sensitivity of thermal sensors

Hi, I am using the Tegra board to test how the core’s frequency/voltage affects the CPU temperature. I change the core’s frequency and read the CPU temperature at each 500 millisecond. Surprisedly the temperature fluctuation can be 3 celsius degree in a 500 millisecond (fan is powered-off). I feel it is too large. Can I trust the reading?

Hi yma1,

Assume what you read are the values of on-die temp sensors. These sensors are very close to CPU/GPU/MEM zones which make them very sensitive to the hotpot temp. Did you measured the input power fluctuation when you change the core’s freq?

Thanks Trumany,

I didn’t have the power fluctuation. It think it is impossible to directly measure the core’s power outside. But I use the power in datasheet and put it in HOTSPOT, a temperature simulator. If the core’s frequency is fixed, the temperature profile from HOTSPOT can match the sensors’ reading.

The large temperature fluctuation is happened only with dynamic voltage/frequency scaling. I follow up a new measurement that changing core’s frequency and reading thermal sensors at each 100 millisecond. The temperature fluctuation is also may be 3 celsius degrees. But for that temperature simulator HOTSPOT, the temperature fluctuation is less than 0.2 degree.

If the frequency is fixed, the temperature fluctuations both from sensor and HOTSPOT are just 0.5 degree in 1 second.

Hi yma1,

It is not suggested to read value when core frequency is unstable, and also as you know the fluctuation in 500ms is not so important.

Hi Trumany,

Thank for your help.