Custom Fan Curves VP66xx it87 inaccurate CPU temp sensor causing fan curves to look off

[philipandag]

Component

Dasharo firmware

Device

Protectli VP6670

Dasharo version

v0.9.0-rc5

Dasharo Tools Suite version

Test case ID

CFC001, CFC002, CFC003

Brief summary

The fans seem to behave differently than defined for the fan curves.

How reproducible

100%, not always apparent, requires observing for some time

How to reproduce

Run the CFC test suite for graphs.

Run a stress test and check the RPM & temperatures manually if preferred https://docs.dasharo.com/unified-test-documentation/dasharo-performance/406-custom-fan-curve/

Expected behavior

The Fans RPM should strictly follow the curves defined for the device.
https://docs.dasharo.com/unified-test-documentation/dasharo-performance/406-custom-fan-curve/

• note, documentation might not be merged yet Test documentation cfc003 fan off docs#966

Actual behavior

The expected fan RPMs cannot be known exactly due to them being physical objects, but should at least look similar to what is expected with some toleration.

The fan RPMs are off, probably due to the temperatures read by the it87 sensor being different than the ones available in the via sensors tool.

Especially visible using the "Performance" curve, where the RPM does not fall below ~1600 RPM despite all the sensors showing the temperatures being slightly below 30 °C.

Screenshots

Additional context

No response

Solutions you've tried

No response

[mkopec]

The expected fan RPMs cannot be known exactly due to them being physical objects

I expect that the manufacturer has a datasheet with a table describing the relation of PWM duty cycle to rotational speed. I'm not sure we'll be able to find them publicly available. I expect it to be very nonlinear and depend on flow impedance (higher impedance = higher speed @ same pwm in most cases). And that's assuming that the fan doesn't have its own onboard speed control IC, adding another point between firmware and the fan speed.

This is why testing by looking only at the RPM won't work well unless we model each fan in each DUT individually (run at the full range of PWM values from 0 to max and note the resulting RPM at each point). That might not be realistic.