AnsweredAssumed Answered

turbostat gives wrong package power consumption when CPU is idle

Question asked by vvsteg on Sep 18, 2019

We have two clusters with 7301 and 7551 CPUs in dual-socket servers. One system uses air cooling, another one uses liquid cooling. Both system runs under SLES12 SP3.

 

On both system we observe that in the idle mode turbostat gives the package power consumption for 2 sockets that is larger than the server power consumption reported via IPMI.

 

The example output of IPMI and turbostat for an idle CPU:

 

# ipmitool dcmi power reading ; turbostat --quiet --show PkgWatt sleep 1

Instantaneous power reading: 92 Watts
Minimum during sampling period: 24 Watts
Maximum during sampling period: 368 Watts
Average power reading over sample period: 82 Watts
IPMI timestamp: Thu Sep 19 03:28:03 2019
Sampling period: 00000005 Seconds.
Power reading state is: activated


1.002633 sec
PkgWatt
134.15
67.72
66.43

 

We use quite a recent version of turbostat 19.03.20.

 

turbostat reports on package power consumption for idle mode do not change if the default frequency is changed in the governor.

 

Under load the reported values seem to be correct (at least, they do not contradict IPMI):

 

# ipmitool dcmi power reading ; turbostat --quiet --show PkgWatt sleep 1

Instantaneous power reading: 286 Watts
Minimum during sampling period: 82 Watts
Maximum during sampling period: 360 Watts
Average power reading over sample period: 84 Watts
IPMI timestamp: Thu Sep 19 03:43:40 2019
Sampling period: 00000005 Seconds.
Power reading state is: activated


1.060027 sec
PkgWatt
270.35
137.57
132.78

 

Please give an advise how to correct these wrong behaviour?

Outcomes