What Kills a GPU: Heat, Dust, Fans, and Voltage Stress

A graphics card rarely dies from a single catastrophic event. Most failures build slowly, across years, from conditions the owner never notices. Understanding what kills a GPU means looking past frame rates and into the physical processes that wear down a board: temperature swings, airborne contamination, bearing wear, and electrical stress. Each of those forces attacks a different part of the card, and each one can be slowed with maintenance that costs almost nothing.

A modern graphics card packs a silicon die under 600mm across, stacked memory linked through micro-bumps, and a power stage that steps 12 volts from the supply down to roughly 0.9 volts while delivering hundreds of amps. Every subsystem on that board has a finite service life, and how you run the card decides how quickly it is spent.

How heat damages a GPU over time

The Arrhenius relationship, borrowed from physical chemistry, sets the pace of every chemical reaction inside a chip. Applied to electronics, it produces a blunt rule of thumb: a 10°C rise in operating temperature roughly halves the service life of a component. That single number explains why cooling is not cosmetic.

Electrolytic capacitors feel the effect most. Their liquid electrolyte slowly evaporates, and as it does, equivalent series resistance climbs while the ability to smooth voltage ripple falls. Parts rated for 105°C operation use lower-vapor-pressure electrolyte and stronger seals than 85°C-rated units, which is why a card like the ASUS TUF Gaming GeForce RTX 5080 specifies solid polymer capacitors rated at 105°C for 5,000 hours at maximum temperature. Budget boards that rely on 85°C electrolytics rated for 1,000 to 2,000 hours age faster under the same conditions.

Vendors size the cooling stack to slow that clock. A vapor chamber spreads heat across a wider area than heatpipes alone, which lowers peak die temperature and the gradient between the center of the die and its edges. The RTX 5080 class of cards pairs a large fin stack with three fans and a phase-change pad, so the interface does not dry out. Fin density matters as well: tighter spacing adds surface area but clogs faster with dust, which is why airflow and filtration have to be maintained together. A typical high-end board carries a 3.6-slot heatsink about 348mm long, with fins spaced under 1mm apart.

Heat also works through cycling rather than steady exposure. Going from idle to a heavy load can push the die up 40°C in seconds. The silicon, the solder balls, the substrate, and the PCB expand at different rates, and that mismatch stresses every solder joint. Repeat the cycle thousands of times, and the joints crack. The symptom is usually intermittent: artifacts under load, crashes after the card warms, or a board that only misbehaves once it is hot. Software such as GPU Tweak III, MSI Afterburner, and HWiNFO reports die, hotspot, and memory temperatures, and a card that suddenly runs 5 to 10°C hotter at the same fan speed is signaling that the thermal interface has aged, dust has built up, or a bearing is failing.

Dust, moisture, and contamination

Computer cases are not sealed. A case pulls in air constantly, and that air carries dust, pollen, and moisture. Dust on heatsink fins acts as insulation. A layer about 1mm thick can cut heat dissipation by 20 to 30 percent, which forces fans to spin faster and pull in more dust. In a typical ATX build, 2 to 3 grams of dust collect on a GPU heatsink each year, adding 0.5 to 1.0mm of insulating material that can raise operating temperatures by 8 to 15°C.

Filters do most of the work here. A case with a 120mm intake fan set against a smaller exhaust runs at positive pressure, so air escapes through the gaps instead of being drawn in through them, and dust that would otherwise settle on the card is trapped at the filter. Cleaning those filters every one to three months keeps the pressure balance and the airflow they were designed for. A filter that clogs raises internal temperature across every component, not only the GPU.

The dust that settles on the board itself is worse. Many particles absorb moisture from humid air, and the wet-dry cycle that follows as the card heats and cools can drive dendritic growth, thin metallic filaments that bridge adjacent traces until they short. That damage is permanent, and no cleaning removes it. Coastal air carries hygroscopic salt, and smoke residue holds moisture, so both speed up the process. Conformal coating, a thin polymer film applied over the PCB, seals circuits against moisture and contaminants, and it is starting to appear on premium cards.

Why the fans fail first

Every other part of a GPU is solid state. The fans are the moving exception, and they wear out. Bearing design decides how long they last. Sleeve bearings spin a shaft inside an oil-impregnated sleeve. The oil migrates away over time, friction rises, noise grows, and the fan eventually seizes. Ball bearings trade quiet operation for durability, and a dual ball bearing design often outlasts a sleeve bearing by two to three times. A sealed housing matters too, since particulate ingress is a leading cause of early failure, and an IP5X dust resistance rating tells you the bearing is protected.

