Extreme Pi Overclocking With Mineral Oil

Liquid cooling is a popular way to get a bit of extra performance out of your computer. Usually this is done in desktops, where a special heat sink with copper tubing is glued to the CPU, and the copper tubes are plumbed to a radiator. If you want dive deeper into the world of liquid cooling, you can alternatively submerge your entire computer in a bath of mineral oil like [Timm] has done.

The computer in question here is a Raspberry Pi, and it’s being housed in a purpose-built laser cut acrylic case full of mineral oil. As a SoC, it’s easier to submerge the entire computer than it is to get a tiny liquid-cooled heat sink for the processor. While we’ve seen other builds like this before, [Timm] has taken a different approach to accessing the GPIO, USB, and other connectors through the oil bath. The ports are desoldered from the board and a purpose-built header is soldered on. From there, the wires can be routed out of the liquid and sealed off.

One other detail used here that  we haven’t seen in builds like this before was the practice of “rounding” the flat ribbon cable typically used for GPIO. Back in the days of IDE cables, it was common to cut the individual wires apart and re-bundle them into a cylindrical shape. Now that SATA is more popular this practice has been largely forgotten, but in this build [Timm] uses it to improve the mineral oil circulation and make the build easier to manage.

31 thoughts on “Extreme Pi Overclocking With Mineral Oil

  1. I’m not familiar with any desktop liquid cooling system that involves gluing copper tubes to a CPU. The closest I can think of is CPU coolers that use copper phase change heatpipes, which would be fluid cooling as the working fluid operates in both a gas and a liquid state. These are rarely, if ever glued to CPUs.

    Desktop liquid cooling usually uses something called a water block, this is a heatsink/radiator style thing (a chunk of metal with fins or channels for water to pass between or through) that is part of a plumbing “loop” that includes a pump, reservoir and radiator. The plumbing is only sometimes copper tubing. The water block is usually connected to the heat spreader of CPU with thermal paste, a thermal pad or “liquid metal” that has little to no adhesive or “glue” like properties. Instead the connection is maintained with a bracket and (often) spring system that imposes constant pressure between the CPU and water block.

    1. I have in my lap a package of “Arctic Silver” two-part, silver-filled epoxy which at least some people claim is the best thermal transfer compound there is. I make no such claims, as I haven’t actually tried the stuff, and yes, glue seems inconvenient but only in a different way than zinc-oxide-filled grease. The grease has its own set of issues, dunno about the epoxy but in other things I’d guess the lack of flexibility could work for or against you.
      I know for a fact that thermal compound creep-out and hardening nearly ruined two of my Intel NUCs, had to rebuild them with new grease after cleaning off the old stuff which was cracked and had voids. I’ve seen the grease cause other failures in machines when it got into connectors and insulated something (possible with crappy connectors, which abound in nature).
      Horses for courses I suppose, now if we really understood the courses…which I’m convinced we don’t. Thermal tempcos make some flexibility a good thing, but creeping is bad too.

      1. I didn’t know there was thermal adhesive marketed to enthusiasts by such a mainstream company. If you want the non creep out of epoxy by the serviceability of paste. You might want to look into thermal pads, I recently saw a video about a new product that apparently performs very well and is reusable.

        Creep out can also usually be avoided with proper application. Something OEM’s sometimes gave a reputation for not being good at.

        I’m amused by your comment about having it in your lap. This suggests that there was an odd coincidence, that you tend to keep thermal epoxy on in your lap, that you put it in your lap for the purpose of writing this comment or that you aren’t being honest. All four are amusing.

      2. The Arctic Silver epoxy is the best thermal transfer *epoxy* there is. Compared to actual thermal transfer compound it’s quite crap. Even Arctic Silver’s own normal thermal compound beats it and it hasn’t taken the crown for years. That said for more extreme applications people these days look towards a liquid metal allow made from gallium and indium.

