Overclock.net banner

Designing a liquid metal cooling system loop (yup, you heard me right)

11552 147
This all started about two weeks ago when I purchased a chiller for experimental purposes. I was scanning ebay/newegg one night for various different types of coolant and brainstorming on how I could come up with some substance a little more exotic than water, or additives I could use to enhance heat exchange. This is when the thought popped into my head that cooling a PC with liquid metal may very well be a viable and feasible option. And something I might actually be able to pull off, given enough time and research. One of my side topics of interest are nuclear reactors, so that was what prompted me in the first place, as some nuclear reactors are actually cooled by liquid metal and it is extremely effective, and known as the "ultimate" coolant. I want to get a few things straight so you guys don't think I've gone off the deep end:

1) This is a long term project - It will likely take at least 6 months to a year to design and build and impliment what I am envisioning.
2) This is not practical in any sense of the word - I find it very intriguing, however, and I have a very curious mind, so yes, while early on it will be experimental, I do eventually hope to come up with a liquid metal cooling "solution" that perhaps could be developed into a real product for overclocking enthusiasts.
3) This is not going to be cheap. Gallium costs about $100 for a 1/2 cup, to put things in perspective.
4) This is totally experimental for the moment, but I think I can make it work, and if I do I will try to come up with a kit or formula that other hardcore OC people can buy and implement themselves.
5) This will be a great covid project as I dont see us coming out of this pandemic anytime soon lol

I first came across Mercury, obviously one of only a few metals that is a liquid at room temperature. But this is nasty stuff, and the vapors alone are enough to keep me away, far away. Then of course there is Gallium, which I think with the right tweaking will make the most effective "liquid metal coolant" for this project. Gallium has some very cool properties and has a melting point of 85*F, which is obviously to high to be used practically in this scenario. It is also very corrosive to aluminum so all the parts used in this liquid metal loop will have to be compatible with a gallium based alloy.

Galinstan - Wikipedia
So I did a little more research and came across a very interesting alloy known as Galinstan, mixture of gallium and indium (and a pinch of tin) which has a melting point of -19*C (-2*F). I need to do more research but if I really want to take advantage of liquid metal as a high performance coolant, I need to involve a chiller in this conversation. Because in a normal loop, with standard radiators you might only see 5-10% increase in thermal dissipation properties of LM vs standard water loop. If you can get the heat out of the liquid metal very quickly, you get more bang for your buck, and can further capitalize on it's superior thermal dissipation properties. In other words, my plan is to utilize a chiller and a plate heat exchanger which will interface the liquid metal loop with coolant loop from the chiller at about 33*F. A plate heat exchanger is a very interesting piece of equipment in that it offers a huge amount of surface area verses a standard radiator, and a radiator is air to water, whereas the plate heat exchanger is liquid to liquid. So right off the bat, we will be using a highly effective form of heat transfer (liquid to liquid is far more effective than liquid to air). The project requires two independent loops, at least as I see it right now (see graphic below). The liquid metal loop goes through the plate heat exchanger where it gets cooled by the chiller (and the chiller will have it's own independent loop and the two will never touch. For all intensive purposes, a plate heat exchanger is the most effective heat exchanger for use with this kind of idea and will provide well over 20x the surface area of a standard radiator. That's the key here in allowing us to flip conventional thinking about radiators on its head, as I will outline my theory below.

My goal is to run the liquid metal loop with a target goal of maintaining it at about 35*F in that loop. And this is where things start to get interesting. In a conventional loop, the radiators get hot, because you are removing all that heat generated from your CPU/GPU. But then it dawned on me, why would I want hot radiators INSIDE my case? I mean, in the conventional sense, a radiator in a house is designed to heat the house. Do I really want 90*+ radiators in a confined space around very temperature sensitive electronic equipment, especially when I'm overclocking? The rational answer is NO.

