Cooling & Airflow
Repasting a GPU, and When Not To
It voids warranties, it is fiddlier than a CPU, and the pads matter more than the paste. When it is genuinely the fix, and the cheaper things to try first.
Contents
Try three cheaper things first, and check the warranty before the screwdriver.
Dust, case airflow and a fan curve account for most GPUs that run hot, and all three are free and reversible. Repasting is the last step, not the first — it frequently voids a warranty, it is fiddlier than a CPU, and on many cards the thermal pads matter more than the paste does.
A note on this guide: this rests on how GPU coolers transfer heat and on manufacturers' published warranty and handling guidance, not on cards stripped here. Pad thicknesses and screw patterns are model-specific — find a teardown for your exact card. See how we test.
The problem, as it appears
A card that used to hold its clocks and now throttles. Fans louder and earlier than they used to be. Sometimes a memory-related instability in demanding titles rather than a temperature reading at all.
The internet's answer is to repaste it, and that is sometimes right and frequently premature.
Why going straight to paste disappoints
Three reasons.
It is usually not the interface. A card that has been in service for a few years is far more likely to be clogged than dried out. Dust in the fin stack is the common cause and it is removable without taking anything apart — cleaning a GPU without removing the cooler.
The pads are doing work the paste is not. The paste sits between the die and the cooler. The pads carry heat from memory and power delivery, and on plenty of cards it is memory temperature that limits things. Disturbing pads without replacing them correctly makes a card worse — a pad that no longer makes contact transfers nothing at all.
It may not be the card. If the whole case is hot and the air leaving it is hot, the card is rejecting heat into an oven. That is an airflow problem and no amount of paste addresses it.
This page contains affiliate links. If you buy through them, Ugly Fixes may earn a commission at no extra cost to you. Products are selected independently.
What actually matters
1. Establishing that it is the interface
The distinguishing test: core hot, heatsink comparatively cool means heat is not getting out of the die — an interface problem. Everything hot, including the exhaust air means heat is getting out of the die and not out of the case.
Record temperatures before you start. Without a baseline you cannot tell afterwards whether the work achieved anything, and "it feels better" is not a measurement.
2. The warranty, before anything else
Many cards carry a warranty seal over a screw. Check the manufacturer's stated policy — some are relaxed, some are absolute.
A card in warranty with a genuine thermal fault is a support case. Opening it converts their problem into yours.
3. Pads, sized properly
If you are going in, plan for pads. They are model-specific in thickness, and guessing is worse than leaving the originals.
Find a teardown for your exact card, note the thicknesses per area, and buy accordingly. Compressing a too-thick pad can hold the cooler off the die; a too-thin one leaves an air gap.
Thermal pad sheet
Thickness is model-specific and is the detail that decides whether this helps. Find a teardown for your exact card before ordering — the wrong thickness is worse than the pads you removed.
See current options(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
4. Not overtightening on the way back
The most common way to turn a maintenance job into a dead card.
Diagonal pattern, in stages, gently. The screws locate the cooler; they are not there to squeeze paste out. Overtightening warps the cooler or cracks a die corner.
Precision screwdriver set
The correct size, seated fully. Stripped GPU backplate screws are a genuinely bad afternoon and are caused by a driver that almost fits.
See current options(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
The options, and who each suits
Clean it, first
Free, reversible, and it is the answer more often than repasting is. Compressed air or an electric duster through the fin stack, fans held so they cannot spin up.
Electric air duster
No propellant, no liquid, and it does not run out mid-job. Hold the fans still while you use it — spinning them backwards can damage them.
See current options(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
Anti-static PC cleaning brush kit
For the fin stack edges and the shroud, where air alone leaves a compacted mat behind.
See current options(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
Fix the case, second
If the exhaust air is hot, the card is not the problem. Fan curve, intake path, and where the card sits relative to the front intakes.
Repaste, third, and only with a reason
For a card several years old, out of warranty, with a clear core-hot / heatsink-cool signature. Expect a meaningful improvement in that specific case and very little otherwise.
Thermal paste kit
A good conventional paste, not liquid metal — that is electrically conductive, attacks aluminium, and a spill on a card is terminal. Enough for several attempts is worth having.
See current options(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
ARCTIC MX-6 thermal paste (4 g)
ARCTIC
Electrically insulating, and ARCTIC publishes the figures rather than the adjective: 1.8 x 10^12 ohm-cm volume resistivity, 7.5 kV/mm breakdown voltage. That is what a bare die surrounded by exposed components wants. 4 g is several applications.
Check price and availability(opens in a new tab, affiliate link)Affiliate link. Price shown on the merchant's site.
The general argument about when paste is actually finished is in when to replace thermal paste, and the pad-versus-paste distinction is in thermal paste vs thermal pads.
The mistake to avoid
Opening a card that is still under warranty. The fault you are trying to fix may be covered, and after the seal is broken it certainly is not.
The second mistake is repasting and ignoring the pads. On a card where memory temperature is the actual limit, a perfect paste job changes almost nothing — and if the old pads were disturbed getting there, the card can come back worse than it went in. GPU thermal pads and VRAM temperatures is the page that argues which of the two your card is limited by.
Questions people actually ask
Does repasting a GPU void the warranty?
What should I try first?
How do I know the paste is actually the problem?
Are the thermal pads more important than the paste?
Can I use liquid metal on a GPU?
How much of a temperature drop should I expect?
What most often goes wrong?
Question this page did not answer? Ask it.