A graphics card that shipped at 65°C under load two years ago and now throttles at 84°C hasn’t gotten weaker. Its thermal paste has. Factory-applied compound on most GPUs starts drying out or migrating away from the die (a failure mode enthusiasts call “pump-out”) somewhere between 18 and 36 months of regular use, and the fix costs less than a fast-food meal. This guide walks through repasting a GPU from first principles: what tools you actually need, how to strip and clean the die without damaging anything, which 2026 thermal paste and pad products are worth buying, and how to verify the job worked with real before-and-after numbers.
Everything below applies to current RTX 40/50-series and Radeon RX 7000/9000-series cards, and the same core process works on older GPUs too. Budget 45 minutes for a first attempt, closer to 30 once you’ve done it once.
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 GPU Thermal Paste Fails and When to Replace It
Thermal paste’s job is simple: fill the microscopic gaps between the GPU die and the cooler’s contact plate so heat transfers efficiently instead of getting trapped in pockets of air. Stock paste on most graphics cards is a cheap, mass-applied compound chosen for manufacturing speed, not longevity. Under the thermal cycling that comes from repeated gaming sessions, that paste dries out, loses viscosity, and in worst cases physically migrates away from the center of the die toward the edges, a phenomenon known as pump-out. The center of the die, where heat concentrates most, ends up with a thinner paste layer than the edges, and core and hotspot temperatures climb even though nothing else about the card has changed.
Pump-out is a documented issue on high-power, vertically mounted, and small-form-factor GPU installations in particular, where mechanical stress and thermal cycling both work against the paste’s original placement. It’s why 2025-2026 thermal interface guides increasingly recommend phase-change materials over standard paste for GPUs that see extended stress: those materials resist the migration effect that undermines conventional compounds over time.
You don’t need to guess whether your card needs attention. Watch for these signals: idle temperatures creeping above 45-50°C in a case with reasonable airflow, load temperatures on an RTX 40/50-series or RX 7000/9000-series card exceeding roughly 80-85°C under sustained gaming, a growing gap between core temperature and hotspot temperature (anything beyond 15-20°C suggests uneven paste coverage), fan curves ramping earlier and louder than they used to for the same workload, or a card that’s simply 2+ years old and has never been opened. None of these alone is proof, but two or more together is a strong signal that a repaste will help.
Factory paste application is also a manufacturing tradeoff, not an engineering failure. Board partners apply paste with automated dispensing arms tuned for speed and consistency across thousands of units a day, not for the ideal dot size on any individual die. Small variances in dot placement, volume, or air entrapment during that automated process mean some cards leave the factory already running a few degrees warmer than an identical unit down the assembly line. A repaste doesn’t just replace aged material, it also gives you a chance to correct any manufacturing inconsistency that was baked in from day one.
Usage pattern changes how quickly paste degrades. A card in a well-ventilated desktop case running a few hours of gaming most evenings ages slowly and predictably. The same card crammed into a small-form-factor enclosure with limited airflow, mounted vertically via a riser cable, or run near-continuously for rendering, mining, or AI workloads sees far more aggressive thermal cycling and can show pump-out symptoms in under a year. Match your repaste schedule to how the card is actually used, not to a generic calendar reminder.
How Often Different GPU Use Cases Need Repasting
Not every GPU ages at the same rate, and matching your monitoring frequency to your actual workload saves you from either over-servicing a card that’s fine or missing a card that’s quietly throttling. The table below reflects general patterns reported across 2025-2026 repaste communities and hardware forums, not a fixed manufacturer schedule.
| Use Case | Typical Repaste Interval | Why |
|---|---|---|
| Casual gaming, a few hours a day, horizontal mount | 3-4 years | Moderate, intermittent thermal cycling; paste ages slowly |
| Daily gaming, well-ventilated case, horizontal mount | 2-3 years | Regular thermal cycling accelerates gradual pump-out |
| Small-form-factor build, vertical GPU mount | 12-18 months | Gravity and riser-cable stress compound pump-out risk |
| Workstation rendering or 24/7 AI/compute load | 12-18 months | Near-constant high load accelerates thermal cycling |
| Laptop GPU (any use pattern) | 18-24 months | Tighter thermal envelope and thinner factory paste application |
Prerequisites: Tools, Materials, Skill Level, and Time
This is a beginner-to-intermediate hardware project. If you’ve ever installed RAM or swapped a storage drive, you have the manual dexterity required. The main risks are electrostatic discharge, over-tightening screws, and touching the die with unclean tools, all of which are avoidable with basic care.
