Liquid metal for PC cooling – asset or liability?

[nexus]

Hi all,

Just finished up a project where I learned quite a bit about thermal interface materials (TIMs) as used in modern electronics. It involves voiding the warranty on a brand new computer of course.

This includes studying the use of liquid metal (room temperature gallium based alloys), which is becoming increasingly popular.

Wrote up some of my findings here, maybe a few folks will find it interesting:

https://neonkev.com/2025/01/20/liquid-metal-for-pc-cooling-asset-or-liability/

Cheers

[Conrad Hoffman]

Wow, that was quite the effort! I'll offer a couple thoughts from some experiments done many years ago. The flatness of the two surfaces, CPU and heatsink, is critical. Ideally, they will be polished and mate perfectly over the whole area. I suppose one could check them with machinist's scraping blue. I used to lap parts, but I've never checked or lapped anything like a CPU. I have to believe they're good, but I suspect a lot of heatsinks aren't. It isn't possible to squeeze thermal paste out of a large area at normal pressures, so it has to be applied far thinner than most people do. If you say, "OMG, is that enough?" it's probably only a little too much. Finally, to help with the layer thickness, I wonder how a phase change compound would work? It's basically a filled wax that goes liquid when it warms up. And electrically nonconductive. See Aavid or Boyd Ultrastick. Once you get the layer thickness down to near zero, the conductivity of the compound becomes less important. Now, if a perfect interface can't be had, the liquid metal starts to get attractive.

[moffy]

At Ferranti in the 80s they did an experiment on the best thermal interface and as Conrad mentioned, perfectly flat surfaces with nothing in between came out on top.

[NiHaoMike]

If the bottom cover is metal, see how much adding thermal pads between the heatsinks and cover helps.

[nexus]

Wow, that was quite the effort! I'll offer a couple thoughts from some experiments done many years ago. The flatness of the two surfaces, CPU and heatsink, is critical. Ideally, they will be polished and mate perfectly over the whole area. I suppose one could check them with machinist's scraping blue. I used to lap parts, but I've never checked or lapped anything like a CPU. I have to believe they're good, but I suspect a lot of heatsinks aren't. It isn't possible to squeeze thermal paste out of a large area at normal pressures, so it has to be applied far thinner than most people do. If you say, "OMG, is that enough?" it's probably only a little too much. Finally, to help with the layer thickness, I wonder how a phase change compound would work? It's basically a filled wax that goes liquid when it warms up. And electrically nonconductive. See Aavid or Boyd Ultrastick. Once you get the layer thickness down to near zero, the conductivity of the compound becomes less important. Now, if a perfect interface can't be had, the liquid metal starts to get attractive.

Thank you!

I know that lapping CPUs is (or used) to be a popular thermal trick. Most CPUs are lidded, so the die wasn't directly exposed. Instead, one would lap the metal IHS (integrated heat spreader) for better contact with the final heatsink.

With direct die cooling, I'm not sure how flat the die surface is or if the surface finish is extremely fine. The heatsinks I've seen were not precision machined either, with milling marks being visible in most cases. Grinding would be optimal for finish and flatness. Add in uneven mounting pressure of most heatsink fixturing, and it only exacerbates the issue of uneven contact.

I haven't stumbled across Aavid or Boyd Ultrastick, but those appear to be some interesting phase change solutions. The most popular phase change material I did read about is Honeywell PTM7950. It appears to be currently popular with the PC modding community and also offered in OEM gaming laptop applications (for mid range hardware, the highest end stuff is still using liquid metal).

At Ferranti in the 80s they did an experiment on the best thermal interface and as Conrad mentioned, perfectly flat surfaces with nothing in between came out on top.

Agreed, perfectly flat surfaces are the ideal scenario (short of the heatsink being a homogenous part of the die).
Any materials between the heat source and heat sink are only adding thermal impedance.

If the bottom cover is metal, see how much adding thermal pads between the heatsinks and cover helps.

That is good idea. Unfortunately the only plastic part on the laptop is the bottom cover
It is metallized (probably for EMI) but otherwise a fairly thin plastic.

For air cooling desktop hardware in a laptop footprint, it probably wont get much better.

I did notice that some laptops like the Asus ROG have much longer fin stacks for the fans (and an extra small fan right next to the GPU)
However, in all the reviews I can find, all of these Raptor lake gaming laptops run right at the thermal limit.

[hanakp]

I experimented with galinstan TIM (Thermal Grizzly Conductonaut) about 6 years ago:

1. For air coolers, the difference was negligible, even heatpipe-based ones conduct the heat away so poorly TIM doesn't help much. This is why @nexus' project improved his laptop's temperatures only minimally.

2. Difference was also negligible for CPUs with heat spreaders, even when fitted with water cooling block. Again, the spreaders conduct heat too poorly.

3. The only useful difference occured when a water block was mounted directly on GPU chip. Full-load temperatures dropped about 8 degrees (GeForce GTX 1070 Ti) compared to good non-metal TIM (Thermal Grizzly Kryonaut).

Be warned galinstan TIMs have one downisde: they will eat into some metals, including copper. After some time, you may not be able to separate the cooler from CPU. And if you do, you will have to re-polish and/or re-plate their surfaces. Fortunately, galinstan doesn't attack silicon, so you can safely put it on a GPU chip (as long as it doesn't short something). TBH I haven't used it since then, it's not worth the hassle.

[Haenk]

I fully agree with that conclusion.
While liquid metal has advantages, the risks are IMHO much too high to damage parts.
Funny enough, I recently found a CoolLaboratory "LIQUID MetalPad" new-in-box in one of my drawers. Must have been there for almost 20 years; and I honestly can't even remember using it. But likely I didn't, as my experiments with liquid metal at that time weren't very promising. Quite a mess to clean up, actually.