What am I getting at? We will use the liquid metal to run the entire loop sub-ambient, so effectively, you will have 35-40* liquid metal flowing through your radiators. In other words, use the radiators to cool your internal case temps, not raise them. Because the liquid to liquid heat exchanger is so much more effective than liquid to air, any decent chiller should be able to maintain the entire loop at sub ambient temperature without to much trouble. There is a very common misconception that running radiators with a chiller is a no no, but actually, when you study it closely, the chiller should have no problem dealing with the temp increase in LM from passing through your radiators. This is effective air conditioning for your system. even if you have two or three large radiators, that's still nowhere near the surface area of the 60 plate heat exchanger I just purchased for use with this project. Yes, there will be condensation and other challenges, but one by one, I will address them.

This LM loop would require a different type of pump (electro-magnetic MDF pumps in particular would be ideal). Taking advantage of the metallic nature of the coolant, the pump pushes the liquid around electromagnetically. Unlike water cooling, the process requires no moving parts, consumes little power, and is silent.

The attraction of liquid metal itself is its excellent conduction of heat, it is 65 times more thermally conductive than water – and 1,600 times better than air cooling.

In short, liquid metal is able to absorb heat more rapidly, and thus cool down chips faster. This property has led to its use as an “ultimate” coolant in some nuclear reactors, which are cooled with liquid sodium or potassium, as well as in the manufacture of high-quality machine components, such as gas turbine blades, where the components are rapidly “cooled” to 660 C with molten aluminum to prevent the formation of defects.

Before you laugh, just know that there are proven liquid metal loops already out there, designed for use with mobile devices all the way up to servers. Nothing holding me back from taking a couple pages out of their playbook for my little project.

This project is in it's intial phases. But I am proceeding with the idea and I will make it happen. Your comments are welcome, both good and bad, just please dont post things like "dont do it" or "it wont work" because I am doing it and WILL make it work.

But your advice is appreciated because I know we have expert opinions out there who can help me or guide me in the right direction to make this whole experiment viable.

Product Guitar accessory Musical instrument accessory String instrument accessory Gadget


Essentially, this layout is similar to what I am envisioning in my head (sump water would be chiller) and liquid-liquid heat exchanger in the center would be the plate heat exchanger)

Product Font Line Parallel Rectangle


Product Rectangle Font Slope Parallel
  • Rep+
Reactions: damniel
121 - 140 of 148 Posts

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #121 ·

· Banned
Joined
·
4,658 Posts
I had some time today to respond to old comments...


Take a hint: It's called exotic cooling.

Actually, you are wrong again. According to gamers nexus, nickle does alright wth gallium and galinstan. Truth fail, again.


can I play with mercury instead? I wanted to get your feedback specifically since you are making a foolish attempt to sound like you actually know what you are talking about here.


No way! Obviously, they will be shielded.


Are you bench racing again? How many times do I have to point out your credibility is in the tank? LOL you didn't even know the typical chiller hot loop / cold loop layout and configuration / arrangement. And two loops are generally used, that's the conventional chiller setup. I suggest you use gasoline to build your next loop. It's fun to live dangerously, right? 😂

Here, a simple reminder (count the number of loops):
View attachment 2532627
Making fun of the concept is getting old. I’ll be back when it’s completed.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #123 ·
Making fun of the concept is getting old. I’ll be back when it’s completed.
Please don't go. It was fun watching you make a fool out of yourself over and over again! 😂
 

· Banned
Joined
·
4,658 Posts
Please don't go. It was fun watching you make a fool out of yourself over and over again! 😂
This thread has too much talk and not enough building. As much as I like to troll, I still have my limits. If you spent your time building instead of replying to comments you’d be done by now.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #125 ·
This thread has too much talk and not enough building. As much as I like to troll, I still have my limits. If you spent your time building instead of replying to comments you’d be done by now.
ok you must have missed the first post where I said it would likely take longer than 6 months to complete. Im still in the study phase.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #126 ·
What CPU block? I don't think your pump pressure will do well forcing chilled liquid metal through microchannels. I would experiment with newer and older CPU blocks for flow rate; the old Fuzion ver 1/2 or even a maze CPU block would probably work better here.
Agreed. Do you guys have any suggestions on what I should start testing with? Obviously it's not going to be a copper water block with micro channels. Perhaps Nickle plated copper. Besides, my heatkiller IV is already in need of another cleaning. In the interim I've added a third pump just to force water through the bottlenecked water block. The two pumps on the other side of the water block worked fine but I wanted to balance it out a little (as in push - pull, not just pull - pull) until I can get around to pulling the water block AGAIN and using an industrial pressure washer on it. So much for "loop filters" they did absolutely nothing to stop smaller particles from getting into the micro channels and creating a block.

