Why doesn't Task Manager show my GPU temperature?

Microsoft's release notes for the feature say it requires a dedicated graphics card and a graphics driver that supports WDDM 2.4 or higher; integrated graphics are not supported. Check your WDDM version by pressing Win+R, running dxdiag, and looking at Driver Model on the Display tab. Updating the graphics driver usually fixes it.

Can thermal throttling damage my computer?

No, throttling is the protection, not the problem. The processor lowers its frequency and power to keep its temperature safe, and if that is not enough Intel processors shut down automatically to prevent permanent damage. The cost is lost performance while it throttles.

Why do different tools show different temperatures?

Processors have several sensors (per core, package, and sometimes a margin-to-limit value instead of an absolute temperature), and tools label and average them differently. Compare readings from the same tool over time rather than across tools.

How to Check CPU and GPU Temperature (Windows, macOS, Linux)

Windows Task Manager shows GPU temperature on its Performance tab (dedicated GPUs with a WDDM 2.4 or newer driver), NVIDIA cards report it with nvidia-smi -q -d TEMPERATURE, and Linux shows CPU and board sensors with sensors from the lm-sensors package. Neither Windows nor macOS has a built-in screen that shows CPU temperature in degrees; macOS offers sudo powermetrics, which reports thermal pressure and, on supported Macs, sensor readings.

Knowing the number is only half of it. What counts as "too hot" depends on your exact processor, and modern chips protect themselves by slowing down long before damage occurs.

GPU temperature in Windows Task Manager

Press Ctrl+Shift+Esc to open Task Manager.
Open the Performance tab.
Select your GPU in the left column. The temperature appears next to its listing and on its detail page.

Microsoft added this in 2019 and listed the conditions in the release notes: "Showing the GPU temperature in Task Manager is currently only supported if you have a dedicated GPU card," it is shown in Celsius only, and the graphics driver must support WDDM 2.4 or higher. To check your driver model, press Win+R, run dxdiag, and read Driver Model on the Display tab.

Task Manager's CPU page shows utilisation and clock speed but no temperature. For a CPU reading on Windows you need the motherboard vendor's utility, the processor vendor's tool, or a third-party hardware monitor.

NVIDIA GPUs on any OS: nvidia-smi

nvidia-smi is NVIDIA's command-line management tool and runs on Windows and Linux. If the NVIDIA driver is installed, open a terminal and run:

nvidia-smi -q -d TEMPERATURE

Example output (the thresholds differ between cards, so read your own):

    Temperature
        GPU Current Temp                  : 46 C
        GPU T.Limit Temp                  : N/A
        GPU Shutdown Temp                 : 98 C
        GPU Slowdown Temp                 : 95 C
        Memory Current Temp               : N/A

NVIDIA's documentation defines the fields:

GPU Current Temp: core GPU temperature.
GPU Slowdown Temp: the temperature at which the GPU hardware starts reducing clocks to cool down.
GPU Shutdown Temp: the temperature at which the GPU shuts down.
T.Limit: on some GPUs, a signed margin to the maximum operating temperature instead of an absolute value. A reading of 0 C or below means the GPU may be reducing clocks for thermal reasons.

Your own card's Slowdown value is the most useful "too hot" line you can get: it is the vendor's threshold for that exact GPU, not a forum rule of thumb. Plain nvidia-smi with no options prints a summary table that includes the current temperature for each GPU.

Linux: lm-sensors

Install the lm-sensors package with your distribution's package manager (on Debian and Ubuntu: sudo apt install lm-sensors), then run:

sudo sensors-detect
sensors

sensors-detect is an interactive program that scans the system for hardware monitoring chips: sensors built into the CPU, Super I/O chips, and chips on the SMBus/I2C buses. Its manual warns that some probes can cause problems, including bus lockups, and advises against running it on production servers. It also says not to force a detection step that would have been skipped by default unless you know what you are doing. sensors then "is used to show the current readings of all sensor chips."

