How can I measure sustained CPU performance on my PC?

OCCT is a good free tool for this because it can apply a continuous CPU load while recording temperatures, clock speeds, power usage and throttling.

A simple test would be:

1. Connect the PC to power, close unnecessary programs and use the power mode you normally use.
2. Let the computer sit idle for about 10 minutes so it starts from a stable temperature.
3. Open OCCT, select its CPU test and set the duration to 30 or 60 minutes.
4. Watch the CPU temperature, effective clock speed, package power and any thermal or power-limit warnings.
5. Compare the readings from the first five minutes with those from the final ten minutes.
6. Stop the test if the computer shuts down, reports errors, or reaches abnormal temperatures and cannot reduce them.

A small clock-speed drop after the initial boost is normal. The important question is whether the speed settles at a stable level or continues falling as the computer gets hotter.

For a more realistic video-encoding test, you can also encode the same long video with HandBrake while monitoring the CPU. Record the average encoding speed near the beginning and end. For gaming, use a game with a repeatable built-in benchmark, since a CPU-only stress test does not include GPU and game-engine limitations.

Keep the room temperature, cooling setup and power settings the same when comparing results. Otherwise, the numbers will not be directly comparable.

For sustained CPU performance, I would use a combination of a real workload and hardware monitoring rather than relying only on a short benchmark.

A simple approach would be:

1. Start HWiNFO in Sensors mode and monitor/log:
   
   
• CPU Effective Clock
• CPU Package Power
• Core/Package Temperature
• CPU Utilization
• Thermal Throttling / Power Limit flags
2. Run a sustained CPU workload for 30–60 minutes.

For a realistic workload, a long HandBrake software encode is a good option because it represents something closer to actual CPU usage than a short synthetic benchmark.

For a more repeatable stress test, OCCT CPU or CPU+RAM can also be used for the same period.

The important part is not only the maximum clock speed. I would compare the first few minutes of the test with the last 10–15 minutes.

If CPU utilization remains close to 100% but the effective clock gradually decreases while temperature or power-limit flags become active, then the CPU is most likely being limited by thermal or power constraints.

I would also record:

• Average effective clock during the test
• Maximum temperature
• Sustained package power
• Any thermal/power throttling events
• Total workload completion time

This gives a much better indication of sustained performance than a single short benchmark score.

For gaming specifically, I would test separately, as most games do not generate the same sustained all-core CPU load and the result can also be limited by the GPU.