4 ms·

Actually, I just put a used HP ProDesk 600 G3 Microtower (Pentium Gold G4560) into operation, and my energy meter shows 0.0W most of the time (which can't be tr

Actually, I just put a used HP ProDesk 600 G3 Microtower (Pentium Gold G4560) into operation, and my energy meter shows 0.0W most of the time (which can't be true, but apparently it's something <1W) with Ubuntu + TLP installed and the HDD in standby. Here are my TLP settings: TLP_DEFAULT_MODE=BAT TLP_PERSISTENT_DEFAULT=1 # Change some BAT settings back to AC defaults CPU_MAX_PERF_ON_BAT=100 CPU_BOOST_ON_BAT=1 CPU_HWP_DYN_BOOST_ON_BAT=1 # Disable options that are undesirable for servers MAX_LOST_WORK_SECS_ON_BAT=0 WOL_DISABLE=0 DISK_DEVICES="sda sdb" # Leave disk power management to hd-idle ('PM_ON=on' means PM off, obviously /s) AHCI_RUNTIME_PM_ON_BAT=on DISK_APM_LEVEL_ON_BAT="keep keep" DISK_SPINDOWN_TIMEOUT_ON_BAT="keep keep" I use hd-idle for spinning down the HDD (sdb) because the shucked WD "White Label"/Ultrastar He10 apparently doesn't care for APM or Standby Timer settings set through TLP: echo 'HD_IDLE_OPTS="-i 0 -a sdb -i 1800 -l /var/log/hd-idle/hd-idle.log"' >> /etc/default/hd-idle systemctl enable hd-idle.service Besides that, I had to systemctl disable udisks2 because udisks2 wakes up the drives every 10 minutes, no configuration possible. Also, smartd (from smartmontools) shouldn't wake the HDD if it's in standby, but it will reset hd-idle's standby timer, so I tell smartd to run every 24 hours instead of 30 minutes: echo 'smartd_opts="--interval=86400"' >> /etc/default/smartmontools

londons_explore 4y ago
Probably best to measure power in the DC side of the power supply to be sure it's under a watt... Measuring AC power is very hard to do accurately - there could for example be something else in your town injecting a bit of power at some other frequency onto the AC power grid (for example 1000Hz), which your PSU ends up absorbing (due to filters in the input of the PSU). Regular power meters will usually only measure the 50/60Hz component of absorbed and reactive power, and can therefore over/under read. Still good to know some desktops are designed with power consumption in mind!
justsomehnguy 4y ago
> which can't be true Yep, but still not <1W. Even at 20% load the PSU have only 80-90%% efficiency. Just test with something with a low and known power usage, like a LED lamp or something.
- currysausage 4y ago
  Actually, that's exactly what I did and the energy meter will gladly measure a 1W LED bulb. That's why I think consumption is <1W most of the time. At the end of the day, everything <5W or even <10W would be pretty great. BTW, there are very short bursts of 1W~10W every second. And FWIW, after ~24h, the energy meter shows 0.0117kWh. Obviously, these measurements are a far cry from lab precision, but they do give me some confidence that I won't really notice the server on my next energy bill.

londons_explore 4y ago

Probably best to measure power in the DC side of the power supply to be sure it's under a watt... Measuring AC power is very hard to do accurately - there could for example be something else in your town injecting a bit of power at some other frequency onto the AC power grid (for example 1000Hz), which your PSU ends up absorbing (due to filters in the input of the PSU). Regular power meters will usually only measure the 50/60Hz component of absorbed and reactive power, and can therefore over/under read. Still good to know some desktops are designed with power consumption in mind!

justsomehnguy 4y ago

> which can't be true Yep, but still not <1W. Even at 20% load the PSU have only 80-90%% efficiency. Just test with something with a low and known power usage, like a LED lamp or something.

currysausage 4y ago
Actually, that's exactly what I did and the energy meter will gladly measure a 1W LED bulb. That's why I think consumption is <1W most of the time. At the end of the day, everything <5W or even <10W would be pretty great. BTW, there are very short bursts of 1W~10W every second. And FWIW, after ~24h, the energy meter shows 0.0117kWh. Obviously, these measurements are a far cry from lab precision, but they do give me some confidence that I won't really notice the server on my next energy bill.