How we clean and test every card

The run

We benchmark each card dusty (as it arrived), then tear it down, clean the heatsink and fans, replace the thermal paste, reassemble, and benchmark again, now clean. Same machine, same driver, same settings, back to back. Nothing about the test changes between the two runs except the dust.

How we clean a card

Every card is fully serviced before its second benchmark. It's the same process you see in the videos:

1. Full teardown, down to the bare board: cooler, shroud, fans, and backplate all come off.
2. Soapy distilled-water wash. Yes, water. A graphics card is a circuit board, and a circuit board doesn't mind water, what it minds is being powered on while wet. Distilled water and a little mild dish soap lift the baked-in dust and grime that compressed air can't touch.
3. Distilled-water rinse to carry off the soap. (Tap water leaves mineral spots; distilled doesn't.)
4. A 99% isopropyl alcohol rinse on the board, which chases out any water hiding in the tight spots, under connectors and in the fin stack, so nothing stays damp where you can't see it.
5. A full dry. This is the part people worry about and the part that matters most. The board goes nowhere near power until it is completely dry. We don't rush it, because under-drying is the one way a wet clean goes wrong.
6. Fresh thermal paste, new pads where needed, and careful reassembly.

Then it's benched again, identical settings to the before run. That's the before-and-after you see.

Is washing a GPU actually safe?Done correctly, yes. The danger everyone pictures is water meeting electricity, and there's none of it here: the card is unplugged, in pieces, and bone dry long before it's ever powered again. Boards die from being switched on wet, and that's the exact step we refuse to hurry.

What we measure

• Load & idle temperatures: sampled once a second through a fixed load, so you see the whole curve, not a single number.
• Throttling: read straight from the GPU's own throttle flags. A dusty card pinned against its thermal limit is the story we're looking for.
• Performance: a cross-generation graphics score plus an FP32 compute figure, so a clean card's sustained clocks show up as real numbers.
• Power, clocks, and fan: to round out why the temperatures moved the way they did.

These are standardized synthetic measurements for fair comparison, not a promise of any specific game's frame rate.

Honesty is the point

We publish every result, including the boring ones. Plenty of cards are only mildly dusty and barely move after a clean, and when that happens, we say so. The value here is trustworthy data, not a highlight reel.