Moving a Graphics Card to the “Wrong” Spot Made It Run 14°C Cooler

October 12 (IT Home) — According to Tom’s Hardware, the YouTube hardware creator TrashBench has published the results of an experiment on cooling in unconventional PC layouts.

He used a standard-layout desktop with a three-fan RTX 3080 graphics card and a 360mm all-in-one liquid cooler as the baseline, locked the fan speeds across the whole machine, and looped 3DMark Time Spy and the built-in 7-Zip benchmark to maintain a constant heat load. He then began moving the graphics card and adjusting the airflow direction of the liquid cooler.

During the experiment he tried a variety of unconventional layouts, such as moving the graphics card to the bottom of the case or mounting it vertically at the side, testing different orientations and angles, and also testing the CPU liquid cooler’s fans blowing outwards and drawing inwards.

The results showed that with the graphics card in certain extreme, unconventional positions, the GPU hotspot temperature fell by as much as 19°C — but at the cost of higher CPU temperatures. In that configuration the GPU’s cooling improved to the greatest extent, while the CPU’s cooling conditions actually got worse.

In later refinements, TrashBench found a setup that balanced the CPU and GPU. His final approach was to mount the graphics card vertically, facing the case’s side intake fans directly, while keeping the CPU all-in-one cooler in the traditional outward-exhaust configuration. In that layout the graphics card’s hotspot temperature fell by 14°C and the CPU temperature also dropped by 3.3°C.

IT Home notes: the GPU hotspot temperature is the reading from the hottest sensor on the chip, not the average core temperature, and is the key indicator that triggers thermal throttling. A 14°C reduction is considerable for a modification that only changes the layout without replacing the cooler.

The experiment was based on a single platform and is not a laboratory-grade controlled study. To reproduce this kind of graphics-card repositioning test, a PCIe riser cable is essential; TrashBench used risers of several different lengths in the experiment. The experiment shows that case airflow and hardware layout still leave considerable room for improvement, but users will need to verify the actual results in different case designs for themselves.