      3. With all thermal interface materials you have to remember, they are orders of magnitudes worse at conducting heat than any metals. However, they are orders of magnitude better than air. Never ever use them to fill substantial gaps, get the metal as close to fitting as possible.

        Comments such as saying aluminum is a much worse conductor than copper, then going on to say you’re using the best thermal compound tend to give the impression that the compound is way better than aluminum can cope with, nope.

        Why copper footed sinks tend to do a lot better on hot CPUs is more a heatsink design problem than anything, it’s that the heat needs to go about an inch sideways and upwards before it gets into any airflow worth speaking of to dissipate it. Fans have huge dead spots in the middle and make no positive pressure worth speaking off, so the sink is only really dissipating from a shallow inverse cone near the top. It’s only this path length that makes copper seem that much better, change the design of your cooling, like blowing the middle of the sink with a blower than can develop a little positive pressure, and you’ll see coppers advantage is very slight.

        Heatpipes hopefully move heat away at near sonic velocity, and copper vs aluminum shouldn’t really see much difference, apart from copper tube is easier to get into the right shape and can be soldered easier. Still your choke point is the thermal interface at either end, CPU to pipe, pipe to fin.

        1. The reason people were talking about aluminum being a bad material is in the context of liquid metal TIM. It costs more and is messy to lay down, so it only makes sense if you’re already at the limit of what better coolers can do.

      4. I’m not claiming Arctic Silver is the best ever, but as much as I’ve used for CPU heatsinks in my life it works. After a few years it’ll “dehydrate” into a gel-like consistency, but hasn’t creeped out of the socket or turned into solid powder.
        It stood up to the overly hot AMD K6 and K7 processors I had as a kid, which is why I still like it.

        Also, Steven13 is correct, thermal pads are good performers as well. Depends on what you need and how much you want to spend.

  2. I’m wondering if oil is the best coolant choice. You want something that absorbs and gives off heat as quickly as possible. Does mineral oil have that property? I know its used in transformers but my understanding is that it is just to couple the heat from the core to the case fins and a transformer is not as spiky in its heat excursions. How does it compare to coolants like fluorinert (3M coolant used in some Cray flood liquid cooling systems).

    1. Not sure. Mineral oil is nice because it’s cheap and you can get it anywhere.

      The big issue is eating things; you have to be careful about what you put in there. The acrylic, silicone sealant, PCB, and most of the electronics should be inert to mineral oil. I’ve read about some concerns with electrolytic capacitors, but it doesn’t seem to be an issue in practice. The PVC coating on wires may get eaten over the course of a few years.

        1. Even if the case is fully sealed against dust or other particle ingress, ionic diffusion from any exposed metals will eventually lead the DI water to provide sufficient conductivity to cause direct shorts or dendrite growth.

    2. The idea is just to have a very large heatsink. You give the heat a very large volume to dissipate into and can either let it passively circulate via convection or actively circulate cooler oil over the heat generator. Trouble is, you need a very large area to radiate that heat back out or what you end up doing is very slowly heating up all of your liquid. He tested it for an hour.
      I’d have run that for a day, at least, at 100% duty cycle. When we did this to naked P90s back in the day, we used relatively shallow containers with a large thermal interface area for the air (large steam cabinet pans, actually) to allow the oil to radiate heat and it still saturated after a day or so, so we started using drinking fountain radiator assemblies.
      P90 at a half-ghz was about the best we could do because at that frequency our oil became capacitively inductive and started shorting the gaps in the CPU socket.
      Might try a similar experiment soon using distilled water on a conformally-coated board. Will post to hackaday if we do. ;)

  3. And what does all this accomplish? Nowhere do I see the bottom line of what clock rate can be achieved before and after all this. What does this buy us? Like most overclocking, it is done as an end in itself. No, I don’t want to watch the whole video just to get those two numbers (or find out we never get them at all). There is an old saying about not seeing the forest for the trees.

Leave a Reply to BNBNCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.