- Thermal paste: Arctic MX-6 (non-conductive, roughly $6-12 depending on tube size, the default recommendation for most gaming GPUs), Thermal Grizzly Kryonaut (12.5 W/mK, about $7-10 per gram) for a step up in performance, or Thermal Grizzly Kryonaut Extreme (14.2 W/mK, around $24.99 for 2g) for heavily overclocked or vertically mounted cards prone to pump-out.
- Thermal pads for VRAM/VRM: Thermalright Odyssey 12.8 W/mK pads (1.0mm thickness common for GPU use) or Gelid GP-Ultimate 15 W/mK pads, both non-electrically-conductive silicone pads safe to use around memory modules.
- Isopropyl alcohol: 90-99% concentration. Anything below 70% leaves residue and evaporates too slowly.
- Lint-free cloths or coffee filters (coffee filters are cheap and genuinely lint-free, a trick borrowed from repair techs).
- Cotton swabs for edges and connector pins.
- Precision screwdriver set with PH0/PH1 and Torx T5-T8 bits. A torque driver rated 0.2-1.0 N·m is worth the roughly $20-30 investment if you plan to open more than one card, since GPU cooler screws are commonly torqued in the 0.3-0.6 N·m range and overtightening can crack the die or warp the backplate.
- Anti-static wrist strap or a habit of touching a grounded metal surface before handling the card.
- A phone camera to photograph screw layout, cable routing, and pad placement before you take anything apart.
- Monitoring software: HWiNFO64 or GPU-Z for temperature logging, MSI Afterburner or a vendor equivalent for fan curve control, and a stress-test tool like FurMark or 3DMark for validation.
Total DIY cost typically lands between $20 and $50 for paste, pads, and cleaning supplies, enough material for several GPUs. A professional repaste and re-pad service at a repair shop generally runs $40-80, or $80-120 if you want premium compounds like PTM7950 or Kryonaut Extreme included. Doing it yourself once pays for the tools.
DIY vs. Professional Repaste: Which Should You Choose
The math almost always favors DIY if you’re comfortable following a checklist and working carefully around static-sensitive electronics. A single tube of Arctic MX-6 and a set of thermal pads runs under $20 combined and services multiple cards over several years, while a shop visit costs that much or more for a single card and requires shipping or dropping the GPU off, adding days of downtime. The break-even point arrives on your very first repaste.
Professional service still makes sense in a few specific situations: if your GPU is still under warranty and the manufacturer’s terms are strict about user disassembly, if you don’t own or want to buy precision tools for a one-time job, or if the card shows signs of a problem beyond paste, like a warped contact plate, damaged fan bearing, or coil damage, that requires diagnostic equipment you don’t have. A repair shop can also validate a suspected paste issue against a hardware fault before you spend money on paste that won’t fix the real problem.
For most gaming PC owners with a card that’s simply aged past its factory paste’s useful life, DIY repasting is a 30-45 minute project with a strong track record of success when the steps below are followed in order.
Step 1-2: Baseline Temperature Log and Safe Power-Down
Before you touch a screwdriver, record what your GPU is doing right now. Without a baseline, you have no way to prove the repaste actually helped, and you lose the ability to catch a bad reassembly if temperatures get worse instead of better.
Open HWiNFO64, enable sensor logging, and run a 10-15 minute gaming session or a FurMark stress test while recording GPU core temperature, hotspot temperature, VRAM temperature (where the sensor is exposed), fan speed, and clock speeds. Save that log somewhere you won’t lose it.
