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 co

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.