New Member- designing a immersion, subzero system

Hi,

Craig here...

I'm an old guy who did overclocking before Al Gore invented the internet. My son asked me to help him with designing a water cooled Gaming PC and I got interested in going extreme on the design.

I'm researching:

1. "Direct Expansion (DX)" cooling. 2-phase in OC jargon
2. Liquid Immersion- to address condensation
3. Ammonia refrigerant to be "green" @ ~ -20C temps
4. DX blocks on CPU & GPU
5. DX coils throughout the water bath for forced convection cooling (liquid stirring) of other MB components

Any tips, tricks or warnings much appreciated!

The year of crazy extreme cooling! First LNL Cooling and now you



Lets see how much I can help on this topic. Thermal isn't my expertise, but I think I can help on how thermal works with the electrical components.

So lets say you have a system immersed in mineral oil, with your DX cooling system proposed above. Using the coils will help with stabilizing temperatures inside the system thus nearly eliminating condensation (if we look at the system from a contained point of view). But now you have forced expansion in a sealed container, you will need a vent process to equalize pressure. With a system like this, you could maintain temperatures and condensation throughout the system. Tricky tricky and complicated system! So lets step back and make the system a bit more managable... Open up the pressure valve so that its full open, no more pressure control. In this case we have a rotation of energy that has to be disipated from the system; in this case it will be in the form of condensation around the hotter areas (GPU/CPU/RAM), but we may get cold built up (and chem reaction) around the cooler areas (SB, SATA, anywhere in the southern area of the mobo). But we can easily control this by setting a floor temperature for the DX system (basically figure out a point where the mobo doesn't freeze in certain areas, but cold enough to OC your toys the max).

You will also need to do some modeling on the convection force of your DX coils, and the resistive force (viscosity and something else) of the mineral oil. You will also need to make sure the mineral oil can go down to the temperatures you want, and not change properties at those temps. An increase in viscosity as temperatures lower is a very bad thing. In the case of low energy areas, cold can rapidly build up causing component issue.

For this project, I suggest looking at any data that has been released by the giant data center companies who have done immersion cooling. All of the major OEMs (Dell, HP, etc) have a secret data center that works on this technology. Some public data should be available.

Dolk said:

The year of crazy extreme cooling! First LNL Cooling and now you

Click to expand...

Hi Dolk- thanks for the welcome and the feedback.

I'm looking to operate the whole liquid bath at about -20C. Mineral oil would be too viscous at that temp, so looking at something like 3M™ Novec™ 7000 Engineered Fluid I wouldn't use like the server farm folks where the fluid boils on the hot CPU @ 34C and is re-condensed using cooling tower water or some other ambient cooling source. I'd be using the fluid as a non-conductive low viscosity heat transfer fluid. I'm thinking I put the DX evap coil in the fluid and move the coldest fluid from the coil to the cPU and GPU. Straight force convective cooling of the chips. The rest of the bath would be at a slightly higher temperature.

Someone tried to do this a couple months ago...when I am not mobile I will look for the tread....

... but search novec at the site and it should bring up his thread.

The Novec is still going to boil at the contact point of CPU and GPU die.
You will still have to seal the system off to condense the fluid and to stop contamination of your room.

This seems like a great idea for a test project but as a daily driver I don't think your son is going to like this very much.

Soulcatcher668 said:

The Novec is still going to boil at the contact point of CPU and GPU die.
You will still have to seal the system off to condense the fluid and to stop contamination of your room...

Click to expand...

roger that... It'll have to be a hermetically sealed tank. The most challenging thing I see so far will be getting all the USB, HMDI, HD, etc connections sealed and external.

Re boiling, I'm thinking if the fluid is cold enough (~-20C), and I get good agitation of the liquid, it won't get hot enough to boil. If it does, the vapor bubbles will immediately condense in the cold fluid.

CraigRochester said:

Re boiling, I'm thinking if the fluid is cold enough (~-20C), and I get good agitation of the liquid, it won't get hot enough to boil. If it does, the vapor bubbles will immediately condense in the cold fluid.

Click to expand...

It is going to boil. That is what the fluid is designed to do. The phase change from liquid to gas is where the cooling power comes from. That is why it works.

Soulcatcher668 said:

It is going to boil. That is what the fluid is designed to do...

Click to expand...

It's design to boil above room temps (37C for Novec 7000). If the temp never gets above its boiling point, it won't boil. I wouldn't us it for its 2-phase properties, only low conductivity and low viscosity at sub zero temps. The stuff is ~$400 a gallon, so will use something else if I can find it.

While certainly interesting, won't this be much less efficient and effective than direct die phase change with an oil bath to prevent condensation? Or is that exactly what you are planning to do, in addition to cooling the oil bath?

matttheniceguy said:

While certainly interesting, won't this be much less efficient and effective than direct die phase change ...?

Click to expand...

I don't know. it's a complicated heat transfer comparison. DX in the die gives the lowest temp at the CPU top. If the whole area surrounding the chip is cold, you've got conduction, convection and radiation cooling in all directions. I need to do more reading on the subject. Just having a evap coil in the bath with good stirring is a hell of a lot easier.

CraigRochester said:

It's design to boil above room temps (37C for Novec 7000). If the temp never gets above its boiling point, it won't boil. I wouldn't us it for its 2-phase properties, only low conductivity and low viscosity at sub zero temps. The stuff is ~$400 a gallon, so will use something else if I can find it.

Click to expand...

It will boil where the fluid contacts the CPU and GPU dies.

I'll be taking bets. What odds do you want?

CraigRochester said:

I'll be taking bets. What odds do you want?

Click to expand...

I have bets on boiling because I have experience in thermodynamics and heat transfer.
Small area, high concentration of power. Phase change will almost definitely happen, I have about a 95% confidence in this.

Even with the added cooling of pulling the fluid temperature down, you'll see this:

ATMINSIDE said:

I have bets on boiling because... I have about a 95% confidence in this...

Click to expand...

I take $1 at 20:1

Seriously though, I'll do some calcs and see what a simple HT model says. Do you have info on the thermal distribution and power density of CPU chips?

put a heatsink on the cpu then havea pump configured to push liquid over the heatsink and ther eye go

CraigRochester said:

I take $1 at 20:1

Seriously though, I'll do some calcs and see what a simple HT model says. Do you have info on the thermal distribution and power density of CPU chips?

Click to expand...

What, no faith in your guess? hehe!

CraigRochester said:

I take $1 at 20:1

Seriously though, I'll do some calcs and see what a simple HT model says. Do you have info on the thermal distribution and power density of CPU chips?

Click to expand...

Only going to pay me $1??

Power draw depends on what CPU you're using. Take the size of the IHS and the TDP of the chip, there's what you need.

And, thinking about it, I'll change to 99.5%.
The thermal conductivity of Novex 7000, until phase change occurs, is abysmal at best

Nope, will be collecting $20. ;-)

Your 20:1 bet shows you are not remotely confident there Craig...