liquid metal cooling loop

[Fast_N_Curious]

Im in the process of designing/eventually building a liquid metal cooling loop. Yes, you heard me right. 

 

The continued popularity in liquid metal as a cooling paste peaked my interest. 

 

My first test piece is going to be on an HP z820 workstation liquid cooler. These are compact, don't hold much fluid and will give me very valuable information / data on how the liquid metal performs relative to that of water. If it performs on par or better, I will continue the project. If it's worse, I will still continue with the project. lol In other words, this project is not practical in any sense of the word. So know that first and foremost. It's also possible that water is a better medium, but we will certainly find out.

 

 

z820 pump/rad combo. Pull one of the lines, drain old coolant then add liquid metal. Super simple.

 

 

One of these liquid coolers alone is actually 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 with pump built in at the base. 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, 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. Galinstan is about 90% gallium, 9% iridium and 1-2% tin. This alloy has a much lower melting point than gallium alone, somewhere along the lines of 32*F! So I could use a chiller on this PC, however, thermal conductivity goes up exponentially as temperature rises, so a chiller might be counter-productive. 

 

The liquid metal I'm using here is almost identical to major suppliers of LM for use as thermal paste. So any research done on that will translate directly to this project.

[Mick Naughty]

So youre expecting that pump to actually move the metal at a sufficient rate and last any amount of time?

[AbydosOne]

23 minutes ago, Storm-Chaser said:

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.

IIRC gallium will attack pretty much any metal routinely used in PCs. You're going to end up with a crumbly mess, more than likely.

[Enderman]

Glad to see someone actually trying this out, it is a widely discussed idea from a long time ago.

 

I don't think it will work great, since the high conductivity is made almost irrelevant due to the fact that the heat is being moved by the physical movement of fluid by the pump, rather than thermal conduction through a solid, but it will still be a cool test

[Enderman]

I would be concerned about the viscosity and surface tension making the metal difficult to flow through the microfins in those CPU blocks though.

[HM-2]

12 minutes ago, Enderman said:

I would be concerned about the viscosity and surface tension making the metal difficult to flow through the microfins in those CPU blocks though.

There's a myriad of good reasons that liquid metal is seldom used as a coolant in anything other than in fast-neutron reactors. In fact, the only reason they're popular in that particular application is the combination of low moderation characteristics, extremely high boiling points and the ability to operate without excess pressure (and thus limit the likelihood of catastrophic failure).

 

Gallium has about three times the viscosity of water and around double that of mercury. You'd actually probably have much better luck with a liquid mercury loop using standard water cooling components, but of course the issue there is toxicity. 

[Levent]

Not gonna work. You gonna kill pumps and corrode radiators before even getting half acceptable testing done. Gallium and mercury have been around for ages, there are reasons why they arent used as coolant in non nuclear applications. 

 

Good luck though.

[Fast_N_Curious]

43 minutes ago, Mick Naughty said:

So youre expecting that pump to actually move the metal at a sufficient rate and last any amount of time?

Liquid metal flow rates have already been tested - it possesses about twice the viscosity of water which at first seems like a large margin, however it is easily doable even with a conventional pump (see what I will be using down below). There is an abstract that discusses the fact that an electromagnetic MDF pump with only 8 feet pressure head was able to provide enough flow. That's actuallly less than my freezemod 800L-H pump... Besides, this cooler uses very little coolant, so the pump should not struggle as much as you might think.

 

Again this is only a test piece. The final loop will have a shielded MDF pump as well. These are ideal for liquid metal for a number of reasons:

1) No moving parts

2) Virtually silent 

3) Consumes less power than a conventional pump

4) Very easy to modulate speed

 

 

[For Science!]

A. I would hazard a guess that this radiator is aluminium, unless there is compelling reason otherwise

B. The thermal conductivity of the coolant is unlikely to make a large impact on the cooling potential of the radiator.

[Fast_N_Curious]

2 hours ago, AbydosOne said:

IIRC gallium will attack pretty much any metal routinely used in PCs. You're going to end up with a crumbly mess, more than likely.

