GPU Temperature

GPU temperature is the reading from your graphics card's main sensor, and in games most desktop cards settle between 65 and 85 °C. Laptop graphics often run in the high 80s. The number that matters is your card's maximum, set by NVIDIA or AMD: as the card nears it, it lowers its clock speed to protect itself, and you lose frames. To check yours, open Task Manager's Performance tab or a tool such as HWiNFO, then run the gpu stress test for five minutes and note the highest steady reading. If it climbs to the limit, or the frame rate falls as the card warms up, the fixes below bring it down.

Quick checks

• Look at the fans under load. Many cards stop their fans completely at idle and start them at about 50 to 60 °C. That is normal. Fans that stay still at 80 °C are not.
• Read it under a real load. Five minutes of a demanding game or the stress test shows the steady temperature. A reading taken in the first minute is too low.
• Check the room and the case. A closed cabinet, a case pushed against a wall or a summer room adds several degrees.
• Close what you are not using. Video tabs, wallpaper engines and background apps that use the GPU keep it warm even when you are not gaming.

Normal GPU temperature range

Typical readings for a card in good order, in a room at about 20 to 25 °C:

Cards with a small cooler, a blower style cooler or a factory overclock sit at the warmer end. Liquid cooled cards sit at the cooler end.

Maximum GPU temperature by card

NVIDIA lists a Maximum GPU Temperature in the full specifications of each GeForce card, for its own Founders Edition and reference designs. Cards from other makers can differ slightly.

AMD Radeon cards report two main readings: the GPU temperature and the junction temperature, which is the hottest point on the chip. AMD's statement on the RX 5700 series explains why the junction can read 110 °C in normal gaming: the card keeps raising its clocks until any one of its many sensors reaches the limit. Intel does not list a maximum temperature on its Arc graphics specification pages.

How to check GPU temperature

Task Manager (Windows 11 and 10): press Ctrl + Shift + Esc, select Performance and then your GPU. A dedicated graphics card shows its temperature next to its name, in °C only. It needs a driver that supports WDDM 2.4 or later, which any current driver does. Built-in graphics show no temperature here.

Inside games:

• NVIDIA app: press Alt + Z to open the overlay and set up the statistics view, then Alt + R toggles it during play.
• AMD Software: Adrenalin Edition: search for Metrics, open Performance Metrics and turn on Show Metrics Overlay. Ctrl + Shift + O toggles it. It can show GPU Current Temperature and GPU Junction Temperature.

For every sensor: HWiNFO and GPU-Z list the GPU temperature, the hot spot where the card reports it, and on cards with fast GDDR6X memory, the memory junction temperature, which often runs hotter than the GPU itself.

Mac: macOS has no temperature readout. Apple silicon reports thermal pressure instead: run sudo powermetrics --samplers thermal in Terminal, where Nominal is normal. Linux: nvidia-smi shows the temperature of NVIDIA cards, and sensors shows AMD's edge, junction and memory readings.

GPU temperature too high: fixes

1. Cap the frame rate. A card rendering 300 frames per second in a menu works flat out for nothing. Set a frame limit in the game, or use Max Frame Rate in the NVIDIA app or Radeon Chill in AMD Software. Capping a little below your screen's refresh rate, which what is my refresh rate tells you, saves heat with no visible loss.
2. Clean the cooler. Shut down, then blow compressed air through the card's fins in short bursts while holding each fan still with a finger, so it does not spin too fast. Clean the case dust filters too.
3. Improve case airflow. Most cards pull air from below and push it into the case. Give the card space below it, use front intake fans and a rear or top exhaust, and keep the side panel on, since it guides the airflow.
4. Set a fan curve. MSI Afterburner works with NVIDIA and AMD cards, and AMD Software has fan tuning under Performance, Tuning. A steeper curve trades noise for temperature.
5. Lower the power limit or undervolt. Lowering the power limit to 80 or 90% in MSI Afterburner usually costs only a few percent of frame rate and cuts heat noticeably. AMD Software has an Undervolt GPU preset under Tuning. NVIDIA cards can be undervolted with Afterburner's curve editor. Test each change with the stress test and a long game session.

On a laptop, the graphics share one cooling system with the processor, so the fixes in laptop overheating apply to both.

GPU temperature while gaming

A browser test cannot load a graphics card as heavily as the most demanding games, so for the hottest case, play your heaviest game for 15 minutes with an overlay showing temperature, clock speed and frame rate. A healthy card climbs for the first few minutes, then holds steady. A card that keeps climbing until it reaches its maximum and then loses clock speed is thermal throttling, and the frame rate drops with it. The fps test helps you see what the drop looks like on screen. If the processor is also running hot, check the cpu temperature guide, because a hot graphics card heats the whole case.

When it is a hardware fault

Look for these signs:

• A fan that does not spin, rattles or grinds under load. Replacement fans exist for many cards, but under warranty, the maker should repair it.
• A hot spot far above the main reading, 30 °C or more apart, or a card that reaches its limit within seconds of starting a game. The thermal paste has dried or pumped out, or the cooler has lost contact with the chip.
• Artifacts, driver crashes or black screens that start only once the card is hot. The gpu test catches drawing errors as the card heats up.

On a card still under warranty, contact the maker before opening it, since replacing the paste and pads yourself can affect the claim. Out of warranty, fresh thermal paste, and new thermal pads of the correct thickness for the memory, often bring an older card back to its original temperatures.

GPU temperature questions