A tube of thermal paste costs less than a fast-food lunch, but in the middle of 2026 it might be the cheapest performance upgrade left for a PC. With DRAM prices up 63% and GPU kit prices climbing another 22.5% at retail in China alone, replacing a graphics card or CPU cooler just to shave a few degrees off your load temperatures is a hard sell. Repasting the hardware you already own is not. This tutorial walks through exactly how to apply thermal paste to a CPU and a GPU, compares the paste and application methods that actually matter, and shows how to build a small logging script that measures the before-and-after difference in real numbers instead of guesses, because “it feels cooler” isn’t proof of anything.
Expect to spend 45 minutes doing the physical work described in the 12 steps below, plus optional time running the verification script at the end. Everything here applies to desktop CPUs and GPUs, and a dedicated section near the end covers what changes on a laptop. None of the steps require soldering, desoldering, or permanent modification. Worst case, you clean everything off and start the pea-sized dot over again.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Why Thermal Paste Matters More in 2026’s Hardware Market
Thermal paste (also called thermal interface material, or TIM) fills the microscopic gaps between a chip’s surface and its cooler’s cold plate. Metal-on-metal contact looks smooth to the eye but is full of tiny peaks and valleys at a microscopic level, and those gaps trap air, which is a poor conductor of heat. Paste displaces that air so heat can move from the die into the heatsink instead of building up. Thermal paste is a solved problem in theory, but it degrades: solvents evaporate, some compounds separate or “pump out” from repeated heat cycles, and a paste job that was fine three years ago is quietly costing you thermal headroom today.
That headroom matters more this year specifically. Tech Insider covered how RAM prices jumped 63% as DRAM hit a record $20 a chip, and separately how Nvidia and AMD hiked GPU kit prices, pushing China retail up 22.5%. When upgrade paths get more expensive, the calculus shifts toward extending the life and performance of the silicon already installed in your case. A GPU running 8-10°C hotter than it should because of dried-out paste isn’t just noisier. Sustained higher temperatures push a chip closer to its thermal design power (TDP) limit, which caps clock speeds under load regardless of how much headroom the silicon itself has. Cooler chips also tend to run more efficiently and can hold boost clocks longer, which is exactly the kind of free performance worth claiming before considering a new purchase.
None of this requires exotic hardware or a degree in materials science. It requires a screwdriver, a syringe of paste that costs somewhere between $5 and $15, about 45 minutes, and a willingness to open your case. The physics is simple even if the marketing around it isn’t: a heat sink can only pull heat away as fast as it can get that heat off the chip in the first place, and the paste layer is the bottleneck at the very start of that chain. Every other cooling upgrade (bigger fans, more radiator surface area, better case airflow) sits downstream of this one $8 fix.
Prerequisites: Tools, Paste, and Software You’ll Need
Gather everything before you start disassembling anything. Working with the CPU or GPU exposed for longer than necessary raises the odds of dust contamination or an accidental knock to an exposed connector.
- Thermal paste. Pick one non-conductive compound for your first repaste. The table below compares real current options.
- Isopropyl alcohol. Use the highest concentration you can find (90%+ is usable, higher is better since it evaporates faster and leaves less residue).
- Lint-free wipes or coffee filters. Paper towels shed fibers that stick to paste residue. Coffee filters are a cheap, genuinely effective substitute for dedicated electronics wipes.
- A Phillips #1 or #2 screwdriver. Check your cooler’s backplate screws before you start, since some AIO and tower coolers use a hybrid or Torx head instead.
- Thermal pads (GPU repastes only). Needed if you’re also servicing VRM and memory contact pads, typically sold in 0.5mm to 2mm thicknesses. Measure your existing pads with calipers before ordering replacements, since the wrong thickness under-fills or over-compresses the gap.