Tantalum will not undergo LME under any circumstances.

1 hour ago, Levent said:

Not gonna work. You gonna kill pumps and corrode radiators before even getting half acceptable testing done. Gallium and mercury have been around for ages, there are reasons why they arent used as coolant in non nuclear applications. 

 

Good luck though.

I will be using a shielded electromagnetic pump. These are actually designed for liquid metal in some situations, so it's already suitable for this application. No moving parts, either.

Which is why the majority of my LM loop including the pipes will be made from tantalum:

This particular metal will form the backbone of the "solid metal" in this loop. Because it is only one of two metals that will not undergo LME at the hands of gallium. For that reason, I am using it. There are also a number of other alloys used by the chemical / industrial industries that are readily available, cost effective, won't undergo LME and won't degrade the metal in my loop significantly.

 

In this test piece however, you are mistaken. According to gamers nexus they have analyzed the thermal heat transfer properties once an oxide layer has been established ((of copper), and there is no loss of performance relative to test samples. It will however, degrade the liquid metal in my loop faster than most other solid metals. Surprisingly, the copper did not undergo very serious LME, because once you form an oxide layer on the walls of the solid metal, the LME process is nearly halted. If I do use copper in my loop it will be nickle plated. Which will also do ok in a loop like this. 

 

Guys the mini test here with the mini rad is just a prototype design and likely after I'm done it will have to be chucked. So don't think this is what the final product is going in.

 

[Fast_N_Curious]

38 minutes ago, For Science! said:

A. I would hazard a guess that this radiator is aluminium, unless there is compelling reason otherwise

B. The thermal conductivity of the coolant is unlikely to make a large impact on the cooling potential of the radiator.

It's copper. I have a few ideas about this:

 

a) I can get around this by using two independent loops. One hot, one "cold" (cold side will be either ambient or with chiller) with a heat exchanger in the middle. Another advantage of doing it this way is I won't need to fill this massive loop with $600 worth of gallium. all I have to do is tap into the 60 plate heat exchanger, which is a liquid-to-liquid design, much like a chiller setup:

 

To give you guys an idea of scale

count the number of pumps... lol

 

baby chiller and 60 plate HE. The heat exchanger should be ok with liquid metal, it's made from 316L stainless steel. 

 

 

 

 

 

 

[For Science!]

2 minutes ago, Storm-Chaser said:

It's copper. I have a few ideas about this:

 

a) I can get around this by using two independent loops. One hot, one "cold" (cold side will be either ambient or with chiller) with a heat exchanger in the middle. Another advantage of doing it this way is I won't need to fill this massive loop with $600 worth of gallium. all I have to do is tap into the 60 plate heat exchanger, which is a liquid-to-liquid design, much like a chiller setup:

 

To give you guys an idea of scale

 

 

 

 

 

 

I know, this is that project where you were adamant to use only a 120 mm radiator initially and got all worked up when people were suggesting you to use more radiator space. 

 

Radiators and chillers will fight each other after reaching ambient temperature, so you will have to control the chiller to not go below ambient, otherwise the radiators will heat up the components.

 

There is absolutely no reason to have a secondary passive liquid metal loop if the chilling is done by the chiler, you should just let the chiller be the primary loop and be done with it.

[For Science!]

Also, still found no evidence of the radiator being a full-copper radiator. AFAIK all OEM liquid coolers use copper cold plates, aluminium radiators, and propylene /ethylene glycol coolant to fight off the corrosions. So would be interesting to know how you are certain that the radiator is not aluminium.

[Fast_N_Curious]

1 hour ago, Enderman said:

Glad to see someone actually trying this out, it is a widely discussed idea from a long time ago.