# Windows: quick temperature check via nvidia-smi (NVIDIA cards)
nvidia-smi --query-gpu=timestamp,temperature.gpu,utilization.gpu,power.draw --format=csv -l 5 > baseline_temps.csv
# HWiNFO64 command-line sensor logging (all GPU vendors)
HWiNFO64.EXE /SILENT /SENSORLOGGING /LOGFILE="baseline_before_repaste.csv"
# Linux: AMD/NVIDIA temp snapshot every 5 seconds for 10 minutes
watch -n 5 "sensors | grep -A2 edge; nvidia-smi --query-gpu=temperature.gpu --format=csv,noheader"With the baseline saved, shut the PC down completely and switch off the power supply at the rear toggle, then unplug the PSU from the wall outlet. Don’t rely on the case power button alone; motherboards and GPUs can retain standby voltage even when “off.” Press the case power button for a few seconds after unplugging to discharge any residual charge, then ground yourself with an anti-static wrist strap clipped to bare metal chassis, or touch the case exterior repeatedly while working if you don’t have a strap.
Step 3-4: Remove the GPU and Open the Cooler Shroud
Open the side panel (typically two thumb screws on the rear of the case) and locate the PCIe retention latch at the end of the primary x16 slot. Unclip it, then unscrew the bracket screws holding the card to the rear I/O panel. Disconnect any PCIe power cables from the card before pulling it free, and support the GPU with both hands as you slide it out since modern flagship cards can weigh over a kilogram.
Lay the card on an anti-static mat or a clean, non-carpeted surface with the fans facing up. Photograph the backplate and shroud screw pattern before removing anything; modern cards often mix standard Phillips screws with Torx security screws, and some hide screws under warranty stickers, rubber feet, or beneath the backplate itself. Remove backplate screws first (if present), then shroud screws, working from the outside in. Set screws aside in a labeled tray or a magnetic parts dish grouped by location, since GPU screw lengths are rarely interchangeable between the shroud, PCB, and backplate.
Gently lift the cooler assembly straight up and away from the PCB. If it resists, don’t force it, old dried paste or degraded pads can act like weak glue. Rock the cooler slightly side to side while pulling up evenly, never twist or pry from one corner. Once separated, note the fan and ARGB cable connectors linking the shroud to the PCB and unplug them carefully before setting the cooler aside.
Step 5-6: Strip Old Paste and Inspect Thermal Pads
Dampen a lint-free cloth or coffee filter with isopropyl alcohol and wipe the GPU die clean using light, single-direction strokes rather than circular scrubbing, which can push residue into the die’s edges. Repeat with a fresh section of cloth until no paste residue transfers. Use cotton swabs dipped in alcohol for the die’s edges and any paste that migrated onto surrounding PCB components, but keep alcohol away from connectors, capacitors, and exposed solder joints where possible.
Clean the cooler’s contact plate the same way. If the plate shows visible pitting, discoloration, or a warped surface, that’s a separate issue from paste degradation and won’t be fixed by repasting alone.
Now inspect the thermal pads covering VRAM modules and VRM components around the die. Press a fingertip against each pad: healthy silicone pads feel slightly tacky and springy. Pads that feel dry, crumbly, or leave a chalky residue on your finger have degraded and should be replaced, not reused. This is also the point where you photograph pad placement and measure thickness with calipers if you have them, since VRAM and VRM pads on the same card often use different thicknesses and swapping them in the wrong spot creates uneven contact pressure elsewhere.
Take your time on this inspection step rather than rushing to the fun part of applying new paste. A card with eight or more individual pad zones can easily have two or three that look fine at a glance but fail the touch test, and those are precisely the ones responsible for a VRAM temperature spike that a paste-only repaste won’t fix. Lay each removed pad on a piece of paper in its original position as you go, so you have a physical template for cutting replacements to the exact right size and shape rather than guessing.