- An anti-static wrist strap. Optional but cheap insurance, especially on dry, low-humidity days. Electrostatic discharge you can’t even feel is still strong enough to damage a chip’s internal circuitry, and a $6 strap clipped to a grounded, unpainted part of your case removes the risk entirely.
- Monitoring software. HWiNFO64 for full sensor logging on Windows, MSI Afterburner or TechPowerUp’s GPU-Z for GPU-specific readouts, and nvidia-smi (bundled with the Nvidia driver) or lm-sensors on Linux. Use whatever the current stable release is for each tool at install time, and update before you start rather than relying on a specific version number.
On the paste itself, the market splits into three tiers: budget silicone-based compounds, carbon/ceramic non-conductive pastes (the right choice for nearly everyone), and electrically conductive liquid metal for advanced users. Prices and thermal conductivity ratings below come from Club386’s 2026 thermal paste comparison testing and Tom’s Hardware’s paste roundup.
| Thermal Paste | Type | Conductivity (W/mK) | Approx. Price | Best For |
|---|---|---|---|---|
| Arctic MX-4 | Carbon-based, non-conductive | 8.5 | ~$6 / 4g | Budget CPU/GPU builds |
| Noctua NT-H1 | Non-conductive paste | ~8.9 | ~$9 / 3.5g | Easiest to apply, beginner-friendly |
| Arctic MX-6 | Carbon-based, non-conductive | Not disclosed by maker | ~$8 / 4g | General-purpose, long service life |
| Endorfy Pactum 4 | Non-conductive paste | 12.0 | ~$5 / 1.5g | Budget performance pick |
| Thermal Grizzly Kryonaut | Non-conductive paste | 12.5 | ~$10-13 / 1g | Enthusiast CPU and GPU builds |
| Thermalright TF8 | Non-conductive paste | 13.8 | ~$4 / 2g | High value-per-gram performance |
| Thermal Grizzly Kryonaut Extreme | Non-conductive paste | ~14.2 | ~$16 / 1g | Overclocked, 250W+ chips |
| Alphacool Apex | Non-conductive paste | 17.0 | ~$3 / 1g | High conductivity on a budget |
| Thermal Grizzly Conductonaut Extreme | Liquid metal, electrically conductive | 80.0 | ~$13 / 1g | Advanced CPU delidding, extreme cooling |
If you’re repasting a GPU, stick to a non-conductive compound unless you have real experience masking off the die. Reviewers and roundups consistently flag that a bare graphics card die sits close to surface-mount components and voltage regulation circuitry, and a stray bead of liquid metal that bridges two of those components can short the card. Arctic’s MX series, Noctua NT-H1, and most carbon or ceramic-based pastes are safe for GPU use precisely because they don’t conduct electricity if they spread further than intended.
Don’t buy based on the conductivity column alone. A 17.0 W/mK paste applied with an air bubble in the middle underperforms an 8.5 W/mK paste applied cleanly with even coverage. For a first repaste, prioritize a compound that’s forgiving to apply (Noctua NT-H1 and Arctic’s MX line are consistently rated easiest to work with by reviewers) over chasing the highest number in the conductivity spec sheet.
Signs Your CPU or GPU Needs a Repaste
You don’t need to repaste on a fixed calendar. Watch for these signals instead:
- Rising idle temperatures over months, with no change in ambient room temperature or case airflow.
- Fan curves ramping earlier under the same workloads that used to run quiet.
- Clock speed instability under sustained load. A GPU or CPU that boosts high for a few seconds, then drops and holds a noticeably lower clock, is often thermal-throttling.
- A pre-built or older laptop that has never been opened. Factory paste application on mass-produced systems is frequently inconsistent, sometimes applied by machine with less precision than a careful manual job.
- Hardware older than 3-5 years that has never been serviced, particularly if it runs demanding workloads regularly.
Before you touch a screwdriver, get a baseline. Open HWiNFO64, run your normal workload (a game, a render, a compile job, whatever actually loads the chip you’re worried about) for 15-20 minutes, and note the peak and average temperatures under the sensor tab. Without this baseline, you have no way to confirm the repaste actually helped, which is the entire point of the logging project later in this guide.