Example output on an Intel machine:

coretemp-isa-0000
Adapter: ISA adapter
Package id 0:  +52.0°C  (high = +100.0°C, crit = +100.0°C)
Core 0:        +49.0°C  (high = +100.0°C, crit = +100.0°C)
Core 1:        +51.0°C  (high = +100.0°C, crit = +100.0°C)

On Intel CPUs these lines come from the kernel's coretemp driver. Its documentation defines crit as the "maximum junction temperature" (TjMax) and high as the "maximum recommended operating temperature (Tcontrol)", both read from the processor, so they reflect your chip's own limits. AMD processors appear under the k10temp driver instead, labelled Tctl (a control value used for fan control rather than a physical temperature) and, on CPUs that support them, Tdie or Tccd1, Tccd2 and so on for each core complex.

macOS: no CPU temperature readout, but powermetrics

macOS has no built-in app that shows CPU temperature in degrees, and Activity Monitor does not show it. The closest built-in tool is powermetrics, run from Terminal with administrator rights:

sudo powermetrics --samplers thermal -n 1

The thermal sampler "displays the current thermal pressure the system is under": a pressure level, not a temperature. -n 1 takes a single sample and exits. The smc sampler reports information from the System Management Controller, which the man page says "on supported platforms" includes fan speed and temperature sensors. Run sudo powermetrics -h to see which samplers your Mac supports; not every sampler is available on every model.

For most Mac users, thermal pressure is the more useful number anyway: it tells you whether the system is limiting performance because of heat, which is the thing you actually care about.

What temperature is normal?

There is no single safe number, because limits are set per processor model. The figure that matters is the maximum junction temperature, often written Tjunction max or Tjmax. Intel defines it as "the maximum thermal junction temperature that a processor will allow" before internal thermal controls reduce power, and says it "varies per product and usually is between 100°C-110°C." For your exact model, look up the Max Operating Temperature or Tjunction field in the Package Specifications section of its page on ark.intel.com. For other processors, use the manufacturer's specification page for the exact model number.

Intel is explicit that it does not publish typical operating ranges: "We do not provide typical temperature operating ranges for each processor or each core," because they depend on system design and workload. It also says that "being at maximum temperature while running a workload isn't necessarily cause for concern."

A practical reading of this:

Under sustained full load, near the limit: expected on many laptops and small desktops. The chip is using its thermal headroom.
At or near the limit while idle or lightly loaded: a cooling problem. Check fans, dust, and the CPU cooler's mounting.
Sudden shutdowns under load: throttling was not enough and the processor protected itself. Fix the cooling before continuing heavy work.

Thermal throttling, explained

When a processor approaches its limit, it lowers its clock frequency and power draw so it produces less heat. Intel describes this as the processor reducing frequency and power to prevent overheating, which "may cause performance loss." If that still cannot hold the temperature, the chip "will automatically shut down to prevent permanent damage." GPUs work the same way: NVIDIA lists "HW Thermal Slowdown" and "SW Thermal Slowdown" among the reasons nvidia-smi -q reports for reduced clocks.

So throttling shows up as slowness, not as damage. Signs include a game whose frame rate drops after several minutes, or a long export that starts fast and then slows. Watch the temperature and clock speed together during the slowdown: if the temperature sits at the limit while the clock drops, heat is the cause.

If temperatures are high

Clean dust from fans, heatsinks and intake filters, and make sure vents aren't blocked (laptops on a bed or sofa are a common case).
Check that every fan spins under load, including the CPU cooler and the graphics card.
On desktops, an old or poorly applied thermal paste between CPU and cooler raises temperatures; reapplying it is a standard fix.
If the machine is also slow when cool, the cause is probably elsewhere: see how much RAM you need or check whether the system disk is a hard drive.

Frequently asked questions

Is it bad if my CPU reaches 100 °C?: Not automatically. Intel says the maximum junction temperature for its processors usually lies between 100 °C and 110 °C, depending on the model, and that being at maximum temperature while running a workload isn't necessarily cause for concern. The processor reduces its clock speed to stay below that limit. If it sits at the limit while idle, or the computer shuts down under load, the cooling needs attention.
Why doesn't Task Manager show my GPU temperature?: Microsoft's release notes for the feature say it requires a dedicated graphics card and a graphics driver that supports WDDM 2.4 or higher; integrated graphics are not supported. Check your WDDM version by pressing Win+R, running dxdiag, and looking at Driver Model on the Display tab. Updating the graphics driver usually fixes it.
Can thermal throttling damage my computer?: No, throttling is the protection, not the problem. The processor lowers its frequency and power to keep its temperature safe, and if that is not enough Intel processors shut down automatically to prevent permanent damage. The cost is lost performance while it throttles.
Why do different tools show different temperatures?: Processors have several sensors (per core, package, and sometimes a margin-to-limit value instead of an absolute temperature), and tools label and average them differently. Compare readings from the same tool over time rather than across tools.