The most repeated advice about how to apply thermal paste is also the least useful. Dot versus line versus X versus a full spread with a spatula gets argued over in every build thread, complete with diagrams and strong opinions, and the shape you squeeze onto the heat spreader is the smallest variable in the whole job.

Three things decide whether your chip runs cool. How clean the two mating surfaces are, how much compound you use, and how evenly the cooler clamps down. Nail those and a dot, a line and an X finish within a rounding error of each other. Miss them and no pattern rescues you.

So this guide spends its time on the boring parts. The amount, the wipe-down, the screw order, and the temperature check that tells you whether any of it worked.

How Much Thermal Paste to Use for Your CPU

A new cooler on the desk, a CPU already seated, and a syringe in your hand with no useful markings on it. Every guide you've read says something different about how much to squeeze out.

Here's the rule that survives contact with real hardware. You want the thinnest continuous film that still wets both surfaces, and the clamping force of the cooler, not a spatula, is what produces it.

Paste does one job. It displaces air from the microscopic pits and ridges on the heat spreader and the cooler base, because trapped air conducts heat badly and bare metal contact never covers the full area. Thermal interface research measured resistance across bond lines from 5 to 35 micrometers and found it rising approximately linearly as the layer got thicker, while a separate thermal-interface study tested thicknesses of 10, 30, 50 and 70 micrometers (published bond-line thickness research). Thickness is the variable you're fighting. You want it small and even, not generous.

What to squeeze, by package type

Match the pattern to the package and keep the quantity modest. Read the last column more carefully than the second one, because that's where the instruction lives.

CPU or packagePatternAmountNote
Smaller standard heat spreaderOne central dotRice grain to pea, about 3 to 5 mmCooler pressure spreads it to the edges
Large mainstream package (AM4, AM5, LGA1700, LGA1851)Central dot plus four small dotsModest, never a thick layerFollow the cooler maker's pattern
Elongated or unusually large packageX or multi-dot patternEnough to reach every edge without squeeze-outCheck edge coverage after mounting
High-power or server processorManufacturer's patternSet by the manufacturerMounting hardware and torque come from the maker

A single centered deposit is the standard approach for a conventional desktop heat spreader, because mounting pressure distributes it across the contact area without trapping air. You're not painting a surface. You're supplying material for the cooler to squash into the voids.

Why a bigger blob doesn't help

Excess compound doesn't build a better bridge, it builds a longer one. The heat now has to cross a thicker layer of material that conducts worse than copper, and the surplus has nowhere to go except out past the edge of the heat spreader and onto the board, where it collects dust and makes the next repaste annoying.

Pattern differences are small once coverage is complete. A controlled test on a 200 W Intel Core i7-14700K measured sustained-load temperature deltas of 50.6 °C for a full spread, 51.4 °C for a central dot and 52.0 °C for an X pattern, a maximum gap of 1.4 °C (Termopasty's comparison of application methods). That's the size of the argument everyone is having.

The rule I use: if you can see a raised cushion of paste after the cooler is seated, there's too much.

Spend the saved effort on the mount. A cooler tightened unevenly costs more than the entire spread-versus-dot debate.

Clean Both Surfaces Before Any Paste Goes On

Most disappointing repastes are decided before the tube is opened. Old compound, skin oil, dust and lint sit between two surfaces that need to touch, and no dispensing pattern compensates for a contaminated interface.

how-to-apply-thermal-paste-surface-cleaning.jpg

Intel's own procedure is blunt on both points. Remove previously applied material completely before installing new paste, and clean the integrated heat spreader so it's free from oil, dust and other debris (Intel's thermal paste application guide). It also warns against adding new material on top of used material, which is the shortcut people take when they're in a hurry.

  1. Power down and unplug the system. Paste work happens around exposed components and alcohol, and a machine that's asleep on the bench is still a machine with a live standby rail.
  2. Lift the cooler and read the old imprint before you wipe it. Dry and crumbly means age. Paste pumped out to the edges in a ring means thermal cycling. Bare patches over the die area mean the last mount was short on compound or short on pressure.
  3. Strip every trace of the old layer. Plastic spudger or a fingernail, never a screwdriver. Clean to bare metal, then keep going until the cloth comes away clean.
  4. Wipe both mating surfaces with high-purity isopropyl alcohol. The heat spreader and the cooler base both need it, because contamination on either face prevents uniform contact. Check the corners and the machined grooves, where residue hides.
  5. Let both surfaces dry completely. Alcohol flashes off in seconds. Trapped solvent under a fresh layer is just a void waiting to form.

