I am working with a Raspi Compute Module 4 that can sometimes heat up to 80C and 85C regardless of the CPU frequency.
In fact to keep the temperature down as low as possible I usually have the kernel cpu governor set to "powersave" which normally pins the CPU frequency to the lowest available.
But when the temperature reaches 80C (`vcgencmd measure_temp`) and the soft temperature limit is reached(`vcgencmd get_throttled`) I see that it actually increases the CPU frequency(`vcgencmd measure_clock arm`) to around 1500Mhz ramping down to 1200Mhz as it gets closer to 85C. When it reaches 85C it forces the clock down to the lowest available as expected.
It seems like the throttling logic completely ignores the requested CPU frequency from the kernel, even if that requested frequency is actually much lower. As a result the throttling logic actually makes the temperature rise faster than it would otherwise. My guess is that there is a hard coded assumption that the CPU was probably at 1500Mhz to begin with to get that hot.
This seems like a bug in the firmware. Is there any way to disable the soft temp limit or otherwise work around this problem? I tried `temp_soft_limit` but that doesn't seem to have an effect on the CM4. And from my testing lowering `temp_limit` to 80C seems to create a new hidden soft temp at 75C or so. My guess is that it is always 5 degrees cooler than `temp_limit`.
In fact to keep the temperature down as low as possible I usually have the kernel cpu governor set to "powersave" which normally pins the CPU frequency to the lowest available.
But when the temperature reaches 80C (`vcgencmd measure_temp`) and the soft temperature limit is reached(`vcgencmd get_throttled`) I see that it actually increases the CPU frequency(`vcgencmd measure_clock arm`) to around 1500Mhz ramping down to 1200Mhz as it gets closer to 85C. When it reaches 85C it forces the clock down to the lowest available as expected.
It seems like the throttling logic completely ignores the requested CPU frequency from the kernel, even if that requested frequency is actually much lower. As a result the throttling logic actually makes the temperature rise faster than it would otherwise. My guess is that there is a hard coded assumption that the CPU was probably at 1500Mhz to begin with to get that hot.
This seems like a bug in the firmware. Is there any way to disable the soft temp limit or otherwise work around this problem? I tried `temp_soft_limit` but that doesn't seem to have an effect on the CM4. And from my testing lowering `temp_limit` to 80C seems to create a new hidden soft temp at 75C or so. My guess is that it is always 5 degrees cooler than `temp_limit`.
Statistics: Posted by casey_8638 — Tue Oct 01, 2024 4:54 pm — Replies 3 — Views 89