It’s also worth ruling out non-paste causes before you assume a repaste is the fix. Dust buildup on a heatsink’s fins or a radiator’s core can raise temperatures just as much as degraded paste, and it’s a five-minute can-of-compressed-air fix rather than a full teardown. Check your case fans are actually spinning at the RPM your fan curve expects, too. A fan header miswired to a “silent” profile will produce the exact same symptoms as bad paste, and no amount of repasting fixes a fan that isn’t ramping.
Step-by-Step: Applying Thermal Paste to a CPU (Steps 1-6)
Step 1: Power down, unplug, and ground yourself. Shut down completely, switch off the PSU, and unplug the power cable. Don’t rely on the OS shutdown alone. Touch a grounded metal surface or wear an anti-static strap before touching any component.
Step 2: Remove the CPU cooler. For tower air coolers, unscrew the backplate bolts in a diagonal pattern (opposite corners first) to release mounting pressure evenly. For an AIO liquid cooler, do the same with the pump bracket screws. If the block doesn’t lift free immediately, twist it gently side to side rather than prying, since old paste can act like a weak adhesive.
Step 3: Clean the old paste from both surfaces. Dampen a lint-free wipe or coffee filter with isopropyl alcohol and wipe the CPU’s integrated heat spreader (IHS) and the cooler’s cold plate until no residue remains. Use a fresh section of the wipe for each pass so you’re not just spreading old paste around. Let both surfaces air-dry for a minute before continuing.
Step 4: Inspect the CPU socket and pins. While the cooler is off, glance at the socket for bent pins (LGA sockets) or debris. This takes ten seconds and can save you a much worse afternoon if something’s wrong.
Step 5: Apply the paste. For most modern CPU IHS sizes, a pea-sized dot in the center is enough. The clamping pressure from the cooler spreads it evenly across the surface without you needing to manually spread it (manual spreading is more prone to trapping air bubbles). For larger IHS surfaces or coolers with known hot spots, a thin X pattern gives more even initial coverage. See the method comparison table further down for when each pattern makes sense.
Step 6: Reseat the cooler. Lower it straight down without sliding it side to side (sliding smears the paste unevenly before clamping pressure is applied). Tighten mounting screws in a diagonal, star-pattern sequence, a little at a time on each screw rather than fully tightening one before moving to the next. This distributes pressure evenly across the die instead of tilting the cold plate.
Step-by-Step: Applying Thermal Paste to a GPU (Steps 7-12)
GPUs are more involved than CPUs because the cooler shroud, fans, and backplate all have to come off, and many cards have separate thermal pads for VRAM and VRM components in addition to paste on the die. Go slowly here. This is also the step where warranty terms matter most, so check your GPU’s warranty policy on opening the shroud before you start. Budget closer to 25-30 minutes for the GPU half of this job versus roughly 15 for the CPU half. Most of that extra time is cable management and the added care needed around a bare PCB rather than anything technically harder.
Step 7: Photograph everything before disassembly. Take clear photos of the shroud screws, fan cable connectors, and backplate from multiple angles. GPU teardown mistakes are almost always reassembly mistakes, and photos are the cheapest insurance against forgetting where a cable or standoff goes.
Step 8: Remove the backplate, then the shroud. Backplate screws are usually straightforward, but disconnect any fan or ARGB cables linking the shroud to the PCB before lifting it away. Yanking the shroud off with cables still attached is a common way to snap a connector.
Step 9: Lift the cooler assembly off the PCB. Unscrew the cold plate mounting screws in a diagonal pattern, same as a CPU cooler. The block may be lightly stuck to the die from old paste. Twist gently, and don’t pry with a tool near the PCB.