The common slip is a kitchen paper towel. It sheds fibers that end up pressed into the paste, so use a microfiber cloth or a coffee filter that leaves nothing behind.

If you'd rather see the whole cycle in motion, including the wipe-down and the screw sequence, this walkthrough covers it:

Milestone for this stage: clean, dry, bare metal on both sides and nothing else on it.

No visible spread beyond the chip, no dried crust at the edges, no lint. That's the state you want to be in before the next section.

How to Place the Paste and Mount the Cooler

The mounting step is where the degrees really live. Get the cooler seated flat and tightened evenly and the compound takes care of itself.

The sequence below is the one I'd follow on a conventional desktop CPU, and the amount matches what Intel specifies, a small central deposit of roughly a rice grain to a pea, about 3 to 5 mm across, placed in the middle of the heat spreader rather than painted over it (Intel's application and mounting steps).

how-to-apply-thermal-paste-cooler-installation.jpg
  1. Place the deposit in the center of the integrated heat spreader. One spot, straight down from the tube, no smearing into the corners.
  2. Lower the cooler vertically onto the chip. Straight down, in one movement, until the base meets the paste. Never slide or twist it into position.
  3. Hold light downward pressure while you catch the first screw threads. This keeps the block from walking as the screws bite, which is where most tilted mounts start.
  4. Bring the screws down in stages, diagonally. A couple of turns on one, then the one opposite, working around the pattern until all are snug. Pressure then develops evenly across the contact area instead of loading one corner.
  5. Check that the cooler sits level and look at the edges. Paste pushed past the heat spreader means you were generous. Non-conductive compound there is untidy, not dangerous, but it tells you the joint had more material than it needed.

On force, the guidance is a target, not a maximum. Intel's line is to use enough pressure to stop the cooler moving and distribute the compound, and not so much that the motherboard bows or the processor is damaged. Snug by hand feel, then stop.

Lateral movement is the quiet killer. Sliding or twisting the block after contact shears the layer into thick and thin regions and drags air in with it. Progressive diagonal tightening avoids the same problem from the screw side, where running one fastener all the way down tilts the cold plate and leaves one edge starved.

Server and high-power workstation platforms are a different job. Mounting hardware, torque values and even the dot pattern come from the chassis or board vendor, and on multi-socket systems the whole bring-up and cooling validation is usually handled as an engineering process rather than a bench task, which is the territory covered in NeoTeo's look at Supermicro's AI deployment services for system bring-up and cooling. Don't carry desktop habits onto a 2U chassis.

A cooler that goes on straight and tightens evenly beats a perfect pattern on a loose mount, every time.

A good install looks boring afterwards. Lift the cooler months later and you'll find a thin, even film across the die area, maybe a small ring at the edge, nothing on the socket, nothing crusted down the side of the package.

Mistakes That Ruin Contact and How to Avoid Them

Every failure on a repaste comes back to contact, not shape. Here are the ones that cost temperature, with the fix attached.

  • A thick layer across the whole heat spreader. The extra material becomes bond-line thickness the heat has to cross. Fix: one modest deposit in the center and let pressure do the work.
  • Too little compound. Dry patches over the die and air pockets under the cooler, which is the one quantity error that reliably shows up on the thermometer. Fix: when in doubt on a large rectangular lid, add a small amount rather than shaving it.
  • New paste on top of old. Uneven residue creates an interface with two different materials and no consistent thickness. Fix: strip to bare metal every single time.
  • Skipping the clean, or wiping with a shedding cloth. Lint and oil break up the contact patch. Fix: lint-free cloth and high-purity isopropyl alcohol, then let it dry.
  • Running one screw fully home before the others. The cold plate tilts and you get a thick region on one side and a starved one opposite. Fix: diagonal order, staged turns, all four snug at the same time.
  • Lifting and reseating the cooler after it has touched the chip. The layer tears and air gets pulled in. Fix: if you lift it, wipe both faces and start again.
  • An electrically conductive or mechanically rigid compound. Silver-loaded and liquid-metal products can bridge onto nearby components, and rigid interfaces can hold the heatsink off the die. Fix: a non-conductive paste for a normal build, and treat liquid metal as a separate procedure with its own masking plan.
Thin, even layerThick layer
Fills the pits, keeps the metal-to-metal gap smallAdds a conductive barrier the heat must cross
Covers the die area completely when mounted wellCovers it and keeps going, out past the edges
Forgiving of small mounting errorsHides a bad mount behind a mess
The goalThe most common result