 

I don't think it will work great, since the high conductivity is made almost irrelevant due to the fact that the heat is being moved by the physical movement of fluid by the pump, rather than thermal conduction through a solid, but it will still be a cool test

Yup good point. Hence the 60 plate heat exchanger and second loop, known as the cold side. The liquid to liquid heat exchanger I will be using has about 40x the surface area of 360mm radiator. So I am hoping for a larger delta T than anything air-liquid presently on the market. 

 

Example layouts 

 

[Fast_N_Curious]

Just now, For Science! said:

Also, still found no evidence of the radiator being a full-copper radiator. AFAIK all OEM liquid coolers use copper cold plates, aluminium radiators, and propylene /ethylene glycol coolant to fight off the corrosions. So would be interesting to know how you are certain that the radiator is not aluminium.

I also have no way to verify.

 

HP used copper radiators in their high end z820s is what I have heard. You won't find that on the internet, however. I've looked as well.

 

I guess we will know for certain if my radiator starts to crumble in my hands lol. 

 

No even if that is the case its not all for not. I can just buy a mini all copper 80mm rad and swap it out and continue testing....

 

 

 

[For Science!]

Just now, Storm-Chaser said:

No even if that is the case its not all for not. I can just buy a mini all copper 80mm rad and swap it out and continue testing....

If you are going ahead with it, I would highly advise you just take that route to begin with. It isn't worth the risk.

As a side note, you should be able to tell from the drained weight if its alu or copper. Alu rads are really light when they dont have liquid in them.

[Fast_N_Curious]

13 minutes ago, For Science! said:

I know, this is that project where you were adamant to use only a 120 mm radiator initially and got all worked up when people were suggesting you to use more radiator space. 

 

Radiators and chillers will fight each other after reaching ambient temperature, so you will have to control the chiller to not go below ambient, otherwise the radiators will heat up the components.

 

There is absolutely no reason to have a secondary passive liquid metal loop if the chilling is done by the chiler, you should just let the chiller be the primary loop and be done with it.

Problem with this idea is that the internal components of the chiller appear to be aluminum. This is going to be a problem unless I upgrade the internals of the chiller as well. 

 

And I was listening, even if I was a little strong willed, I wasn't completely disregarding your recommendations. And turns out I did eventually take you up on them, and for the reasons you all specified. Now the rig is at maximum radiator capacity lol...

1) one 360mm

2) one 240mm (passive)

3) one 120 x 80mm alphacool monsta (essentially two 120s stacked)

4) one 120mm

5) 2 freezemod 800L/H PWM pumps (after waterblock)

6) 1 barrow 450L/H pWM pump (before waterblock)

7 heatkiller full copper water block

 

I've had too much fun upgrading this rig. but it is less ugly since the last time you saw it. lol ill post some blacklight pics eventually.

[Fast_N_Curious]

4 minutes ago, For Science! said:

If you are going ahead with it, I would highly advise you just take that route to begin with. It isn't worth the risk.

As a side note, you should be able to tell from the drained weight if its alu or copper. Alu rads are really light when they dont have liquid in them.

Right, I was just thinking that's probably best. start with known copper radiator. 

[Fast_N_Curious]

4 hours ago, Storm-Chaser said:

I don't think it will work great, since the high conductivity is made almost irrelevant due to the fact that the heat is being moved by the physical movement of fluid by the pump, rather than thermal conduction through a solid, but it will still be a cool test

Can you elaborate on this. I already responded to this post but I want to be sure I'm correct in my interpretation. 

[fonzz1e]

are you doing a research paper?

 

• do u have 2 identical sys for this setup?
• 1 sys should be a base aircool or watercool
• have both of them running thermal stress test, prim95 small fft loop

 

i'm interested in the result after long term testing (min 6 months)

[Fast_N_Curious]

Just now, fonzz1e said:

are you doing a research paper?

