Clarification on Valid Initial CPU Temperature Range on iMX8QXP MEK

[Ram2]

Hi NXP Team,

I hope this message finds you well.

We are currently validating thermal behavior on the i.MX8QXP-C0MEK platform as part of our startup diagnostics. During testing, we observed that immediately after powering on the system (cold boot, ambient temperature ~25°C), the CPU temperature reported by /sys/class/thermal/thermal_zone0/temp is around 30–34°C.

We would like to confirm:

1.  Is this range (30°C to 34°C) considered valid and expected for initial junction temperature, as measured by the TMU (Thermal Monitoring Unit)?

2.  Could you share any official documentation or SoC-level reference that defines expected thermal startup behavior or typical idle CPU temperature values?

3.  Where can we find detailed guidelines for TMU calibration or accuracy specifications on the i.MX8X family?

We assume the TMU reads the internal die temperature (not ambient), and that it is normal for the CPU to report slightly elevated values on first power-up due to internal clocking and voltage activity.

Looking forward to your clarification so we can align with NXP's thermal expectations.

[Ram2]

Hi @JorgeCas 

Any update on this ?

Please can you reply waiting for your inputs ?

[JorgeCas]

Hello,

I was waiting for your confirmation, I apologieze for the delay.

In my MEK board, the temperature reading was 26°C.

You may need to check if the heatsink is correctly installed on the board. 

Also keep in mind that I'm using a B0 version with Linux 5.15.52.

The software version and silicon version could cause a variation in this reading.

Best regards.

[Ram2]

Hi @JorgeCas 

Thank you for your response and for testing on your MEK board.

We acknowledge your observations on the B0 silicon + Linux 5.15.52 environment where the CPU temperature at boot was 26°C.

We'd like to highlight that:

• We are using i.MX8QXP C0 silicon with Linux 6.1.3 (Yocto Nanbield).

• On cold boot, we consistently observe CPU temperature in the range of 30–34°C, as reported by /sys/class/thermal/thermal_zone0/temp.

We will further reinspect the heatsink assembly and thermal contact quality on our hardware to rule out any dissipation issues.

Additionally, to help isolate the cause, could you kindly confirm:

 Whether the thermal trip thresholds and factory calibration values differ between B0 and C0 revisions?

Are there any known thermal behaviour changes or fixes introduced between Linux 5.15.x and 6.1.x for the thermal drivers or TMU sensor?

This will help us align our expectations and rule out software-based causes in the thermal recovery issue we’re still analysing.

Looking forward to your inputs.

[JorgeCas]

Hello,

We will further reinspect the heatsink assembly and thermal contact quality on our hardware to rule out any dissipation issues.

Please let me know the results.

Do you have other boards to test and compare if the result is the same?

Whether the thermal trip thresholds and factory calibration values differ between B0 and C0 revisions?

The thermal threshold is the same for both silicon version but the factory calibration values is different for each IC, is defined in the factory after a test.

Are there any known thermal behaviour changes or fixes introduced between Linux 5.15.x and 6.1.x for the thermal drivers or TMU sensor?

There are changes in newer driver version but are documented in repository.

Best regards.

[JorgeCas]

Hello,

1.

We do not have an expected temperature value a powering on the system since this value can change by several factors such as BSP version, etc. I can try it with our prebuilt BSP and compare it with your board.

2.

No, we do not have documentation related to CPU temperature values.

3.

The temperature read out does not reflect case or ambient temperature. It reflects the temperature in proximity of the sensor location on the die. Temperature distribution may not be uniformly distributed; therefore, the read-out value may not be the reflection of the temperature value for the entire die.

This calibration is done with values defined in factory and does not need to be modified.

Best regards.

[Ram2]

Hi @JorgeCas 

Thank you for your detailed clarification regarding the TMU behaviour and factory calibration.

We understand that the CPU temperature reflects the internal die-level thermal state and may vary depending on the BSP and runtime conditions. Your confirmation that the TMU is factory-calibrated and not user-adjustable helps us validate our sensor path.

As a follow-up, we are currently observing a thermal behaviour concern during runtime:

---

Thermal Concern We’re Facing:

• On the i.MX8QXP MEK, when our HMI application runs continuously, the CPU temperature rises steadily to 85–90°C

• In response, our software actively disables high-load features such as:

  • Bluetooth and Wi-Fi services

  • HMI switches to a reduced mode (lower rendering load)

• However, despite this effort, the temperature does not significantly fall

  • Only drops by ~5% at most

  • CPU continues running hot even after load reduction

 We'd Like to Request:

• Based on your past experience with i.MX8X family, could you please share possible causes for CPU temperature not falling as expected?

• Are there any recommended best practices or SoC-specific limitations (e.g., TMU polling delay, cooling map limitations, low-power states, DDR activity) that might influence thermal recovery?

• Is there a way to verify whether the thermal governor and passive cooling (via frequency scaling or GPU throttle) are working properly?

We have validated that:

• The passive trip point is defined at 65°C

• The CPU governor is set to ondemand

• Cooling devices are mapped correctly in the DTSI and visible under /sys/class/thermal/cooling_device*/
  
  As suggested, it would be helpful if you could please compare the initial CPU temperature values from your side using the official prebuilt BSP on the i.MX8QXP MEK. This would help us validate that our setup is thermally aligned with reference behavior.

Your suggestions or feedback based on your validation tools or internal insights would be highly appreciated, especially as we’re approaching a release deadline.

Thank you once again for your continued support

[JorgeCas]

Hello,

The actions taken were correct to reduce the load of the system, those are the recommended modifications to reduce power consumption and temperature.

You also could try to reduce overall processor frequency. 

This reported temperature is directly related to the processor's load.

Could you please share the part number and labels in BSP? These temperatures are abnormal.

Is there a way to verify whether the thermal governor and passive cooling (via frequency scaling or GPU throttle) are working properly?

There is not a flag or similar to check if is working property but let's compare them with prebuilt BSP to discard hardware issues.

As suggested, it would be helpful if you could please compare the initial CPU temperature values from your side using the official prebuilt BSP on the i.MX8QXP MEK. This would help us validate that our setup is thermally aligned with reference behavior.

Regarding this request, in my MEK board, the temperature reading was around 26°C.

Best regards.