Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Chrysalis]

This is really interesting, the Power Preference slider makes a HUGE difference, but not quite how I would expect.

If I crank it all the way to efficiency it doesn't seem to let any core boost above 1600Mhz, even under load.  This halves OpenVPN throughput, kinda sucks as I had hoped it would just boost a single core.

But if I put it right in the middle the cores drop to between 1200Mhz and 1600Mhz and boost to 2922Mhz under load, but ALL cores boost when the traffic is going over OpenVPN, which I shouldn't think is necessary.    Idle temp has dropped 3-4C though.

Using sysctl -a dev.cpu | grep 'freq:' to show core speed.

I noticed after I did a CMOS reset the intel speed shift settings now appear on this unit.

Changing to core level control should from the description allow independent core clock speeds.

But what I can tell on my unit, the clock speeds behave the same no matter where the slider is, whether its 50 or 100.  Its much more aggressive than powerd, the cores never go near idle speed and are typically moving between about 1.3ghx and 1.8ghz.  Setting it to 0 does lock the cores at max clock speed though, so maxing it out does something at least.

--

Did some testing using openssl bench and observing live clock speeds.  All done on core level control.

With it set to 70 I got almost max clock speeds, and the result was as expected for the N100.  Like you said however I observed max clocks or close to max clocks on all cores.  As far as I know its a single threaded test.
75 they drop rapidly down to about 2500mhz, at 80 further down to about 2000mhz, 85 is not much different to 80.

All from 50-100 seem to have similar idle behaviour, where clocks are quite high for an idle load. Runs about 10C hotter than idle Windows OS, however running on 0 does impact idle clocks as it idles at max clocks.  I didnt check anything between 5 and 45.  So a value of 70 seems about optimal if you want the peak performance to be able to hit over 90% of what the CPU can do.

--

Looking at the temp graphs, moving from 50 to 70 in itself is making no measurable difference on idle (including the change from package to core), enabling C2 states (I wont go higher then that as bit dodgy in FreeBSD), knocks about 4C of the idle temps so that one does help.

'sysctl hw.acpi.cpu.cx_lowest=C2' add to /etc/sysctl.conf.local to make permanent (or add to system tunables page).

Disabling hardware p-states and and handing back control to powerD gives similar temps to windows about 10C lower.  But with a hit to interactive performance. (which would be browsing the UI, it shouldnt hurt things like VPN usage).