Best GPU Thermal Paste and Pads for 2026
Paste choice matters less than people assume once you’re comparing two decent products, independent CPU-load testing in 2026 found roughly a 1.5-2.5°C spread between budget and premium non-liquid-metal pastes. The bigger gains come from fixing bad factory application or pump-out, not from chasing marginal W/mK improvements. Still, for GPUs prone to vertical mounting stress or extended overclocking, the higher-viscosity options genuinely resist pump-out better over multi-year use.
| Product | Type | Conductivity | Typical Price | Best For |
|---|---|---|---|---|
| Arctic MX-6 | Standard paste | Manufacturer doesn’t publish a figure; independent CPU testing placed it near 72°C load vs. 70-71°C for premium pastes | $6-12 (4-8g) | Most gaming GPUs, default pick |
| Thermal Grizzly Kryonaut | Standard paste | 12.5 W/mK | $7-10/g | High-TDP RTX 40/50 and RX 7000/9000 cards |
| Thermal Grizzly Kryonaut Extreme | Standard paste | 14.2 W/mK | ~$24.99 (2g) | Overclocked flagship GPUs |
| Noctua NT-H2 | Standard paste | ~8-9 W/mK (independent testing) | ~$14.95 (3.5g, includes wipes) | Long-term stability, infrequent re-service |
| Honeywell PTM7950 | Phase-change pad | ~8-15 W/mK effective | $15-20/sheet | Vertical mounts, laptops, pump-out resistance |
| Thermalright Odyssey 12.8 | Silicone pad (VRAM/VRM) | 12.8 W/mK | $10-20/set | VRAM and VRM repadding |
| Gelid GP-Ultimate | Silicone pad (VRAM/VRM) | 15 W/mK | $12-25/set | Premium VRAM repadding |
Phase-change materials like Honeywell PTM7950 sit solid at room temperature and melt to flow like a paste once the die warms up under load, then re-solidify when it cools. That behavior is specifically why 2026 thermal guides increasingly recommend them for GPUs that see repeated thermal cycling: unlike conventional paste, they don’t gradually migrate away from the die’s hottest point. If you’re repasting a laptop GPU, a small-form-factor build, or a card mounted vertically in a case, a phase-change pad is worth the extra cost over standard paste.
Step 7-8: Apply New Paste to the Die
With the die and contact plate both clean and dry, apply paste using either the pea-sized dot method (a single dot roughly 3-5mm in diameter centered on the die) or the thin line method for larger, rectangular GPU dies common on flagship cards. Do not spread the paste manually with a card or spatula before reassembly; the pressure from the cooler mounting screws distributes it far more evenly than manual spreading, and pre-spreading risks trapping air bubbles.
# Rough paste volume guide for GPU dies (die contact area in mm^2)
# Rule of thumb: 0.05-0.1mm layer thickness after compression
def paste_volume_mm3(die_length_mm, die_width_mm, layer_mm=0.07):
area = die_length_mm * die_width_mm
return round(area * layer_mm, 2)
# Example: RTX 4080-class die, ~20mm x 18mm
print(paste_volume_mm3(20, 18))
# Output: ~25.2 mm^3 -- roughly a pea-sized dot, err smaller rather than largerUsing too much paste is a more common mistake than using too little. Excess paste squeezes out past the die and can bridge onto surrounding SMD components, and if the paste is electrically conductive (most modern GPU pastes are not, but check the label), that’s a short-circuit risk. Err toward slightly less than you think you need; you can always disassemble and add more.
Step 9: Replace VRAM and VRM Pads
Cut new thermal pads to match the exact footprint of the components they’ll cover, using the old pads as a template if they came off cleanly, or your earlier photos and measurements if they didn’t. Match thickness closely to the original; a pad that’s meaningfully thicker than stock will lift the cooler off the die on that side and create a rocking gap, while a pad that’s too thin won’t bridge the space at all.
Peel the protective film from one side, press the pad onto the PCB component (VRAM module or VRM MOSFET/choke), then peel the second film layer just before you set the cooler back down, so the exposed adhesive doesn’t collect dust while you finish the rest of the pads. Work through each component methodically rather than doing them all at once, since GPUs commonly have 8-16 individual pad locations across VRAM and VRM sections and it’s easy to lose track of which ones are done.
Step 10-11: Reassemble With the Correct Torque Pattern
Reconnect the fan and ARGB cables between the shroud and PCB before you set the cooler down, this connector is easy to forget once the shroud is back in place and screwed down. Lower the cooler straight onto the PCB without sliding it sideways, which can smear the paste dot unevenly before the screws even engage.
