24/7 data cruncher. Just a thought but if you know anyone that's good with programming (I'm not) then maybe write a program to run a variable OC off of the CPU's TJ Max sensor.

I thought about that. I think it will definitely take some tunning to find what my max and min system temps will be base on outside air temps. This is also one of the reason I think I will be building a heatpump into the system to raise the outside outgoing hot well above the outside temp to aid in heat rejection. For this im thinking of just using a few CPU block with Pezos between them and using PWM to regulate it. I can do the programing in things like Pi or even on an Old PLC i have laying around. Some simple PID loops will help keep things stable

 

[Johan45]

Tír na nÓg is talking about radiator area. A 240 rad for CPU and 2x 360 rads for the GPUs is what he's saying.

 

[DieselTwitch]

Ahh! thank you! I was really confused there but now I'm kicking my self.

One of the big things i want to be able to do is reject heat inside when its cool inside. no sense in pointlessly heating the outside air. I know it will be complex but I'd like to have control of where that heat is going. we dont have AC up here so dumping heat inside in the summer is a bad idea

 

[Blaylock]

You could even look into TEC (peltier plates) coolers to regulate the temps.

 

[DieselTwitch]

You could even look into TEC (peltier plates) coolers to regulate the temps.

That's exactly what i will be doing, I just call them eHeatPumps or Pezos. Those combined with some PID loops, and PWM and I can very tightly regulate temps

 

[Blaylock]

Looks to be very cool. (Pun intended)

 

[EarthDog]

Should be awesome...but id delid along with it. The biggest issue with these chips are getting the heat out. Cold doesnt do much good if it isnt getting. Personally, as cool as this is, not sure the reward is worth the effort for a couple hundred mhz. Particularly if you need it to run stable...etc.

Youd also need board with a top notch vrm and keep it cool.

 

[DieselTwitch]

Should be awesome...but id delid along with it. The biggest issue with these chips are getting the heat out. Cold doesnt do much good if it isnt getting. Personally, as cool as this is, not sure the reward is worth the effort for a couple hundred mhz. Particularly if you need it to run stable...etc.

Youd also need board with a top notch vrm and keep it cool.

Im just worried about voiding warrenty on a 1000$ chip... lol but your right about having the greast cooling system in the world and the CPU not beable to tansfer heat.

I also agree with you about the reward vs work and money. that being said. if it was just for one machine it would be a total waste. However I have plans on build bigger and bigger processing machine. eventually rack mounting them. My next build will be a Xeon E5-2699 and then a Dual Xeon E5.... Im hoping to come up with a powerfull enough machine to scale it for my opperation. Normally this is the point that people ask why i dont just get a AWS or off site. well my product has some very strick restrtions on where i can take the data. Privacy and all that.

 

[Blaylock]

Diesel, I just came across this by a member here known as SilverSurfer and at OCN as OCAddict. I think you'll find it right up your alley. LOL

http://www.overclock.net/t/1633988/chilled-water-cooling-vs-3-0-build-log