Thirty-Three Kilograms Before the Motherboard Goes Under

Walked into the aquarium shop to pick up calcium supplement. The owner was behind the counter drilling holes in acrylic. Not a tank. A PC case.

Full mineral oil immersion build. Computer components completely submerged in clear dielectric fluid, running, operating, live—CPU at load pushing 87°C in a liquid that should conduct heat four times better than air but only if you solve the convection problem, which is apparently the entire hobby.

He said thermal stratification is harder than the hardware. Hot oil rises, cold oil sinks, and without forced circulation you get a 20°C gradient top to bottom. The CPU cooks while the GPU idles cold.

I’d been thinking about thermals as water flow—reef parameters at ±0.001 SG/5°C, protein skimmer bubble density, calcium reactor circulation rates. Never occurred to me you could drown a motherboard.

Bought a 40-litre acrylic tank on the way home. Same shop, different aisle. Tank rated for 1.025 specific gravity seawater, which means it’ll hold mineral oil at 0.84 SG without stress. Already know how to drill acrylic for bulkhead pass-throughs—same technique as the paludarium build.

Stripped an old PC to the frame. Removed every fan. Case fans, GPU fans, CPU cooler—all of it. Fans don’t move oil. The oil is the cooling system. Removing them is mandatory, not optional, and seeing the motherboard naked like that felt wrong in a way I didn’t expect.

CPU thermal paste: gone. Standard silicone compounds degrade in oil immersion. Replacement options are indium foil (which I don’t have), thermal epoxy (which is permanent), or just bare metal contact (which the forums say works but nobody recommends). Ordered indium foil. Two-day shipping. Build stalled until Friday.

Hard drives can’t be submerged. Breather holes equalize pressure as the platters spin—oil seeps in, catastrophic failure. SSDs are fine, hermetically sealed, but spinning rust stays outside the bath or doesn’t spin at all.

The tank weighs 2 kg empty. Forty litres of mineral oil at 0.84 g/mL: 33.6 kg. Add the hardware, add the oil displaced by components, total system mass around 40 kg. Put that on a desk without reinforcement and the desk fails, not the tank. Weight engineering isn’t optional.

Dielectric breakdown voltage: 10-15 kV/mm for pure mineral oil. That’s why you can submerge live circuits. But contamination—dust, moisture, metal particles—drops that number fast. Same purity discipline as reef trace-element dosing. Can’t let it get dirty.

Single-phase immersion: the oil never boils. Datacenter builds use two-phase fluorocarbons that boil at 50°C, carry heat away as vapor, condense in a heat exchanger. Those systems are closed-loop, pressure-controlled, expensive. Mineral oil stays liquid the whole time. Simpler. Heavier. Slower.

Forty litres of Nytro 10XN ordered. Food-grade white mineral oil, same viscosity class as transformer insulation fluid. Arrives Thursday. Tank sits empty on the workbench until then, acrylic walls spotless, waiting for a computer I haven’t finished disassembling.

Thermal stratification is the real problem. Oil convects too slowly for modern CPUs. Without an external pump and radiator, the system eventually reaches equilibrium at whatever temperature the CPU can sustain—probably 95°C, thermal throttling, exactly what the oil was supposed to prevent.

Ordered a 12V DC submersible pump (800 L/h, same model as the reef powerhead) and a 120mm aluminum radiator. Loop will pull hot oil from the top, push it through the radiator, return cold oil to the bottom. Forced convection. Active thermal management.

It’s just a reef sump with a CPU instead of coral.

The indium foil arrives Friday. The oil arrives Thursday. The tank is ready now.

Nothing’s powered yet but I keep staring at the motherboard, trying to imagine it breathing underwater.