GIGABYTE Server Shows Two-Phase Immersion Liquid Cooling

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
While I know exactly squat about designing an immersed system, I can probably suggest that this looks cool and serves as a novel approach to the cooling game. It is hard to say if this setup has any real future in data centers, however—as the article points out, there is that little issue of replacing hardware.

GIGABYTE and 3M had submerged a full 8-GPU, dual CPU system with 24 memory modules and nothing more than large copper heatsinks on the CPU/GPU, and had even removed the power delivery heatsinks. To cool the vapor as it rises through the system, a cold radiator is placed inside the sealed system. Well, I say sealed, but during the demo it was being opened and the demonstrator was clearly putting his hand inside. There seemed to also be a system in place to add/remove hydrocarbon material through a pump as well. So the point in all this is more efficient cooling – no need for massive air conditioning units in a data center, no need to pump chilled water into water blocks.
 
I don't know much about this, but how do you keep the liquid from heating up too much? Does it get pumped through radiator like a normal closed-loop system?
 
I don't know much about this, but how do you keep the liquid from heating up too much? Does it get pumped through radiator like a normal closed-loop system?

Watch the video, it has a "secondary cooler" that the fluid condenses on and drips back down to the cooling tank.
 
Watch the video, it has a "secondary cooler" that the fluid condenses on and drips back down to the cooling tank.

Gotcha, the link above for the other system looks like it is similar. Still, I have to wonder if the surrounding liquid would heat up locally but not enough to vaporize and cause temperatures to rise over time.
 
Gotcha, the link above for the other system looks like it is similar. Still, I have to wonder if the surrounding liquid would heat up locally but not enough to vaporize and cause temperatures to rise over time.

From what I know, the biggest issue was cavitation, where there would be more bubbles than liquid. It would then become a feedback loop until either load was reduced to cool the part, or something burned up.

Heres another video with higher power parts.

 
The liquid itself boils off at around 35C so all components in the liquid are kept cool. Also to change hardware all you do is remove the existing hardware and replace it with new hardware. Unlike traditional oil based immersion coolers you don't need to clean off the contacts and such, it's just pull, plug and play.
 
This chemical compound they are using looks like a new twist on the older HFE, HydroFlouroEther from about what? 12 or so years ago?
 
Now the question is, how much does this liquid cost and how often do you have to replenish it?

Rough estimate going off of 3M Novec searches looks to be around $400 / gallon.
 
Now the question is, how much does this liquid cost and how often do you have to replenish it?

Rough estimate going off of 3M Novec searches looks to be around $400 / gallon.


That is similar to what I remember HFE going for about 10+ years ago. I was enough to make experimenting with it as an amateur a bit too much. That and the Material Handling Sheet said that it could be dangerous in vapor form and when it came to using it in a heat-change setting I figured vapor state would be surety so ...
 
its actually fairly easy and cheap to do somthing like this (not like the fancy 3m shit) if i remember right some dude owns all the patents for submerged pc's and thats why you have never seen it come to market. pudget systems even had a retail modle for abit before they had to take it down.
 
I'd make an argument here for the $40 HSF solution for dissipating unwanted heat....but I don't want to get shouted-down by you oil-crazed zealots...........
 
My question is, why not distilled water? I know that it is hard to keep it pure enough to not conduct electricity, but it does work.
 
its actually fairly easy and cheap to do somthing like this (not like the fancy 3m shit) if i remember right some dude owns all the patents for submerged pc's and thats why you have never seen it come to market. pudget systems even had a retail modle for abit before they had to take it down.

pudget and Abit were teaming up? I guess stranger things have happened.
 
My question is, why not distilled water? I know that it is hard to keep it pure enough to not conduct electricity, but it does work.
The Novatec boils at something like 34 C. The heat is carried with the boiled gas to a radiator at the top. If the tank were filled with water, it and your components would have to heat to 100 C before you achieved the same effect.
 
The Novatec boils at something like 34 C. The heat is carried with the boiled gas to a radiator at the top. If the tank were filled with water, it and your components would have to heat to 100 C before you achieved the same effect.
if i pull my water down to 1 psi i could achive this just fine.
 