EDIT: CNC machine is set to replicate the maze from the shining, only very small of course.

It's going to be hilarious when I eventually get this project up and running to see the look on your face. Like I said, keep making yourself look foolish and then we can all laugh a little more at the end when I complete what I set out to do and all you did was sit in the peanut gallery and act like a nimrod.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #127 ·
Okay new pump placement isn't so great but it's only temporary, at least I can once again go to 5.3 for benching with this rig....



 

· Registered
Joined
·
2,618 Posts
Agreed. Do you guys have any suggestions on what I should start testing with? Obviously it's not going to be a copper water block with micro channels. Perhaps Nickle plated copper. Besides, my heatkiller IV is already in need of another cleaning. In the interim I've added a third pump just to force water through the bottlenecked water block. The two pumps on the other side of the water block worked fine but I wanted to balance it out a little (as in push - pull, not just pull - pull) until I can get around to pulling the water block AGAIN and using an industrial pressure washer on it. So much for "loop filters" they did absolutely nothing to stop smaller particles from getting into the micro channels and creating a block.

EDIT: CNC machine is set to replicate the maze from the shining, only very small of course.

It's going to be hilarious when I eventually get this project up and running to see the look on your face. Like I said, keep making yourself look foolish and then we can all laugh a little more at the end when I complete what I set out to do and all you did was sit in the peanut gallery and act like a nimrod.
Sounds more like poor loop cleaning if you have had to clean the block out once before, have filters, and it's already clogged again..
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #129 ·
Sounds more like poor loop cleaning if you have had to clean the block out once before, have filters, and it's already clogged again..
Yes to some extent (however, the filters have now been removed because small particles would go right through them anyway) and the fact that I didn't have an adequate pressure washer the first time I had it apart. I'm just limping along until the new coolant arrives, then it will get a proper cleaning, including both reservoirs and fresh coolant. Just didn't want to waste the new coolant on a half way clean loop.
 

· Registered
Joined
·
891 Posts
Won't the gallium pull heat out of the microscopic fins faster than pure water despite lower flow rates? It only needs to transfer the heat away from the coldplate to the cooling system on the 2nd loop, right?
What's the flowrate/cooling perf of the gigantic plate heat exchanger in the OP?
Does the 2nd loop go to window/outside the home by any chance?
Are you testing this with a 10980XE or the W-3175X/LGA3647 or the LGA4677 cpu in ~10months?
Light Rectangle Slope Font Triangle
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #131 · (Edited)
Won't the gallium pull heat out of the microscopic fins faster than pure water despite lower flow rates?
It should, after all gallium is a wet liquid metal, meaning it should theoretically flow through the micro channels but it is in fact twice as viscous as water, so the micro channels may need to be cut again just a little bit wider to account for the difference. idk then again it might do just fine. It's not very much more difficult to push liquid metal vs water. But the flow rate must remain high because the LM will more readily absorb heat when compared to water. So this is crucial, I must have fairly high flow rate to fully capitalize on LM performance attributes.

It only needs to transfer the heat away from the coldplate to the cooling system on the 2nd loop, right?
Correct. Still working out the details of the second loop but more or less:
Product Rectangle Font Slope Parallel


What's the flowrate/cooling perf of the gigantic plate heat exchanger in the OP?
Careful with the term gigantic. Doesn't even come close. :ROFLMAO:
Textile Automotive tire Sleeve Gas Engineering

It's totally dependent on your pump not the heat exchanger - it's (my 60 plate HE) not restrictive at all. I did some testing last week on flow and concluded the liquid to liquid heat exchanger does not pose a restriction whatsoever in regards to flow rate through the loop. in fact for kicks i tested the mini pump that came with the chiller and it had no trouble pushing water through it either.

