On a more serious note, would it be possible to achieve said silence without too much compromise by some variant of water cooling, with a good sized reservoir? And use the heat to drive circulation rather than a pump?
Because there is no pump the water will flow much slower, therefore you need much bigger pipes. Think at least 3x the diameter you would think you need.
Also because you are depending on thermo siphon you have to ensure you have to ensure the water flow works with gravity. The heat source must be at bottom, and the radiator must be at the top. You don't get any flexibility go around something that might be in the way, the pipes must always be sloped in the correct direction.
If you are using anything other than water you need to pay attention to specific gravity (and how it changes with temperature), and viscosity. Probably something else that I'm not thinking of.
It probably can work. I have an antique tractor that uses such a system. A 2.2L, 16hp engine, has 24L of coolant, and the pipes between the engine and radiator are noticeably bigger than my truck with a 7.3L 250HP engine. Remember that this was designed to run at just under the boiling point of water (antifreeze is too modern). You probably want your computer considerably cooler, which implies even larger systems than a similar active one.
Yes, just use a flat (i.e., no high fins) passive cooler on the CPU (this won't fit with stock parts), lay the motherboard flat and submerge about 1cm deep in a refrigerant boiling in the range (max room temp 5K) and (max healthy cpu temp -20K). Flourinert and Pentane are suitable, the former is really expensive and the latter as flammable as gasoline.
If you replace the cooler with a 1mm copper plate and use a pump to archive a flowspeed across the cooler between 5 and 20m/s, you can get crazy overclocking. You might get subzero clocks without any active refrigeration.
The benefit is that all your ancillary components are cooled too, and you have superior performance with a pumnp, and even performance without one.
Ensure some radiator with gravity-backflow gets the vapor, e.g. just use a central heating one above and ensure liquid can't pool in the piping, and that that is not too narrow.
From what I recall mineral oil has long term effects on the HW, or in other words, I think I've seen some pictures of mineral oil cooled MBs with degraded/melted plastics. And the oil it's a logistic mess.
Instead the other day on Youtube I don't recall how I landed to a 3M infomercial about their Novec dielectric liquid used for immersion cooling. That stuff looked interesting, but they didn't spilled (pun ^__^;) details about cost or health issues tho.
The difference between Novec 649/1230 (they seem to be the same) and Perflourohexane/FC72/'Flourinert (well, 'the' classical kind)' is that the former is somewhat less inert and apparently slightly toxic to the central nervous system (high doses only), and the latter is sufficiently non-toxic to be used for allowing burn victims to breathe, as it shields their lungs from damage, and is able to provide enough oxygen by dissolving it to prevent suffocation. Rats don't drown in it if submerged and if it is sufficiently aerated to contain enough oxygen.
The other difference is that Novec has a Halon-style effect on fire, whereas Flourinert is at least not sold with that intention. Also FC72 decomposes slowly into a toxic, but afaik not dangerous to the elctonics, product, which has to be scrubbed.
Both of those have a vastly higher global warming effect than 'just' using Pentane, but that is rather flammable so one would probably want to ensure the oxygen concentration in the room is too low to result in a flammable mixture. This would probably mean that a human couldn't breath without external oxygen supply, but that should not be more expensive if it is a suitably low-physical-maintenance location, i.e. the systems are installed once and at most swapped out when obsolete or in case of component failure. A human could probably make due with an oxygen tube in his nose and some way to prevent the exhaled oxygen from sticking around in the (probably low-airflow) room.
The cost for FC72 is about 300$/kg, keep in mind the density is about 2.3 times that of water. Novec should be somewhat below, I think (otherwise there is little reason to offer it).
The difference is that mineral oil cools by convection and my solution cools by nucleate boiling. With my solution, and a flow or iirc about 20m/s, you could cool a delidded AMD EPYC 32 core at 600W. I.e. you are not limited by cooling, you are limited by how much current the LGA pins can handle before they loose stiffness (due to resistive heating in the pin and at the contact area) and then spiral out of control due to the weak contact pressure increasing resistance and leading to the ping disconnecting. This easily goes over to the others, there are documents from iirc Intel about how this happens/works, and how to prevent it.
I considered soldering the LGA pins to the CPU, but due to the craziness of this idea I haven't researched it further.
The thing is, that a dual-socket 32 core/socket EPYC clocked/overvolted for full stability and an overvolting-induced lifespan of 5 years, which does not have problems with power delivery on the way between the interposer (the thing the dies are mounted to and that connects them to each other and provides LGA pads) and the wall socket, would probably be the most affordable (TCO, i.e. including electricity and maintenance) system for non-distributed compiling of reasonably parallelizable software.
It would be 'the' ultimate workstation in the sub-$15k range, this would be the full system including the single-quantity amount of, in this case, Flourinert (which is not toxic; there was even research to fill the lungs with it for high-G environments, as it prevents lung collapse. There were issues circulating the liquid fast enough due to the much higher viscosity compared to air, so they dropped it.) and a custom-manufactured (from a local metal shop) containment case for the system and work from a plumber to provide suitable piping. The only non-included thing is the manufacturing and mounting of the nozzle that provides the required flow rate across the CPU to keep up with the heat.
Short answer: mineral oil does not provide suitable cooling for high-TDP processors.
It's possible using a low temperature phase-change system, but you need some absolutely immense radiators. The Calyos NSG-S0 chassis will passively cool a high-end gaming PC, but it weighs 22kg and costs €1000.
Probably not - you need to push water over the radiator blocks at a decent clip and I'm pretty sure you need pump. You could try a heatpipe into a big radiator that sits in a big tank of water, though.
Peltiers are only really worth it if you need sub-ambient temperatures (i.e. extreme overclocking). You then need to carry away the heat of the Peltier in addition to that of the processor, making expelling the heat even more difficult than it was before.
On a more serious note, would it be possible to achieve said silence without too much compromise by some variant of water cooling, with a good sized reservoir? And use the heat to drive circulation rather than a pump?