# GPU cooler screw tightening sequence (4-screw pattern around die)
#
# [1]-----[2]
# | die |
# [4]-----[3]
#
# Tighten in this order, 1/4 turn per pass, repeat 3-4 passes:
# 1 -> 3 -> 2 -> 4 (diagonal/star pattern, NOT clockwise around the edge)
# Target torque: 0.3-0.6 N*m -- stop as soon as resistance increases,
# do not tighten until the screwdriver clutch or driver stallsTighten screws in a star or diagonal pattern rather than working around the perimeter in a circle, applying a quarter-turn at a time across multiple passes until each screw is snug. This distributes clamping pressure evenly across the die instead of concentrating it on one corner first, which can crack the die or create an air gap on the opposite side. If you have a torque driver, stop at 0.3-0.6 N·m; if you don’t, stop as soon as you feel resistance increase noticeably rather than tightening until the screwdriver stalls.
Reattach the backplate if your card has one, then reinsert any screws you removed from the shroud edges and I/O bracket.
Step 12: Reinstall, Power Up, and Stress Test
Slide the card back into the PCIe x16 slot until the retention latch clicks shut, screw the bracket to the rear I/O panel, and reconnect PCIe power cables from the PSU. Close the case, plug the PSU back into the wall, and boot normally. Confirm the display outputs from the GPU (not an onboard motherboard port) and that Windows or your OS detects the card correctly in Device Manager or equivalent.
# FurMark command-line stress test (15-minute GPU burn-in)
FurMark.exe /width=1920 /height=1080 /msaa=4 /run_mode=1 /max_time=900000 /log_temperature=1 /log_gpu_usage=1
# Log results to CSV for direct comparison against baseline_before_repaste.csv
FurMark.exe /nogui /log_temperature=1 /max_time=900000 /output=after_repaste.csvRun the same 10-15 minute stress test or gaming session you used for the baseline, with HWiNFO64 logging active again. Compare core temperature, hotspot temperature, VRAM temperature, and fan speed directly against your earlier log. A successful repaste on a card that genuinely needed one typically shows core temperatures down 5-10°C and hotspot temperatures down 8-12°C under sustained load, sometimes more on cards with severe pump-out. If temperatures are unchanged or worse, don’t assume the paste is bad, it almost always means a reassembly issue: uneven screw tension, a paste dot that didn’t spread evenly, or a pad placed on the wrong component.
Before vs. After: Real Temperature Results
These figures reflect typical results reported across 2025-2026 repaste case studies and independent testing on cards running stock factory paste that had reached 18+ months of use. Your results depend heavily on how degraded the original paste was and your case airflow, so treat these as realistic ranges rather than guarantees.
| Card | Load Temp Before | Load Temp After | Hotspot Delta |
|---|---|---|---|
| RTX 4090 (stock cooler) | 70-80°C core | 63-67°C core | -8 to -10°C |
| RTX 4080 | 65-75°C core | 58-67°C core | -6 to -9°C |
| RX 7900 XTX | 80°C core / up to 95°C hotspot | 72-75°C core | -8 to -12°C hotspot |
| RX 7900 XT | 70-85°C core | 63-77°C core | -5 to -8°C |
Notice that hotspot temperature typically improves more than core temperature. That’s the direct signature of fixing pump-out: the factory paste had already migrated away from the die’s hottest zone, so a fresh, evenly-applied layer closes that specific gap more than it lowers the overall average.
Common Pitfalls When Repasting a GPU
- Applying too much paste. A dot larger than a pea on a modern GPU die usually squeezes out past the edges under mounting pressure, risking contact with nearby SMD components.
- Spreading paste manually before reassembly. Letting the cooler’s mounting pressure do the spreading gives more even, air-bubble-free coverage than a card or spatula.
- Uneven screw torque. Tightening one screw fully before moving to the next tilts the cooler and leaves one side of the die with a thin paste layer and the other with excess. Always work in a star pattern with gradual passes.
- Reusing degraded thermal pads. A pad that’s gone dry and crumbly won’t make proper contact even after reassembly; it needs replacing, not reinstalling.
- Mismatched pad thickness. Installing a pad thicker than the original lifts the cooler off the die on that section, which can actually make core temperatures worse even while VRAM temperatures improve.
- Skipping the baseline log. Without a documented “before” state, you can’t tell whether a disappointing result means the repaste didn’t help or the reassembly introduced a new problem.
