What thermal throttling is
Thermal throttling is a built-in protection. When a CPU reaches its temperature limit, it lowers its clock speed and voltage to cool down. A GPU does the same when it reaches its own limit. The chip keeps running, just slower, until it cools.
GPUs also trim their clocks in small steps as temperature rises, well before they reach the maximum. NVIDIA’s GPU Boost works this way. So a card can lose a little speed before any warning sign appears.
Signs you’re being throttled
- Clock speeds drop under sustained load. Clocks may fall noticeably, or bounce up and down in a sawtooth pattern while the load stays high.
- Frame rates fall after several minutes. The game runs well at the start and slows down later in the session.
- A second benchmark run scores lower. The first run is fast, and the second drops sharply.
- Temperatures are near the limit. CPUs usually reach their limit somewhere around 90–105°C, depending on the chip. Many GPUs begin trimming clocks in the mid-80s°C or higher, depending on the card.
A high temperature by itself does not prove throttling. Check the clock speed and the throttle flags as well.
Check a CPU for throttling
- Open HWiNFO64 in Sensors-only mode.
- Reset the minimum and maximum values, then start the game or stress test.
- Watch the throttle flags:
- Intel: the per-core “Thermal Throttling” Yes/No flag, “Package/Ring Thermal Throttling” and “PROCHOT CPU/EXT”.
- AMD: the “Thermal Limit” percentage. At 100%, the chip is limited by temperature.
- Compare the effective clock with the reported clock. The effective figure is the one that matters.
Intel XTU shows “Thermal Throttling” and “Power Limit Throttling” indicators. ThrottleStop’s Limits window shows the reason for each limit.
Some Intel chips use Thermal Velocity Boost, which drops boost steps above about 70°C. That is designed behaviour, not throttling.
Check a GPU for throttling
- GPU-Z shows a PerfCap Reason. Thrm means the card is limited by heat. Pwr (power) and VRel (voltage reliability) are normal during games and do not mean a cooling fault. Util appears at idle.
- HWiNFO64 shows a “Performance Limit – Thermal” flag for the GPU.
- MSI Afterburner can show the core clock on screen next to the temperature. Watch whether the clock falls as the temperature climbs.
Thermal, power and current limits look alike
A chip can slow down for three different reasons:
- Thermal: the temperature reached its limit.
- Power: the chip hit its power target, such as PL1 or PL2 on Intel, or PPT on AMD. This is common in heavy games, even with cool temperatures.
- Current: the chip hit a current limit, such as IccMax or EDC.
Check the flag that matches the reason, not just the clock. Two other cases are not throttling at all. A laptop on battery drops its clocks for power reasons, and low clocks at idle are power saving.
The two-run benchmark test
To check for heat-driven slowdown, run the same test twice in a row:
- Run Cinebench R23 in a 10-minute loop, or a game’s built-in benchmark.
- Let it finish, then run it again straight away.
- Compare the two scores.
A big drop on the second run suggests the chip is throttling under sustained heat. A similar score on both runs suggests it is not. Check the throttle flags during the second run to confirm.
Normal throttling and a real cooling problem
Some limit-managed behaviour is by design. Ryzen 7000 and 9000 chips can reach 95°C with high clocks as part of their normal boost behaviour. That is not harmful throttling.
A cooling problem is more likely when:
- The chip throttles steadily in games at normal room temperatures
- The clocks drop well below their normal range, and the fans already run at full speed
- The same workload was fine before cooling changed, such as after a dust build-up or a cooler swap
For the causes and fixes, see how to stop a CPU from overheating. For a stress test that brings out throttling, see how to stress test CPU temperatures with Prime95 and Cinebench.