PC Temperatures

How to Stress Test CPU Temperatures With Prime95 and Cinebench

Stress test CPU temperatures with Prime95 and Cinebench, set up HWiNFO64 monitoring, pick a realistic load and read peak temps, clocks and throttling flags.

On this page
  1. Before you start
  2. Set up monitoring
  3. Prime95: choose the right mode
  4. Cinebench: a realistic sustained load
  5. Reading the results
  6. When to stop the test
  7. What counts as a normal result

Before you start

  • Note the room temperature. Room temperature shifts CPU readings roughly one-for-one, so a result from a hot afternoon is not comparable with one from a cool evening.
  • Use stock settings. Test with the BIOS defaults unless you are testing a specific overclock. Turn off any manual overclock, and any PBO or memory profile you are not testing.
  • Close the side panel. Keep the case closed so the airflow matches normal use.
  • Close other programs. Background apps add load and heat that you did not intend to test.

Set up monitoring

  • Open HWiNFO64 in Sensors-only mode, and reset the minimum and maximum values before the test.
  • Track the CPU package or Tctl temperature, the clock speeds, the CPU power draw and the throttle flags.
  • Do not run several monitoring tools at once. Their sensor polling can conflict and skew the results.

Prime95: choose the right mode

Prime95’s torture test has four main modes:

  • Smallest FFTs: stresses the L1 and L2 caches, with high heat.
  • Small FFTs: produces the most heat and power draw. This is the cooling test.
  • Large FFTs: stresses RAM and the memory controller more than the cores.
  • Blend: mixes the other modes.

Small FFTs with AVX, AVX2 or AVX-512 enabled produce the highest and least realistic heat. Untick the AVX options if you want a more typical load. Run the test for 15–30 minutes, long enough for temperatures to settle. A temperature check does not replace a stability test, which can take hours.

Prime95 Small FFT results are a worst case. They do not match gaming temperatures, so do not judge your gaming cooling by them. Watch the VRM temperatures on budget boards during long AVX runs, as the power delivery can run hot.

OCCT’s CPU test is an alternative, and it has temperature-stop limits you can set.

Cinebench: a realistic sustained load

Cinebench is a rendering benchmark that gives a steady multi-core load, which is closer to real work than Prime95 with AVX.

  • In Cinebench R23, set “Minimum Test Duration” to 10 or 30 minutes so the run is long enough to heat the chip.
  • Cinebench 2024 also has a minimum-duration setting. Its default is a single run.
  • A single short run is too brief for a thermal test, because the cooler has not yet soaked up the heat.

Cinebench is the better choice if you want to know how the chip behaves during long real work.

Reading the results

Record these for each test:

  • Peak and average CPU package or Tctl temperature
  • CPU power in watts
  • Clock speed over time, using the effective clock rather than the reported figure
  • Throttle flags, to see whether the chip reduced its clocks because of heat
  • Room temperature

Compare the peak against your chip’s TjMax. See what TjMax means for the limits by family.

Heat soak also matters. Air coolers settle in about 10–15 minutes, while AIO liquid coolers can take 20–30 minutes. A 5-minute run does not show the heat-soaked temperature of an AIO.

When to stop the test

  • Stop if clocks fall sharply while the temperature is at its limit. The chip is throttling heavily.
  • Stop if the system becomes unstable, such as a crash or an error during the run.
  • Stop if the board or case gets alarmingly hot. Check the VRM readings and the airflow.

A chip with a 95°C limit may reach that point during a Prime95 test and still be running as designed. Watch the clocks and stability, not only the number.

What counts as a normal result

  • Ryzen 7000 and 9000 reach 95°C under all-core stress by design. The chip is managing its own boost.
  • Intel 13th and 14th gen chips often hit 100°C with unlimited power limits.
  • A clean 30-minute run does not prove the system is stable. Stability needs longer tests.
  • Stress-test temperatures are a worst case. Games are usually cooler, so check gaming temperatures separately.

For the next steps, see thermal throttling. If the temperatures are high and the system is unstable, see how to tell if a CPU is damaged from overheating.

The complete guide to pc temperatures

PC Temperature Monitoring: Software, Sensors and Safe Temps
Read the guide

Keep exploring

Guides from other topics

All topics →