- Breaking a warranty seal without checking the policy first. Some board partners treat any broken sticker as automatic warranty voidance regardless of workmanship.
Troubleshooting: 8 Repaste Problems and Fixes
1. Temperatures are the same or worse after repasting. Almost always a mounting issue. Reopen the cooler, check for uneven paste spread (a good spread leaves a roughly circular, evenly-thin film covering 80-90% of the die), and verify the star-pattern torque sequence was followed.
2. Card doesn’t display any output after reassembly. Reseat the PCIe power cables fully, they can look connected while sitting slightly loose. Confirm the display cable is in a GPU port, not a motherboard port, and reseat the card itself in the PCIe slot.
3. Fans spin but the GPU isn’t detected in the OS. Check the fan/ARGB connector between shroud and PCB wasn’t left unplugged, then check the PCIe slot latch is fully clicked closed; a card that isn’t fully seated in the slot can still spin fans off standby power without establishing a data connection.
4. VRAM temperatures spiked after repadding. The most common cause is a pad cut slightly off-position, missing partial contact with the memory module. Reopen and verify pad alignment against your reference photos.
5. Coil whine or new fan noise after reassembly. Usually unrelated to the paste itself and more likely a cable resting against the fan blades, or a fan cable pinched under the shroud during reassembly. Open the case and visually trace cable routing near the fans.
6. Paste squeezed out onto the PCB. Power down, remove the cooler, and clean the excess with isopropyl alcohol and a cotton swab before it has a chance to affect nearby components. Reduce the dot size before reapplying.
7. A screw won’t tighten or feels stripped. Stop immediately rather than forcing it, a stripped screw hole in a GPU backplate is difficult to repair. Try a slightly larger driver bit for better grip, or in worst cases a screw extractor kit.
8. Hotspot delta is still large (20°C+) even after repasting. This can indicate a warped cooler contact plate rather than a paste problem. Check the plate for visible unevenness with a straightedge; if it’s warped, repasting alone won’t fully resolve the gap, and the cooler itself may need lapping or replacement.
Warranty Rules by Brand: ASUS, MSI, Gigabyte, Zotac
Every major board partner treats cooler disassembly as user-initiated risk to some degree, but the practical enforcement varies. Check your specific card’s warranty terms before opening it, since policies can differ between product lines within the same brand.
| Brand | General 2026 Policy Pattern |
|---|---|
| ASUS | Cooler disassembly is not officially user-serviceable; RMA review typically focuses on tamper stickers and evidence of physical damage rather than the fact of disassembly alone. |
| MSI | Thermal paste/pad replacement is not officially supported; damage discovered during disassembly is grounds for warranty denial. |
| Gigabyte | Similar to MSI; warranty stickers over screws are generally treated as void-if-removed. |
| Zotac | Historically the strictest of the four on user disassembly beyond basic external cleaning. |
Photograph the card in its original, unopened state before starting, and again after reassembly, so you have documentation the card was reassembled correctly and undamaged if a warranty question ever comes up. If your GPU is still within its first year and running fine, weigh the modest performance gain against the warranty risk; a card actively throttling or exceeding safe temperatures is a stronger case for repasting regardless of warranty status.
Advanced Tips: Phase-Change Pads, Vertical Mounts, and Liquid Metal
Once you’re comfortable with a standard paste repaste, a few upgrades are worth considering for specific situations. Phase-change materials like Honeywell PTM7950 or Thermal Grizzly PhaseSheet are increasingly the 2026 recommendation for GPUs mounted vertically or subjected to heavy thermal cycling, since their solid-at-rest, liquid-under-load behavior resists the migration that eventually degrades conventional paste. They cost more upfront but are commonly cited as lasting 5-10+ years without needing another service.
Graphene-based pads, such as Thermal Grizzly’s KryoSheet, are another option worth knowing about for GPU dies specifically, they behave more like a solid conductor than a paste and avoid pump-out entirely, though application requires more precision since they don’t self-level the way a compound does.
