I have a card with semi-passive cooling (Sapphire RX Vega 64 Nitro+) When the fan turns off, non-zero speed is still reported through hwmon fan1_input. If I'm not mistaken, the same problem was present with my previous card - Sapphire RX 580 Pulse (so I guess it's the same for all cards). On Windows zero rpm was correctly reported for both cards, of course.
Probably a silly question but still: Did you confirm (visually) that the fans are actually turned off? AFAIK on Windows there is extra software running which turns of the fans when the GPU temperature is below a certain threshold. IIRC you need third-party scripts to achieve the same.
Oh and probably AMD developers would like to know which kernel version you are using.
(In reply to Felix Schwarz from comment #1) > Probably a silly question but still: Did you confirm (visually) that the > fans are actually turned off? Yes > AFAIK on Windows there is extra software > running which turns of the fans when the GPU temperature is below a certain > threshold. IIRC you need third-party scripts to achieve the same. On Windows I have no additional software, only AMD driver (downloaded from amd.com, not modded by Sapphire or anything). Actually, I have my own script that controls fans on Linux (because there's no sysfs knobs to tweak the "automatic" fan curve), it just checks the temperature every second and writes desired fan speed to hwmon/pwm1. I tried to manually write '0' to pwm1 (with pwm1_enable set to '1') - exactly the same bug - fan1_input stops changing (when the fan is actually spinning, fan speed is always slightly changing, +/- few rpms), but doesn't show "0" (and, again, the fans are actually stopped) This problem occurs with both "pwm1_enable=2" and "pwm1_enable=1, pwm1=0"
(In reply to Felix Schwarz from comment #2) > Oh and probably AMD developers would like to know which kernel version you > are using. It's here, in "Kernel Version" field in bugzilla. 4.20-rc7