 

• do u have 2 identical sys for this setup?
• 1 sys should be a base aircool or watercool
• have both of them running thermal stress test, prim95 small fft loop

 

i'm interested in the result after long term testing (min 6 months)

No research paper... more or less pure curiosity. 

 

I have a z820 dual processor rig. This rig has two E5 2696 v2 processors, 12 cores each and 24 threads. The front LGA 2011 socket tends to run a little cooler than the back one. But with some simple math, accurate results can still be interpreted with one running just coolant/water and the other running liquid metal. If for some reason this doesn't work I have another z820 system I can swap my other 2696 v2 in and then we will know for sure there will be accurate results.

 

So yes, a couple of the radiators I will leave stock. The other two will get liquid metal. 

 

 

[fonzz1e]

if your LM loop is setup, run the stress test & monitor cpu temps

compare to your normal setup stress test cpu temp

[Fast_N_Curious]

9 minutes ago, fonzz1e said:

if your LM loop is setup, run the stress test & monitor cpu temps

compare to your normal setup stress test cpu temp

I still need to get a lot of supplies to make this happen including the galinstan. ETA 3 months

[For Science!]

3 hours ago, Storm-Chaser said:

Can you elaborate on this. I already responded to this post but I want to be sure I'm correct in my interpretation. 

This stems from the fact that the bottleneck in conventional liquid cooling does not lie in the thermal conductivity of the coolant.

 

Energy is only removed from the system by transferring heat from an area of high surface area (I.e. the radiator) into the ambient air. This interface copper-air is incredibly inefficient and thus needs high surface area. Heat is moved from the chip to the area via the coolant but the coolant is not a static entity, it is being pumped around and so does not rely on the thermal conductivity to fulfill its purpose. In fact another big player is the specific heat capacity which dictates how much energy a coolant can absorb before it actually heats up.

 

in a chiller context where the copper-air bottleneck is alleviated, the next biggest bottlenecks tend to be the die-IHS or IHS-waterblock, so people move to direct die cooling (with a waterblock). Perhaps in this extreme scenario then coolant identity starts to make a difference.

 

As extra, PC coolants can vary wildly from almost water, to pure propylene/ethylene glycol. Theoretically glycols have a signinicantly lower thermal conductivity than water, but performance is the same, because it is not the limiting factor.

[Fast_N_Curious]

6 hours ago, For Science! said:

This stems from the fact that the bottleneck in conventional liquid cooling does not lie in the thermal conductivity of the coolant.

 

Energy is only removed from the system by transferring heat from an area of high surface area (I.e. the radiator) into the ambient air. This interface copper-air is incredibly inefficient and thus needs high surface area. Heat is moved from the chip to the area via the coolant but the coolant is not a static entity, it is being pumped around and so does not rely on the thermal conductivity to fulfill its purpose. In fact another big player is the specific heat capacity which dictates how much energy a coolant can absorb before it actually heats up.

 

in a chiller context where the copper-air bottleneck is alleviated, the next biggest bottlenecks tend to be the die-IHS or IHS-waterblock, so people move to direct die cooling (with a waterblock). Perhaps in this extreme scenario then coolant identity starts to make a difference.

 

As extra, PC coolants can vary wildly from almost water, to pure propylene/ethylene glycol. Theoretically glycols have a signinicantly lower thermal conductivity than water, but performance is the same, because it is not the limiting factor.

Am I correct in assuming the heat spreader and thermal paste are the top two major bottlenecks? So I probably need to go direct die if I want to make a decent result here. Makes sense.

 

Yes liquid metal has a lower specific heat but this is a two edged sword. In other words, it will heat up quicker than water but it will also lose heat faster, and require less energy.to do so.  This comes into play with pump speed. I will not be using a liquid to air system as you might see in a conventional loop. Liquid to liquid has many advantages over air cooling. Also note I can stick the entire heat exchanger in the freezer for benching. It's a sizeable heat exchanger made with 316L stainless steel so there should be no problem running liquid metal through it, if that's what I decide to do.