Does the 2nd loop go to window/outside the home by any chance?
No. Second loop will be completely under the hood with the first, might go pipe in pipe design, it's slightly radical but it might just be the ticket. I'm not using the above computer for this LM project, wasn't my intent to imply that. z820 has numerous advantages for testing this concept. Tool less case, Lots of room, lots of power and excellent airflow.

Are you testing this with a 10980XE or the W-3175X/LGA3647 or the LGA4677 cpu in ~10months?
I'm going to be testing with one of my dual processor z820 rigs.
This particular rig has two OEM Xeon E5 2696 v2 processors, which surprisingly outperform Intel's flagship retail 2697 v2 in multi core performance due to +100Mhz all core turbo (3.1GHz all core, 3.5GHz single). It makes for 24 cores and 48 thread when paired up. Enough to just take down a 10980XE in the cpu-z multi-core performance benchmark.
Rectangle Computer Font Screenshot Parallel


So here is the plan guys:
Pull two z820 liquid coolers from storage. I'm stocked up for the end times lol.
Grille Motor vehicle Automotive lighting Automotive tire Hood


This is what they look like from the back:
Watch Automotive lighting Camera accessory Flash photography Cameras & optics


Believe it or not, these little buggers are capable of cooling a 2687w v2! In fact they are mandatory for that processor, which is a 150W TDP processor. As you can see, its a closed system. But very easy to take the sleeve off and drain, flush and fill with gallium (galinstan) and get right to testing. This is probably the most compact way in which I can create a prototype design to test and get meaningful results from, since I can always go back and reference performance characteristics with water as a coolant as well.

This is going to be a sacrificial test. Nothing is going to suffer from LME but I will probably have to chuck the gallium after testing is complete because it will react much more readily with copper, of which 99% of the radiator is. It wont compromise the cooling capacity of the LM, however, which is why it's applicable to test in this scenario and I can still get accurate results I can then use towards a slightly larger loop.

In this case, the z820 liquid cooling system is ideal for testing liquid metal as a coolant. First off, the pump is built into the base, second, it requires very little coolant to fill the loop. Third, everything is self contained. Forth, I have a very acute idea how these coolers perform so it will be interesting to see if the LM is going to be better right off the bat.
 

Attachments

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #132 ·
What's the flowrate/cooling perf of the gigantic plate heat exchanger in the OP?
You should be able to do the math and translate over from the much smaller one you posted earlier.
Mine is more or less dimensionally the same in terms of footprint? But get the spect sheet and then compare how many plates it has. Mine has 60 stainless steel 316L plates which should hold up just fine. So clearly it's all about surface area. I want to focus on delta T with LM. Plus liquid to liquid HE are much more effective at extracting heat than any other type of heat exchanger, including air to water.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #133 ·
It only needs to transfer the heat away from the coldplate to the cooling system on the 2nd loop, right?
Right, after passing CPU, heat will go from the liquid metal(hot side) to water (cold side) via plate heat exchanger, and then once the heat is transferred into the 2nd loop (the cold loop) that second loop will be cooled by:
a) Chiller
b) conventional means
c) a simple, conventional, standard, custom loop.

One of the benefits of running a liquid to liquid heat exchanger and having radiators on the cold side is that I wont have to deal to much with LME or degradation of LM itself since the aluminum radiators are on a secondary loop which only has water in it. And I can initially build the LM portion of the loop with a very minimal amount of volume and no reservoir, so I wont have to spend $500 on LM coolant just to build my prototype.
So at first, I will be starting with a very small LM loop and a much larger water (cold side) custom loop, just like you see here. Part of the reasoning is that if I have twice the volume of water as opposed to LM, I'm going to be able to extract more heat from the LM side and never max out the cooling capabilities of the secondary cold loop. Hope this makes sense to you guys.
 

· OC...D
Joined
·
28,261 Posts
Those z820 coolers work real well! Waiting to see that thing pumping metal! 🤞
 
  • Rep+
Reactions: storm-chaser

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #135 ·
Those z820 coolers work real well! Waiting to see that thing pumping metal! 🤞
I think I need to spend like $80-90 for the Galinstan, much better than the $500+ I had originally intended to purchase. Plus, this operation should be very straight forward and we will get meaningful data immediately.