Step 10: Clean the die and inspect the memory/VRM pads. Clean the GPU die exactly as you did the CPU IHS. While you’re in there, check the condition of the thermal pads over VRAM modules and VRM components. If they’re compressed flat, cracked, or crumbling, replace them with pads matched to the original thickness rather than skipping this step. Those components can run hotter than the core itself under load.
Step 11: Apply paste to the die. GPU dies are frequently smaller and more rectangular than a CPU IHS. A slightly smaller pea-sized dot, or a short thin line down the center for rectangular dies, tends to spread more predictably than an X pattern on these smaller surfaces.
Step 12: Reassemble in reverse order and do a visual check. Cold plate first (even diagonal screw tightening again), reconnect all fan and ARGB cables before closing the shroud, then backplate last. Before you close the case, boot the system and confirm the fans spin and the display outputs correctly. It’s much easier to fix a loose connector now than after you’ve closed everything up.
Pea, X, Line, or Spread? Comparing Application Methods
There’s more community debate about paste application patterns than the physics really justifies, but the patterns below each have a genuine best use case rather than being purely aesthetic preference.
| Method | Best For | Difficulty | Air Bubble Risk |
|---|---|---|---|
| Pea (single dot, center) | Most CPUs, small-to-mid GPU dies | Easiest | Low |
| X pattern | Larger IHS surfaces, uneven cooler pressure | Moderate | Low-Moderate |
| Line (single stripe) | Rectangular GPU dies | Easy | Low |
| Five-dot / spread | Very large dies, delidded CPUs | Moderate | Moderate |
| Manual spread (card/spatula) | Liquid metal, precise thin-layer control | Hardest | Highest if done poorly |
The pea method wins on simplicity because clamping pressure does the spreading work for you. Squeeze a cooler down evenly and a single dot of the right size will spread edge to edge without manual intervention or trapped air pockets. Manual spreading with a card or spatula gives more visual control but introduces more opportunity for streaks and air bubbles unless you’re experienced, which is why it’s mainly reserved for liquid metal application where a thin, even, fully-covered layer is critical for both performance and safety.
Choosing Monitoring Software: HWiNFO64 vs Afterburner vs GPU-Z vs nvidia-smi
You’ll want at least one of these running before you touch a screwdriver, both to capture the baseline mentioned earlier and to build the verification log in the next section. They overlap, but each has a reason to exist rather than being interchangeable clones of each other.
| Tool | Platform | Best For | CSV Logging |
|---|---|---|---|
| HWiNFO64 | Windows | Full sensor breadth: CPU, GPU, VRM, storage in one view | Yes, built in |
| MSI Afterburner | Windows | GPU-focused live overlay while gaming, fan curve control | Yes, built in |
| TechPowerUp GPU-Z | Windows | Quick GPU sensor snapshot and card validation | Limited |
| nvidia-smi | Windows/Linux | Scriptable Nvidia GPU polling, no GUI needed | Via redirect, as shown above |
| lm-sensors | Linux | Kernel-level CPU, motherboard, and many GPU sensors | Via redirect/parsing |
If you only install one thing, make it HWiNFO64. It reads more sensors than any single alternative on this list, and its CSV export is what feeds the comparison script below if you’d rather not touch the command line at all. MSI Afterburner earns its place for its on-screen overlay, which is the fastest way to eyeball temperatures in real time while a game or stress test is running. Command-line tools matter for a different reason: they’re scriptable, which is what turns a one-off temperature check into a repeatable, automated before-and-after log. Whether you’re running an Nvidia GeForce card with nvidia-smi or an AMD Radeon card monitored through AMD’s own driver and monitoring stack, the goal is the same: numbers you can compare, not a temperature you half-remember from before you opened the case.
Building a Temperature-Logging Script to Verify Your Repaste
“It feels cooler” isn’t data. If you’re going to spend 45 minutes on this, spend another 10 building something that tells you objectively whether it worked. This section builds a small before/after logging toolkit using tools you likely already have installed.
