This is one enthusiast's test on one machine, not a lab study. Even so, it shows a trade-off that matters to anyone who builds or tunes a PC: where you put parts decides where the air goes.
What was tested
TrashBench's plan was to build an RTX 3080 desktop, test it normally, then start moving parts around. The baseline was conventional: a CPU cooled by a 360mm AIO and a triple-fan RTX 3080 in the usual PCIe slot.
He tested that baseline under constant load by looping 3DMark Time Spy and 7-Zip's built-in benchmark. The system fans had their RPMs locked for consistency. Locking fan speeds is a sensible choice. Without it, the motherboard's fan curves would respond to each layout and muddy the comparison.
Tom's Hardware says the changes covered several variables:
- The GPU was tried at the bottom of the case, and vertically to the side. It faced different directions and sat on either side of the three side fans.
- The CPU AIO radiator fans were tested pushing air out and pulling air in.
- PCIe riser cables, in several lengths, made the GPU repositioning possible.
The headline numbers
The coverage gives three figures, and they describe different things:
| Figure | What it describes |
|---|---|
| 19°C | Largest GPU hotspot reduction, in a setup that hurt CPU temperatures |
| 14°C | GPU hotspot reduction in the balanced final layout |
| 3.3°C | CPU temperature reduction in the balanced final layout |
The video's title says "17°C Cooler", while the article's headline says 19°C. Tom's Hardware doesn't explain the gap, so treat both as the creator's and the outlet's stated figures. Don't read either as the final configuration's result.
The article also doesn't say which exact placement produced the 19°C result. It doesn't give the CPU penalty for that layout either.
The balanced layout
After testing, TrashBench settled on a compromise. Some configurations mostly benefited the GPU or the CPU, but the sweet spot cooled both and was still worthwhile. Tom's Hardware describes it as the card mounted vertically directly against the side intake fans, with the AIO radiator in a traditional exhaust arrangement.
That fits a general airflow principle. A GPU dumps a lot of heat into the case, and the card sitting directly in the path of fresh intake air gets the coolest supply. This is general industry knowledge, not something the report states. Pulling cool air through a radiator first, or hot air off a card, can raise the other component's temperature. That would explain why the best GPU result hurt the CPU. Tom's Hardware doesn't say that this is the cause.
Why the hotspot number matters
GPU hotspot is the hottest sensor reading across the die, not the average core temperature. Hotspot readings tend to be the ones that trigger thermal throttling. A drop of 14°C or more there is large for a layout change with no new cooler. The article doesn't report any performance, noise, power or stability results for the repositioned builds, so we can't say whether the cooler card also ran faster.
What the test doesn't show
- Limited data. The report gives no ambient temperature, absolute CPU or GPU temperatures, run duration, CPU model, case model, fan RPM values or run-to-run variation. Only temperature deltas are reported.
- One machine. Results will differ with other cases, GPUs, fan layouts and spacing. Nothing here shows that GPUs are generally "installed wrong."
- No independent corroboration. I found no second source testing this specific experiment. My search turned up mostly other TrashBench stunts, such as sub-zero coolers, a server AIO strapped to an RTX 3080 and a water-cooled laptop.
- Riser caveats. The coverage offers no riser compatibility guidance and no signal-integrity measurements. A long or low-quality riser can in principle cause PCIe link problems. That is general knowledge, not a finding from this test.
If you want to experiment
The report calls a PCIe riser cable essential for this kind of GPU repositioning. A cautious approach, based on common testing practice rather than the source:
- Record a baseline with your current layout, under a repeatable load and with fixed fan speeds.
- Change one variable at a time, such as radiator direction or GPU position.
- Log both CPU and GPU hotspot temperatures, since a gain on one side can cost the other.
- Confirm the GPU still links at its expected PCIe speed after adding a riser.
The takeaway
The conventional layout is a reasonable default, but it isn't necessarily the best one for every case. Tom's Hardware says TrashBench found a balanced arrangement that cooled both the CPU and the GPU. Treat the numbers as a lead worth testing on your own hardware, not as a universal fix.