Some simple test results (kernel 5.2-rc5 1000 Hz tick rate ("lowlatency") and 250 Hz tick rate ("generic"):
Idle system (and by "Idle", I really mean it, as my test server has no GUI and I disabled a bunch of services):
250 Hz kernel: Processor package power: 3.70 watts ( - just a little).
1000 Hz kernel: Processor package power: 3.70 watts ( + just a little).
Note: I would have to do a longer test to be certain, but say 10 milliwatts difference (maybe 0.3%).
pipe-test on 1 core (meaning using both CPUs on the one core, instead of across two cores) X 2 instances. i.e. two cores active, and two idle.
Why this test? Because it spends a lot of time in idle state 0 (polling), which has had some changes in the last 1/2 year.
250 Hz kernel: Processor package power: 40.4 watts ( - just a little) and performance of 4.90 uSec per loop.
1000 Hz kernel: Processor package power: 40.4 watts and performance of 4.95 uSec per loop.
O.K. so that was unexpected, but the 250 Hz kernel seems to go into and out of idle state 0 about twice as often.
The above tests were done using the intel_pstate CPU frequency scaling driver and the powersave governor and the teo idle governor.
Some additional pipe test numbers for the 250 Hz kernel:
intel_cpufreq driver (i.e. intel_pstate in passive mode) ondemand governor, teo idle governor: 32.3 watts 5.5 uSec per loop.
intel_cpufreq driver, ondemand governor, menu idle governor: 32.2 watts 5.6 uSec per loop.
Why such a big difference? I think (have not yet proved) because the intel-pstate driver in active mode (I do not have HWP) includes idle state 0 in its busy calculation, but passive/ondemand does not, and so it scales the frequency down a bit.
For reference, the main turbostat command used for these tests:
Code:
sudo turbostat --quiet --Summary --show Busy%,Bzy_MHz,PkgTmp,PkgWatt,GFXWatt,IRQ --interval 60
Note: My system never hits thermal throttling triggers. actually, thermald was disabled.
Bookmarks