Start with a quick one-off temperature check. On Windows with an Nvidia GPU, the driver ships with nvidia-smi, which can poll and log temperature, utilization, and power draw on an interval:
nvidia-smi --query-gpu=timestamp,temperature.gpu,utilization.gpu,power.draw --format=csv -l 5On Linux, lm-sensors covers both CPU and many GPU/VRM sensors once installed and configured:
sudo apt install lm-sensors
sudo sensors-detect --auto
watch -n 2 sensorsWindows doesn’t expose reliable per-core CPU die temperatures through a simple built-in command the way Linux does through the kernel’s hwmon subsystem, but you can pull a general ACPI thermal zone reading through PowerShell as a fallback (HWiNFO64 remains the more accurate option for real per-core CPU readings):
Get-CimInstance -Namespace "root/wmi" -ClassName MSAcpi_ThermalZoneTemperature | ForEach-Object {
"{0:N1} C" -f (($_.CurrentTemperature / 10) - 273.15)
}Now combine polling with a timer so you get a full CSV log across a stress test window instead of a single snapshot. Save this as repaste-log.ps1. Run it once before you repaste (rename the output before-repaste.csv) and once after (after-repaste.csv), starting a GPU stress test in another window each time so both logs capture the same kind of sustained load:
# repaste-log.ps1
param(
[int]$DurationMinutes = 15,
[string]$OutFile = "repaste-log.csv"
)
"timestamp,temp_c,util_pct,power_w" | Out-File $OutFile
$endTime = (Get-Date).AddMinutes($DurationMinutes)
while ((Get-Date) -lt $endTime) {
$line = nvidia-smi --query-gpu=timestamp,temperature.gpu,utilization.gpu,power.draw --format=csv,noheader,nounits
$line | Add-Content $OutFile
Start-Sleep -Seconds 5
}
Write-Host "Logged $DurationMinutes minutes to $OutFile"Finally, a short Python script turns the two CSV files into a plain-language before/after summary instead of two spreadsheets you have to eyeball:
# compare_repaste.py
import csv
def summarize(path):
temps = []
with open(path) as f:
for row in csv.DictReader(f):
temps.append(float(row["temp_c"]))
return {
"samples": len(temps),
"min_c": min(temps),
"max_c": max(temps),
"avg_c": round(sum(temps) / len(temps), 1),
}
before = summarize("before-repaste.csv")
after = summarize("after-repaste.csv")
print(f"Before: avg {before['avg_c']}C, max {before['max_c']}C, samples {before['samples']}")
print(f"After: avg {after['avg_c']}C, max {after['max_c']}C, samples {after['samples']}")
print(f"Delta: {round(before['avg_c'] - after['avg_c'], 1)}C average improvement")This whole toolkit is deliberately small (under 40 lines combined) because the goal isn’t a monitoring platform. It’s a repeatable, objective before/after comparison you can point to instead of guessing. Swap the nvidia-smi calls for an AMD equivalent or HWiNFO64’s CSV logging export if you’re not on an Nvidia card. The CSV parsing script works the same either way as long as the column names line up.
Reading Your Results: Sample Output and What Good Numbers Look Like
Here’s an example of what the comparison script’s output format looks like. These are illustrative placeholder numbers to show the format, not a benchmark claim. Your actual delta depends heavily on how degraded your old paste was, your case airflow, ambient room temperature, and your cooler’s mounting pressure:
Before: avg 78.4C, max 84.0C, samples 180
After: avg 71.2C, max 76.0C, samples 180
Delta: 7.2C average improvementA meaningful improvement is usually visible in both the average and the max columns together. A drop in average with no change in max often just means your workload wasn’t consistent between the two runs, which is why running the same stress test for the same duration both times matters more than people expect. If your after numbers come back worse than before, don’t panic and don’t assume the paste is bad. It almost always means a mounting or seating problem covered in the troubleshooting section below.