Check back around the 5th of next month and around that time I should have it up and running.
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #136 ·
Okay Guys, I think I found a suitable work around for extracting as much heat from the liquid metal as possible. The problem that I face with a liquid metal loop is that the "heat" will be moved quickly away from radiators/ heat exchangers, calling into question the efficacy of the loop to actually cool the liquid metal effectively.

Product Rectangle Font Parallel Cylinder


OR

Rectangle Font Parallel Diagram Magenta


Rectangle Font Slope Parallel Technology



It's called a counter flow double heat exchanger... get the idea? We have the liquid metal flowing in red, and cold coolant flowing the other direction, this way we can extract as much heat from the LM metal as possible and loop velocity will not compromise cooling performance.

Rectangle Line Font Automotive design Parallel


And this is standard operation of the 60 plate heat exchanger, which will be used in conjunction with the above double pipe heat exchanger for max heat transfer..

Slope Rectangle Parallel Font Triangle
 

· Registered
Joined
·
2,089 Posts
Won't the gallium pull heat out of the microscopic fins faster than pure water despite lower flow rates?
That depends. Convective heat transfer requires a materials thermal conductivity yes... But velocity and turbulence matters as well. Thank about air, it has no business moving heat away with its "conductivity" rating. But when forcefully pushed over fins with some turbulence, it extracts heat and that's due to flow rate!

The convection coefficient is a function of both conduction and velocity of fluid flow via the Nusselt Number (Nu).


h = (Nu * k) / L where Nu is found via empirical data and some partial differential equations :D, Nu is dependent on fluid flow and fluid flow characteristics (laminar, turbulent).

TL;DR: h is not a easy parameter to follow without testing in a lab.

Even though the k value is high, what if your Nu is very very low... then h might not improve much. And h is the major contributor to heat transfer in convection.
 
  • Rep+
Reactions: storm-chaser

· FX-5000
Joined
·
994 Posts
This reminds of the time I almost built a water loop with Acetone and DryIce as the chiller. Instead of using a DryIce pot, I thought this would be a neat approach to a water loop. Keeping in mind the dangers, issues with rubber o-rings dealing with acetone and then boiling issues actually inside the waterblock, I decided against this project.

Neat to see someone come out with a similar idea and actually go for it. I had to chicken out. It's just easier to use a Dice pot and be done with it I guess is the final thought.

Speaking of thoughts, through many years of cooling experiments, TEC is one of my favorites.

Let me share with you the number 1 issue I always came across.

The IHS plate on modern chips are just too small. Too thin. And the surface area of modern silicon hasn't changed much, there's just more of them now. 2 to 3 chips to cool under a single IHS plate. Which again, is just too small.

The Very first problem with cooling a processor is only being able to pull heat in a single direction for a single surface area. I call this in my head 2D cooling. The IHS plate has 6 sides, you only remove heat from one side. So that becomes and issue. Then the thermal mass of such a small plate (IHS) makes it so you need to remove heat basically immediately.

I found that removing the plate and replacing it with one with about 4 times the amount of mass, I was able to slow the exchange process from CPU to the cooling, in my testing cases a TEC heat exchanger. To put it simply, it took longer for the CPU to heat the enlarged plate, or, because a TEC does the chilling, longer for the TEC to get the plate cold. But once cold, would take a considerable amount more energy and time to warm the plate up again. Despite the cpu core reading a 15c positive temperature at load, I was able to keep the new enlarged plate frozen. This helped quite a bit for my experiments.

Anyhow, that's the basics of any findings of my recent experiments. Transistor density is our issue more so than the coolers we use, but I really enjoy experiments like this one. I hope it works well, but if it doesn't, then you're learning something. Good Luck!
 

· Twin Turbski
Joined
·
2,313 Posts
Discussion Starter · #139 ·
This reminds of the time I almost built a water loop with Acetone and DryIce as the chiller. Instead of using a DryIce pot, I thought this would be a neat approach to a water loop. Keeping in mind the dangers, issues with rubber o-rings dealing with acetone and then boiling issues actually inside the waterblock, I decided against this project.