A slightly thin layer with a properly torqued cooler beats a generous blob with a lazy mount, and it's easier to clean up next time. If temperatures climb after a repaste, don't assume the paste is at fault either. Firmware behaviour can move the numbers as much as a bad mount, as with the reported MSI Vcore cap on Ryzen 8000 chips.

Keep a dry cloth within reach while you work. Spills wiped immediately come off in one pass. Spills left to skin over take alcohol and patience.

When to Replace Paste Instead of Waiting

The trigger for a repaste is condition and service, not a calendar. Paste doesn't expire on a date. It dries as its carrier fluid escapes and it pumps out sideways as the metal on either side expands and contracts through thousands of heat cycles.

Manufacturer guidance summarized by HP puts quality paste at about 3 to 5 years in typical use, says heavily stressed gaming systems may benefit from replacement after roughly 2 years, and names increased fan noise, reduced performance or thermal throttling after 3 or 4 years as the practical warning signs (HP's thermal paste guidance). The same body of guidance says paste should be replaced whenever the cooler comes off, since separating it can disturb the existing layer and create uneven coverage or air pockets.

SituationTypical timingAction
Quality paste in typical useAbout 3 to 5 yearsPlan a repaste at that age
Heavily stressed gaming systemAbout 2 yearsInspect and consider repasting
Fan noise, lower performance or throttlingOften after 3 or 4 yearsRepaste and clean the cooler
Cooler removed for any reasonEvery timeRepaste, since the old layer may now have uneven coverage
Sustained temperature rise under the same workloadAny timeCheck mounting, then repaste if the rise persists

Reading a temperature rise

A rise only means something when it's measured like for like. Same workload, same fan curve, same room temperature, same amount of time for the reading to stabilize. Compare against a baseline you recorded earlier rather than a number you remember.

A machine holding its normal temperatures doesn't need repasting just because it's old. Age alone isn't a symptom. A steady climb under an unchanged workload is.

When the numbers do climb, check the mount before you blame the compound. A cooler that's worked loose, a screw that was never fully seated, or a plate sitting slightly tilted all produce the same warmer readings as dried paste, and all of them are cheaper to check than a repaste.

Check the Install With a Temperature Test

A repaste you can't compare to anything is just optimism. Verify it properly and you'll know within half an hour whether the job worked.

  1. Write down the room temperature and your fan settings. A warm afternoon and a silent fan profile will both make a good install look bad.
  2. Run the workload you already have data for. A sustained all-core stress test, or a fixed game scene you've benchmarked before. Sustained results are what count, which is why NeoTeo's Asus ProArt GR1X sustained CPU and AI results tell you more than a single peak benchmark.
  3. Let temperatures level off. Stop reading at the first high number. Wait for the curve to flatten, since a fresh mount takes minutes to reach steady state.
  4. Compare package or core temperature to your earlier baseline. Not the idle number, which moves with background noise, and not a screenshot you half remember.
  5. If it's worse, open it up again. Tilt, uneven screw pressure, a protective film left on the cooler base, or air pockets from a cooler that slid on the way down.
A bad application shows up as a jump. Test variation shows up as a wobble.

Run-to-run differences are normal even when nothing changed, so judge a clear shift rather than a single degree. A large rise after a repaste points at the way you applied and mounted it, not at the paste itself. Almost every time, it's coverage, film, or clamping pressure.

If temperatures match or improve, you're done. Note the date, the pattern and roughly how much you used, then keep that note with the build. Next time you open the case, you'll be comparing against a real baseline instead of guessing.


If you want the rest of your build checked against tested numbers rather than forum folklore, NeoTeo covers cooling and CPU performance with the same practical detail, from sustained-load benchmarks to repaste walkthroughs. Their hardware and tutorials sections are worth a read before you buy a cooler or plan a maintenance day. Head over to NeoTeo and keep your next build running cool.