If you don’t want to script anything, HWiNFO64’s built-in logging (Sensor status window, right-click, “Log to CSV file”) captures the same kind of data with a GUI instead of PowerShell, at the cost of manually opening two spreadsheets to compare them.
5 Common Thermal Paste Mistakes That Ruin a Repaste
1. Using too much paste. More paste does not mean better heat transfer. Paste is actually a worse thermal conductor than the direct metal-to-metal contact it’s replacing gaps between. A pool of excess paste squeezed out the sides can also spread onto the socket, motherboard, or GPU PCB, and with a conductive liquid metal compound that’s a real short-circuit risk. A pea-sized dot is almost always enough.
2. Not fully removing the old paste first. Applying fresh paste over old, partially dried residue creates an uneven base layer, which defeats the purpose of the whole exercise. Clean both surfaces down to bare metal before reapplying anything.
3. Sliding the cooler after it makes contact. Twisting or sliding the cooler once it touches the paste smears it unevenly and can drag air bubbles into the middle of the contact patch. Set it straight down and don’t reposition it once it touches.
4. Uneven mounting pressure. Tightening one screw fully before moving to the next tilts the cold plate and creates a wedge-shaped paste layer, thick on one side and nearly squeezed out on the other. Always tighten in a diagonal star pattern, a little at a time per screw.
5. Using electrically conductive paste near exposed components without masking. Liquid metal compounds are excellent thermal performers but conduct electricity. On a GPU especially, where surface-mount components sit close to the die, an inexperienced first application is genuinely risky. Non-conductive paste removes this failure mode entirely, which is why it’s the right default for a first repaste.
Advanced Techniques: Liquid Metal, PTM7950 Pads, and Delidding
Once you’re comfortable with a standard paste repaste, a few more advanced options exist for squeezing out additional thermal performance. None of these are necessary for a typical repaste. They’re worth knowing about, not worth reaching for on your first attempt.
Liquid Metal
Compounds like Thermal Grizzly Conductonaut Extreme use a gallium-based alloy with roughly 80 W/mK conductivity, several times higher than even premium carbon-based pastes. The tradeoff is real risk: it’s electrically conductive, it can amalgamate with (corrode) bare aluminum, and it needs to be applied in a thin, fully controlled layer, usually spread manually with the included applicator rather than dotted and clamped. Mask the die edges and any nearby exposed metal with electrical tape before application, and never use liquid metal on a cooler with an aluminum cold plate. Copper and nickel-plated surfaces are the safe pairing.
PTM7950 Phase-Change Pads
PTM7950 is a solid pad at room temperature that transitions into a liquid-like, paste-consistency state once it reaches operating temperature, then re-solidifies when the system cools. That behavior makes it popular for laptops and handhelds, where thermal cycling and orientation changes cause standard paste to pump out from beneath the cooler faster than in a stationary desktop. It’s sold as thin pre-cut or roll-form sheets rather than a syringe, and reapplication is more about replacing the pad outright than cleaning and reapplying compound.
Delidding
Delidding removes the CPU’s integrated heat spreader entirely to replace the factory-applied internal TIM (between the silicon die and the IHS) with liquid metal or a higher-performance paste, then reseals or clamps the IHS back on. It’s the most effective single thermal upgrade available for supported CPUs, and also the most likely to end in a dead chip if you crack the die during removal. Dedicated delidding tools that use a controlled shearing motion rather than a blade meaningfully lower that risk, but this step is worth attempting only after you’re fully comfortable with standard repasting and understand it typically voids the CPU’s warranty.
Troubleshooting: 8 Post-Repaste Problems and Fixes
Most repaste problems trace back to one of two root causes: uneven mounting pressure, or a cable/connector that got bumped loose during reassembly and was never the paste’s fault at all. Work through these in order before assuming you need to redo the whole job.