Lol everything old becomes new again.

They usually use a special 3m mineral oil and there are a few other types available too.

The main reason no one has done this large scale in last 20+ years is servicing becomes much more difficult, transportation becomes difficult and oil wicking has to be addressed with cables if off shelf parts are used.
 
Lol everything old becomes new again.

They usually use a special 3m mineral oil and there are a few other types available too.

The reason no one has done this large scale in last 20+ years is servicing becomes much more difficult.

and patents :p but yea i remember someone showing a submerged dual x5698 (2core 4.2ghz) overclocked to 6+ a long time ago. And i may give a cool variant of somthing like this when i get the free time to mess around.
 
and patents :p but yea i remember someone showing a submerged dual x5698 (2core 4.2ghz) overclocked to 6+ a long time ago. And i may give a cool variant of somthing like this when i get the free time to mess around.

Thing is I bet there is a way around it, often many ways to circumvent patents and design something a little different or slightly differently functioning, that I feel it's more than just patent reasons.
 
Thing is I bet there is a way around it, often many ways to circumvent patents and design something a little different or slightly differently functioning, that I feel it's more than just patent reasons.
either way ill be giving somthing "slightly differnt" a try soon and ill see if its practical then maybe go form their.
 
either way ill be giving somthing "slightly differnt" a try soon and ill see if its practical then maybe go form their.

Nice, will you post about it?

This is a long term project I've always wanted to do. Key is to get decent passive heatsinks. It seems large fin pitch like passive designs (my OG scythe ninja rev 1 has this) work best in these environments. If you think about, it's a passive system in a thicker medium...
 
Nice, will you post about it?

This is a long term project I've always wanted to do. Key is to get decent passive heatsinks. It seems large fin pitch like passive designs (my OG scythe ninja rev 1 has this) work best in these environments. If you think about, it's a passive system in a thicker medium...

i will defenently get a post when i do it however i dont want to give away too much as it is still just a bubble in my head and afew sketches. however feel free to cheack out my curent cooling project. i need to do abit more to finish it up and i need to update the thread :p
https://hardforum.com/threads/smart-thermo-electric-cooling-system.1917470/
 
Nice, will you post about it?

This is a long term project I've always wanted to do. Key is to get decent passive heatsinks. It seems large fin pitch like passive designs (my OG scythe ninja rev 1 has this) work best in these environments. If you think about, it's a passive system in a thicker medium...

If you're going down a boil-off route, you want a flat metal copper plate on the chip, with lots of nucleation zones (think like golf ball dimples on a much much smaller scale). That's the fastest for that type of cooling. If you plan on using conventional oil based then yes, big HSF is best. The fan will still be fine too.
 
Oooh, I have an idea. We should use this liquid, just on the CPUs, but have it in a tube for each CPU. Then when it boils off, it will go to the top of the tube, and then we can cool it off there, and it will condense back down and drip down inside the tube.

We could call it, Thermo Engine.
 
I went on a tour of the Cray facilities in Chippewa Falls WI a few years ago. They said they had experimented with submerged systems but due to the way they ran their processors the liquid would boil too quickly and couldn't provide enough cooling. The system I saw was described as using a form of liquid Kevlar which was sprayed onto the the motherboard in a constant stream. The liquid would be collected below then run through a cooling radiator before being sprayed again.
 
it would be neat if they found some type of silicon replacement that didn't heat up or it cooled as a reaction (man made diamonds!). I think we are only a few years away from alternatives to traditional wafers (nano carbo, cheap man made diamonds, bio-living devices etc).
 
What was that movie where the supercomputer cores dipped directly into a pool of some kind of fluid.... Sunshine?
 
What was that movie where the supercomputer cores dipped directly into a pool of some kind of fluid.... Sunshine?

Know the movie, can't think of the name, but they've been doing immersion cooling in old-style (no off the shelf components) supercomputers for decades. I've got a cpu board from a Cray 1a that was immersion cooled back in the day.
 
Back
Top