Blade geometry and hub size also shape how hard a bearing has to work. A smaller hub allows longer blades, which move the same air at lower RPM and lower bearing load. Reinforced composite blades resist warping over thousands of hours, and balanced blades cut the vibration that shortens bearing life. A card that ramps to 100 percent fan speed at moderate load is a sign that airflow has degraded, and that higher speed wears the bearing faster.

Failure follows a familiar sequence: grinding noise as metal meets metal, then a struggle to hold target RPM, then intermittent operation, then a dead stop. Modern cards throttle clocks when a fan fails, but throttling is damage control, not repair. Running a card with a failed fan degrades the thermal interface and can cook the VRM.

Power delivery stress and voltage transients

The voltage regulator module converts 12 volts down to the 0.8 to 1.1 volts the die needs, at currents that can pass 400 amps. That conversion is not lossless, and the energy lost becomes heat concentrated in the VRM. In a 14-phase design, each phase carries roughly 28 amps at full load. A high-end card running at 90 to 93 percent efficiency at 450W of board power throws off 30 to 40W as heat. When one phase fails, the survivors must carry its share, which raises their current, temperature, and stress, so a single component failure can cascade.

Component choice inside the VRM separates a durable card from a fragile one. DrMOS stages integrate the high-side switch, the low-side switch, and the driver into one package, which shortens the current path and reduces switching loss. Digital controllers regulate voltage with finer steps and respond to load changes faster than analog designs, so the die sees fewer overshoots. Japanese solid capacitors with tight tolerances filter the output, and their longer rated life is one reason premium boards cost more.

Power supplies are not perfectly stable either. Large load steps produce voltage transients, and a quality VRM absorbs them while a marginal one passes spikes or droops through to the die. An 850W 80 Plus Gold or Platinum supply wastes less energy, delivers cleaner voltage, and usually uses better internal parts. A loose 16-pin 12V-2x6 connector adds resistance and heat at the contact, and an uninterruptible supply with line conditioning protects against the surges and brownouts that stress the regulator.

Thermal interface material degradation

The material between the die and the heatsink matters more than its price suggests. Thermal paste fills the microscopic gaps that would otherwise trap air. A layer only 0.1mm thick still does the job. Over two to three years, silicone oil separates from the filler, and thermal cycling pumps the paste outward from the center of the die, leaving dry spots. Thermal resistance can climb 30 to 50 percent, and conductivity can fall from 8 to 12 W/mK to 2 to 4 W/mK, which raises temperatures by 5 to 10°C.

Monitoring is how you know when the interface has aged. If a card that once ran at 75°C under a heavy load now runs at 82°C at the same fan curve, and the fins are clean, the paste is the likely cause. Replacing it means removing the old material, cleaning both surfaces, and applying a thin, even layer. For owners who would rather not disassemble the card, a professional repaste costs far less than a replacement and can restore near-original temperatures.

Phase-change pads avoid that decline. They sit solid at room temperature and soften under heat to conform to the surfaces, holding 3 to 5 W/mK for five to eight years without drying out or pumping away. Several RTX 5080 cards use a phase-change pad for exactly this reason. Applying paste correctly also matters: too much creates an insulating layer, and too little leaves air gaps.

Extending GPU lifespan with a simple routine

Most of the wear described above can be slowed with a short maintenance list. Watch temperatures and investigate sudden increases. Clean intake filters every one to three months and clear heatsink fins with compressed air every six to twelve months. Keep case pressure positive so air leaves through gaps rather than entering through them. In humid or coastal climates, favor cards with conformal coating and sealed bearings. Pair the card with a supply that has headroom, and replace thermal paste after four to five years if temperatures creep up.

None of this requires special tools. A can of compressed air, a filter you remember to clean, and a temperature monitor cover most of it. The parts that fail first, bearings and thermal paste, are also the cheapest to address, and catching them early keeps a small problem from turning into a dead card.

A GPU is a consumable. It will fail eventually, but the schedule is not fixed. A card that dies at three years of neglect can run for a decade with regular care, and the difference comes down to heat, contamination, bearings, and power.