1. Temperatures are the same or worse after repasting. Almost always a mounting issue. Recheck that all screws are evenly tightened in a diagonal pattern and that the cooler is fully seated flat against the die.
2. System won’t POST after reassembly. Reseat the CPU in its socket and confirm no pins are bent (LGA sockets). Reseat RAM and the 24-pin/8-pin power connectors too, since it’s easy to bump one loose while working around the cooler.
3. Paste squeezed out visibly around the edges of the die. You used too much. Power down, remove the cooler, clean both surfaces completely, and reapply a smaller amount.
4. AIO pump is running but temperatures are high. Check that the pump bracket made full contact. A common mistake is leaving one corner of the bracket resting on a standoff instead of flush against the block. Also confirm the radiator fans are wired to a header that’s actually being controlled by a fan curve, not running at a fixed low RPM.
5. Fans don’t spin after GPU reassembly. Almost always a fan cable that wasn’t reconnected when the shroud went back on. This is the single most common GPU repaste mistake, and it’s why Step 7 tells you to photograph cable routing before you start.
6. No display output after a GPU repaste. Reseat the card fully in the PCIe slot and confirm the PCIe power connectors are clicked in, not just resting in place. If you also removed and cleaned drivers during the process, reinstall the latest GPU driver before assuming it’s a hardware fault.
7. Thermal throttling persists even after a clean repaste. Confirm the fan curve is actually ramping under load using FanControl or your motherboard’s fan software. A repaste can’t fix a fan curve that’s capped too low, or case airflow that’s fighting itself. Check fan airflow direction if you haven’t verified it recently.
8. Temperature sensor readings look obviously wrong (like 0°C or 128°C constant). This is typically a software/driver reporting issue rather than a hardware fault. Update HWiNFO64 and your GPU driver to their current versions, and confirm you’re reading the correct sensor. GPUs often expose several die-adjacent sensors alongside the primary core temperature, and monitoring tools don’t always default to the right one.
How Often You Should Reapply Thermal Paste
There’s no single universal number here. Lifespan depends on the specific compound, how many thermal cycles (power-on to full load to idle) the system goes through, and ambient conditions. Reviewers and paste manufacturers’ own product positioning give a reasonable general range to plan around rather than a guaranteed expiration date:
| Paste Type | Typical Usable Life | Check Sooner If… |
|---|---|---|
| Budget silicone/carbon paste (MX-4 tier) | ~2-3 years | System runs 24/7 or in a hot ambient environment |
| Long-life carbon paste (MX-6 tier) | Up to ~8 years per maker positioning | Heavy sustained workloads (rendering, mining, AI training) |
| High-performance enthusiast paste (Kryonaut tier) | ~3-5 years under sustained high wattage | Chip is overclocked or running 250W+ |
| Liquid metal | Several years, but monitor for pump-out/corrosion | Any visible corrosion on aluminum-adjacent surfaces |
| Factory pre-applied (laptops, pre-builts) | Highly variable, often the weakest link | First repaste on any unopened system 3+ years old |
Treat these as planning ranges rather than hard deadlines, and let the symptom checklist earlier in this guide (rising idle temps, earlier fan ramp, unstable boost clocks) be the actual trigger. A chip that’s still hitting its expected temperatures three years in doesn’t need to be opened just because a chart says so.
Thermal Paste for Laptops: What’s Different
Laptop repasting follows the same core principles with tighter constraints. A few differences worth planning around before you open one:
- Less clamping pressure. Laptop coolers use spring-loaded screws with far less mounting force than a desktop tower cooler, which is exactly why phase-change pads like PTM7950 have become popular in laptops and handhelds. They tolerate lower, less even pressure better than traditional paste.
- Smaller, more fragile connectors. Fan cables, coin-cell CMOS batteries, and ribbon cables are far more delicate than their desktop equivalents. Go slower, and don’t force any connector that resists.