Liquid metal compounds deliver the lowest possible die temperatures but come with real risk on GPUs: liquid metal is electrically conductive, can corrode bare aluminum contact plates over time (nickel-plated copper is safe, bare aluminum is not), and a single stray drop near VRAM or VRM components can cause a short. This is an expert-only technique, and most repaste guides steer casual builders away from it on GPUs specifically because the failure mode is a dead card rather than just disappointing temperatures.
After reassembly, it’s worth setting a slightly more relaxed fan curve than the factory default, since a fresh repaste often gives you enough thermal headroom to run quieter at the same load without sacrificing clock speeds. Export your current curve first so you can revert if needed.
# MSI Afterburner fan curve profile (example, post-repaste headroom)
# Save as a .cfg profile before adjusting, so you can revert instantly
[FanProfile]
Point1=30C,20%
Point2=45C,35%
Point3=60C,50%
Point4=75C,70%
Point5=85C,100%
HysteresisC=3
ApplyOnStartup=1For a small-form-factor build where the GPU sits vertically via a riser cable, prioritize either a phase-change pad or a high-viscosity paste like Thermalright TFX or Kryonaut Extreme over a standard compound; both are specifically marketed around pump-out resistance, which is the exact failure mode a vertical mount accelerates.
Laptop GPU Repaste: What’s Different
Repasting a laptop’s discrete GPU follows the same core principles but with tighter tolerances and more disassembly overhead. Laptop coolers frequently share a single vapor chamber or heat pipe assembly across both the CPU and GPU dies, meaning you’re often servicing both chips in the same session whether you planned to or not. Factory paste layers on laptops also tend to be thinner than on desktop cards, since laptop chassis have far less clearance to absorb an uneven or over-applied dot.
Getting to the GPU die usually means removing the entire bottom chassis panel, disconnecting the battery first as a hard safety requirement, and carefully documenting the significantly larger number of screws (often 20-40+, frequently in different lengths) compared to a desktop card. Because of the shared heat pipe design and the risk of disturbing a delicate vapor chamber seal, laptop repastes are exactly where phase-change materials like Honeywell PTM7950 earn their reputation: their pump-out resistance matters more in a chassis that flexes slightly with heat and daily handling than it does in a stationary desktop tower. If your laptop is still under manufacturer warranty, check the terms carefully first, laptop warranty enforcement around chassis disassembly tends to be stricter than desktop GPU policies.
Complete Project Walkthrough: A Full Repaste Case Study
To tie the whole process together, here’s how a full repaste plays out end to end on a two-year-old flagship-class card showing early signs of degraded paste (idle around 48°C, load hotspot climbing to 92°C under sustained gaming, and a fan curve that had gotten noticeably louder over the prior six months).
- Logged a 15-minute FurMark baseline: 78°C core, 92°C hotspot, fans at 68% (2,400 RPM).
- Powered down, unplugged the PSU, and grounded with a wrist strap clipped to the case.
- Removed the card, photographed the 14-screw shroud pattern (10 shroud screws, 4 backplate screws, two different lengths).
- Disassembled the shroud, exposing a die roughly 22mm x 20mm and eight VRAM/VRM pad zones.
- Cleaned the die and contact plate with 99% isopropyl alcohol and coffee filters until no residue transferred.
- Found three of the eight thermal pads dry and crumbly; replaced all eight with Thermalright Odyssey 12.8 pads cut to match the originals.
- Applied a 4mm pea-sized dot of Thermal Grizzly Kryonaut to the die center.
- Reassembled the shroud using the diagonal star pattern, four passes, quarter-turn increments, torque driver capped at 0.5 N·m.
- Reinstalled the card, reconnected PCIe power, and booted successfully on the first attempt.
- Reran the identical 15-minute FurMark test: 66°C core, 79°C hotspot, fans at 51% (1,850 RPM).
- Net result: -12°C core, -13°C hotspot, and a fan noise reduction noticeable from across the room.
- Logged the results and set a reminder to check temperatures again in 18 months.
That 12-13°C swing is on the higher end of what to expect, consistent with a card that had genuinely degraded pads in addition to worn paste, but it illustrates why the pad inspection step matters as much as the paste itself. A repaste that only replaces paste while ignoring crumbling VRAM pads leaves real performance on the table.
