• State
  • Replies 2 replies
  • Subscribers 98 subscribers
  • Views 49 views
  • Users 0 members are here
Support feedback
Options
Options
Related

Original question:

SK-AM62P-LP: Unexpected CPU Temperature Drops on AM62P During Continuous Real-Time Workloads

SK-AM62P-LP: SK-AM62P-LP: Unexpected CPU Temperature Drops on AM62P During Continuous Real-Time Workloads

Anaswara P
Prodigy 20 points
Part Number: SK-AM62P-LP
Other Parts Discussed in Thread: AM62P

Tool/software:

Hi team,

In my earlier post on SK-AM62P-LP, I had reported unexpected CPU temperature drops (~4–5°C) during continuous workloads (CAN, I2C, SPI, GPIO, LIN, GPU rendering with glmark2).

Previous link:

The previous response clarified:

  • Thermal zones are defined in k3-am62p-j722s-common-thermal.dtsi.

  • The VTM module only triggers shutdown (or throttling if configured), not automatic scaling.

  • There is an ADC quantization delay in thermal readings.

  • No direct equation exists for AM62P temperature vs. CPU load, but AM57x principles apply.

My follow-up concern:


The workloads remain active, CPU frequency is constant, and system load does not change. Still, the CPU temperature drops intermittently by 4–5°C.

I want to understand what the SoC is doing internally in such cases. In any case, we can also plot a graph and see what’s happening internally.

Regards,

Anaswara P

  • 0
    TI__Mastermind 18440 points

    Hi Anaswara,

    Can you help me understand what the SoC is doing when there is a temperature drop? Meaning, is it only the GPU doing work? Are the A53, MCU R5F, DM R5F running any application?

    The typical heat producers for AM62P will by the GPU and the various cores.

    What is LIN in this case? Is it referring to ethernet?

    How is the temperature captured? Does the case and junction temperature both show the temperature drop?

    Thanks,

    Anshu

  • 0 in reply to Anshu Madwesh
    Prodigy 20 points

    Hi Anshu,

    Thanks for the detailed follow-up. Let me clarify:

    Workload distribution:
    In both MCU R5F and A53 sides, we are working with CAN, GPIO, I2C, and UART. Additionally, on the A53 side, we are also using WLAN services, and GPU activity is from running glmark2.

    Heat producers:
    As you mentioned, the GPU and A53 are the primary contributors in our setup.

    LIN:
    By LIN, I mean Local Interconnect Network (automotive LIN bus communication), not Ethernet.

    Temperature capture:
    The readings are from /sys/class/thermal/thermal_zone0/temp. As far as I know, this corresponds to the junction temperature from the VTM sensor. I don’t have direct case temperature readings available.

    The main observation remains that with the workloads continuing (no change in CPU frequency or utilization), the CPU temperature sometimes drops by ~4–5°C. I would like to understand what could cause this behavior internally in the SoC.

    I have already plotted a graph with temperature vs. workload timeline, which is attached below for reference.

    Regards,
    Anaswara