Neat to see someone come out with a similar idea and actually go for it. I had to chicken out. It's just easier to use a Dice pot and be done with it I guess is the final thought.

Speaking of thoughts, through many years of cooling experiments, TEC is one of my favorites.

Let me share with you the number 1 issue I always came across.

The IHS plate on modern chips are just too small. Too thin. And the surface area of modern silicon hasn't changed much, there's just more of them now. 2 to 3 chips to cool under a single IHS plate. Which again, is just too small.

The Very first problem with cooling a processor is only being able to pull heat in a single direction for a single surface area. I call this in my head 2D cooling. The IHS plate has 6 sides, you only remove heat from one side. So that becomes and issue. Then the thermal mass of such a small plate (IHS) makes it so you need to remove heat basically immediately.

I found that removing the plate and replacing it with one with about 4 times the amount of mass, I was able to slow the exchange process from CPU to the cooling, in my testing cases a TEC heat exchanger. To put it simply, it took longer for the CPU to heat the enlarged plate, or, because a TEC does the chilling, longer for the TEC to get the plate cold. But once cold, would take a considerable amount more energy and time to warm the plate up again. Despite the cpu core reading a 15c positive temperature at load, I was able to keep the new enlarged plate frozen. This helped quite a bit for my experiments.

Anyhow, that's the basics of any findings of my recent experiments. Transistor density is our issue more so than the coolers we use, but I really enjoy experiments like this one. I hope it works well, but if it doesn't, then you're learning something. Good Luck!
Hi Shrimp. Still delidding? Perhaps I will send you my 9600KF (or 1680 xeon v2) to delidd so I can put your theory to work here in my setup. People have been saying, I need direct die cooling here to really capitalize on heat transfer. I just like the exotic challenge of making liquid metal work in a loop.

Most of the parts are here for the new z820 build. Yes, I'm building a brand new z820 despite the fact I already have two of them. This is going to be the experimental machine and I didn't want to compromise the other two, that's why I'm starting fresh with a different unit.

CPU showed up today. It's a xeon 1680 v2, that has an unlocked multiplier.
Memory also showed up. It's DDR3 1866R. It's a 48GB kit, (6x8GB) but I will be adding two more identical dimms to get to quad channel memory (64GB)
I didn't want to destroy any z820 liquid coolers from my stock so I purchased one specifically for this project and it is also here.

Just waiting on the z820 computer itself AND the wireless usb card.

With the unlocked multiplier I will really be able to push the CPU hard and that will give me pretty good insight into how liquid metal performs during overclocked circumstances. Sources tell me I should be able to go to 4.6 or 4.7GHz no problem.
 
  • Rep+
Reactions: ShrimpBrime

· FX-5000
Joined
·
994 Posts
Hi Shrimp. Still delidding? Perhaps I will send you my 9600KF (or 1680 xeon v2) to delidd so I can put your theory to work here in my setup. People have been saying, I need direct die cooling here to really capitalize on heat transfer. I just like the exotic challenge of making liquid metal work in a loop.

Most of the parts are here for the new z820 build. Yes, I'm building a brand new z820 despite the fact I already have two of them. This is going to be the experimental machine and I didn't want to compromise the other two, that's why I'm starting fresh with a different unit.

CPU showed up today. It's a xeon 1680 v2, that has an unlocked multiplier.
Memory also showed up. It's DDR3 1866R. It's a 48GB kit, (6x8GB) but I will be adding two more identical dimms to get to quad channel memory (64GB)
I didn't want to destroy any z820 liquid coolers from my stock so I purchased one specifically for this project and it is also here.

Just waiting on the z820 computer itself AND the wireless usb card.

With the unlocked multiplier I will really be able to push the CPU hard and that will give me pretty good insight into how liquid metal performs during overclocked circumstances. Sources tell me I should be able to go to 4.6 or 4.7GHz no problem.
I haven't done a cpu in a year'ish. Last one was an EYE-7 930. It did help, but the chip wasn't the greatest clocker in the world...

4.6ghz... I'd hope for higher frequency. But this frequency seems normal to the generation of the chip.....
 
121 - 140 of 148 Posts
Top