For further reading on thermal interface material testing, Tom’s Hardware maintains an extensive thermal paste comparison tested across dozens of compounds, and Thermal Grizzly’s official Kryonaut Extreme page documents the manufacturer’s own conductivity specifications. For a visual step-by-step reference, iFixit’s overheating graphics card repair guide covers the disassembly process with photos, Arctic’s product page for MX-6 lists the manufacturer’s current specifications, and Igor’s Lab regularly publishes independent GPU thermal teardown analysis worth reading before you buy paste.
Frequently Asked Questions
Does replacing GPU thermal paste void the warranty?
Potentially, depending on the brand and whether a warranty seal is broken in the process. ASUS, MSI, Gigabyte, and Zotac all generally treat cooler disassembly as outside official support, though enforcement tends to focus on evidence of physical damage rather than disassembly alone. Check your card’s specific warranty terms before opening it, and if the card is still early in its warranty period and running within normal temperatures, weigh the risk against the benefit.
How often should I repaste a GPU?
Most gaming GPUs under normal use don’t need attention for 2-4 years. Cards running near-constant heavy loads (mining rigs, workstation rendering, 24/7 servers) or mounted vertically may show pump-out symptoms within 12-18 months. Watch temperature trends over time rather than repasting on a fixed schedule; if load hotspot temperatures haven’t crept up and the fan curve hasn’t gotten noticeably louder, there’s no urgency.
Can I use CPU thermal paste on a GPU?
Yes, non-conductive CPU pastes like Arctic MX-6, Noctua NT-H2, and Thermal Grizzly Kryonaut all work identically well on GPU dies. There’s no CPU-specific or GPU-specific chemistry difference in mainstream non-conductive compounds; the only real distinction to watch for is electrically conductive or liquid metal pastes, which carry higher risk near a GPU’s densely packed VRAM and VRM components.
What temperature should my GPU run at after repasting?
Target a load core temperature under roughly 75-80°C for current RTX 40/50 or RX 7000/9000-series cards in a well-ventilated case, with hotspot temperature staying within about 10-15°C of the core reading. VRAM should generally stay at or below 90°C. Exact numbers vary by case airflow, ambient room temperature, and the specific card’s cooler design, so compare against your own baseline rather than a universal target.
Is GPU thermal paste the same as GPU thermal pads?
No. Thermal paste is a compound applied to the GPU die itself, the single largest and hottest chip on the board. Thermal pads are pre-cut solid silicone sheets used on VRAM modules and VRM components, which sit at a different height than the die and need a compressible material to bridge the gap to the cooler’s contact plate rather than a spreadable paste.
How much does professional GPU thermal paste replacement cost?
A basic repaste and re-pad service at a PC repair shop typically runs $40-80, rising to $80-120 if you request premium thermal interface materials like PTM7950 or Kryonaut Extreme. DIY costs land around $20-50 for enough paste and pads to service several GPUs, plus a one-time investment in a precision screwdriver set if you don’t already own one.
Do I need to replace thermal pads every time I repaste?
Not necessarily. Inspect them first: pads that still feel slightly tacky and springy can usually be reused. Pads that feel dry, crumbly, or leave residue on your finger have degraded and should be replaced. Since pads are inexpensive relative to the labor of opening the cooler a second time, most guides recommend replacing them proactively whenever the card is already open.
Will repasting improve my FPS in games?
Indirectly, and only if your card was thermal-throttling before the repaste. Lower temperatures let the GPU sustain its boost clock longer under load instead of stepping down to stay within thermal limits, which can translate into more consistent frame rates during long gaming sessions. A card that wasn’t throttling in the first place won’t see a measurable FPS gain from a repaste, even though temperatures and fan noise will still improve.
Related Coverage
- How to Overclock a GPU: RTX 5080/5090 & RX 9070 XT in 12 Steps [2026]
- Custom Water-Cooling Loop Build: 13 Steps, 4 Hours [2026]
- Intel Crescent Island AI GPU Slips, Nvidia Holds 80% [2026]
- DualSense Edge Stick Module Swap: 12 Steps, $19.99 [2026]
- How to Test RAM for Gaming Crashes: MemTest86 in 12 Steps [2026]