- Warranty implications are usually stricter. Check your laptop manufacturer’s policy before opening the chassis. Many use a visible warranty seal over a screw, and some ultrabooks are not designed to be opened by the end user at all.
- Smaller paste amounts. Laptop CPU and GPU dies are frequently tiny compared to desktop equivalents, and a full pea-sized dot is often too much. Use a rice-grain-sized amount instead.
- Reassembly torque matters more. Laptop chassis screws strip more easily than desktop case screws. Use a properly sized screwdriver bit and stop as soon as you feel resistance, not when it feels “desktop tight.”
If your laptop is still under warranty, weigh the thermal gains against the cost of voiding coverage. For a system that’s misbehaving out of warranty, the same 45-minute process described above applies, just with smaller tools and more patience.
Put together, this is one of the few PC maintenance jobs where the cost-to-benefit ratio hasn’t moved with the rest of the hardware market. A tube of paste is still a few dollars, the tools are reusable, and the logging script above turns “I think it’s better” into an actual number you can point to. In a year where a single memory kit or graphics card costs meaningfully more than it did twelve months ago, that’s a rare thing to get essentially for free.
Frequently Asked Questions
How often should I reapply thermal paste?
Most builds go 2-5 years before it’s worth checking, depending on the paste used and how hard the system runs. Watch for rising idle temperatures or earlier fan ramp-up rather than following a fixed calendar. A system under a light home-office workload can often go longer than the ranges in the maintenance table above, while a workstation running sustained renders or training jobs around the clock should check sooner.
Can I use too much thermal paste?
Yes. Excess paste can squeeze out around the die and, with conductive compounds especially, create a short-circuit risk on nearby components. A pea-sized dot is enough for the vast majority of consumer CPUs and GPUs.
Is liquid metal thermal paste safe for GPUs?
It’s riskier than standard paste because GPU dies sit close to exposed surface-mount components, and liquid metal conducts electricity. It’s used successfully by experienced builders who mask the die edges carefully, but it isn’t the recommended starting point for a first repaste.
Do I need to remove all the old thermal paste before reapplying?
Yes. Applying new paste over old residue creates an uneven layer that undermines the whole point of repasting. Clean both the chip surface and the cooler’s cold plate down to bare metal first.
What concentration of isopropyl alcohol should I use to clean thermal paste?
Higher concentrations (90% and up, ideally 99%) are commonly recommended by PC builders because they evaporate faster and leave less residue behind than lower concentrations, which contain more water.
Can I apply thermal paste without removing the cooler?
No. The cooler has to come off completely so you can clean the old paste and access the bare die or IHS. There’s no way to top up or refresh paste without full removal.
Does thermal paste expire before you even open the tube?
Unopened paste stored in reasonable conditions lasts a long time, but syringes can dry out or separate after years in storage. If a tube feels unusually stiff, grainy, or won’t extrude smoothly, it’s better to buy a fresh one than risk a poor application.
Will repasting void my GPU or laptop’s warranty?
It can. Many GPU manufacturers and most laptop makers apply a warranty seal over at least one shroud or chassis screw. Check your specific product’s warranty terms before opening it, especially if the hardware is still within its coverage window. If the card or laptop is still covered and running within normal temperatures, it’s usually worth waiting until the warranty expires rather than trading guaranteed replacement coverage for a few degrees of thermal headroom.
Related Coverage
- How to Stress Test a GPU and CPU: 12 Steps, 60 Min [2026]
- How to Overclock a GPU: 12 Steps, 60 Min [2026]
- How to Undervolt a GPU: 12 Steps, 45 Min [2026]
- Fix PC Fan Airflow Direction: 12 Steps, 45 Min [2026]
- How to Install a Graphics Card: 12 Steps, 45 Min [2026]
- FanControl: 12-Step Fan Curve Setup, 30 Min [2026]
- RAM Prices Jump 63% as DRAM Hits Record $20 a Chip [2026]
