Scrap AC Radiator Cools RTX 3080 to 28C: 4-kW Bench Test

By Grace Kim

A discarded 4-kW air conditioner radiator held an Nvidia RTX 3080 at 28 degrees Celsius in a bench test published this week, barely above ambient room temperature and down from 65 degrees on the card's stock cooler.

The builder, an Australian hardware experimenter who publishes under the name TrashBench, salvaged the outdoor radiator-fan unit from a split air-conditioning system. The unit came without refrigerant gas. He stripped it to its essential parts: the radiator, the fan, and the pump. He discarded the remaining internals and mounted the assembly on casters for mobility. A standard reservoir went into the main enclosure.

How did he connect the radiator to the PC?

The plumbing work was improvisational. TrashBench converted a T-joint into a Y-joint with heat and force. He then designed and 3D-printed mounting brackets to attach the various cold plates to the hardware under test. The video records the cost of the build in blood. At the 02:49 mark, his knuckles had paid what the source calls the requisite toll to the machine god.

He opened the presentation with an impression of Crocodile Dundee, then moved straight to instrumentation and benchmarks.

What did the numbers show?

Two graphics cards went onto the loop. The first was a modern RTX 5060, an efficient part that ran at 52 degrees Celsius on its standard cooler. The 4-kW loop brought that figure to 33 degrees, roughly the temperature of lukewarm water, a 16-degree drop.

With thermal headroom to spare, TrashBench pushed the RTX 5060's boost clock from 2902 MHz to 3300 MHz. That is a 14 percent clock increase. Average frame rates rose only 3 percent. The result confirms how tightly Nvidia has tuned its power and performance curve since the RTX 2000 series: extra clocks buy almost nothing when the voltage-frequency relationship is already dialed in.

The RTX 3080 presented a harder thermal load. This power-hungry card lost decisively to the Fujitsu-derived radiator. Its temperature fell from 65 degrees Celsius on the stock cooler to 28 degrees on the loop. Overclocking this card also returned roughly 3 percent in frame rate, a marginal gain for the effort involved.

Could it cool a CPU as well?

The final test combined a graphics and processor load. The processor was arguably the worst consumer chip available for the purpose: the Intel Core i9-14900K, a fast and notoriously hot part. The loop dropped CPU temperatures by 18 degrees to 76 degrees Celsius, a strong result for this silicon outside liquid-nitrogen setups.

The graphics card sensor did not move at all during the combined run. That reading suggests the radiator retained far more capacity than the available hardware could generate. Four kilowatts of rated heat dissipation vastly exceeds the roughly 800 watts a high-end CPU and GPU pair can produce at full load.

What does the experiment prove?

The test is a data point on how oversized commercial cooling hardware compares with purpose-built PC components. A scrap split-system condenser, stripped of its refrigerant circuit and repurposed as a passive water-loop heat exchanger, outperformed stock air cooling on every component measured, with double-digit temperature drops across the board.

The overclocking results cut the other way. Both cards gained only 3 percent in frame rate despite substantial clock increases, which indicates that consumer GPU performance is limited by architecture and voltage curves, not by cooling, once temperatures are under control.

TrashBench's earlier experiment, which placed a graphics card in an unconventional position for better cooling results, set the stage for this larger build. Future projects along these lines will likely target even higher heat loads, and the spare capacity measured in the combined CPU and GPU run leaves room for a much more power-hungry system before the 4-kW radiator reaches its limit.