GPU Thermal Pads and VRAM Temperatures

Measure the pads that come out before buying replacements. Thermal pads bridge a specific gap, and a thicker pad is not a safer pad — too thick lifts the heatsink off the core, which trades a small memory improvement for a large one everywhere else.

A note on this guide: this rests on how graphics cards are assembled and how pad materials age, rather than on cards stripped here. See how we test.

The problem, as it appears

Monitoring software reports the memory temperature and it is dramatically higher than the core — twenty or thirty degrees higher is not unusual — and it looks like something is wrong.

Then on an older card the same figure creeps up over a couple of years with no change in what the machine is doing.

Why comparing it to the core temperature disappoints

Because they are cooled in completely different ways, so the gap between them is expected rather than diagnostic.

The core is a flat die pressed directly against the heatsink base through a very thin paste layer, with the fans immediately above it. That is about as good a thermal path as the card can offer.

Memory is a set of packages at a different height, bridged to the heatsink by a compressible pad spanning a physical gap. A pad is a far worse conductor than a paste joint under pressure, and on some designs the memory sits on the back of the board with almost no cooling arrangement at all.

So memory reading much hotter than the core is normal. What matters is not the gap between them — it is whether your own card's memory temperature has changed.

What actually matters

Change over time, not the absolute figure

Manufacturers publish memory limits, and cards protect themselves by throttling as they approach them. An isolated reading is therefore hard to act on.

A reading that is clearly higher than it was a year ago, under the same load in the same case, is the signal worth following. That is either airflow or pads, in that order of likelihood.

Pads harden, and the decline is gradual

Thermal pads are soft so they can conform to two surfaces that are not parallel and not touching. Over years they dry out and stiffen, and a stiffer pad contacts at fewer points.

Nothing visibly changes. The card looks identical, nothing is loose, and the memory simply runs a little hotter each year. This is the mechanism behind most "my old card runs hot now" cases that are not dust.

Thickness is a measurement, not a preference

This is where people damage cards, so it is worth being explicit.

A pad has to fill its gap exactly. Too thin leaves an air gap and conducts almost nothing. Too thick holds the heatsink away from the core — and the core is the part with the good thermal path you have just ruined.

So the rule is to measure what comes out, or consult a reliable teardown for that specific card, and match it. Different gaps on the same card frequently need different thicknesses.

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.

The broader question of which material belongs where is covered in thermal paste vs thermal pads.

Do the free thing first

Memory temperature responds to the air the card receives. A restricted intake, a dust-loaded filter, or a card pressed close to a glass side panel raises everything.

Fixing airflow costs nothing and it is the correct first move, exactly as it is with the rest of the symptoms in what goes wrong with a gaming PC after year one.

Opening the card is a real decision

Repadding means removing the cooler, which usually means removing the backplate, disconnecting fan headers and handling a bare board.

It commonly voids the warranty, and while the cooler is off the paste on the core must be replaced too — so the job is never only pads.

A heavy card has a second problem

Cards heavy enough to have substantial pad-cooled memory are usually heavy enough to sag, and sag puts load on the slot and can distort the contact the whole cooling assembly depends on.

That is a separate issue with its own fix — why heavy graphics cards sag.

The mistake to avoid

Buying one thickness of pad for the whole card. It is the convenient assumption and it is how people end up with worse temperatures than they started with — a single sheet cut for every location will be wrong for most of them, and wrong in the direction that lifts the heatsink is the expensive one.

The second mistake is treating a high memory reading as a fault without a baseline. Memory runs hot by design because it is cooled through a pad rather than a paste joint, and a card that has always read that way and is not throttling is a card working as intended.