vt(4) not power friendly? (was: High CPU temperature and high fans level)
demelier.david at gmail.com
Thu Aug 18 21:45:31 UTC 2016
On 20/07/16 15:20, Ian Smith wrote:
> On Wed, 20 Jul 2016 12:46:32 +0200, David Demelier wrote:
> > Le 18/07/2016 15:41, Ian Smith a écrit :
> > > In freebsd-questions Digest, Vol 632, Issue 8, Message: 21
> > > On Sun, 17 Jul 2016 11:59:32 +0200 David Demelier
> > > <demelier.david at gmail.com> wrote:
> > > > 2016-07-17 0:25 GMT+02:00 Polytropon <freebsd at edvax.de>:
> > > > > On Sun, 17 Jul 2016 00:06:07 +0200, David Demelier wrote:
> > > > >> Hello,
> > > > >>
> > > > >> I was trying FreeBSD 10.3 on my laptop (hp probook 4510s) and was surprised
> > > > >> to see high CPU temperature and fans running high.
> > > > >>
> > > > >> No apps running, I get a temperature of 57C in dev.cpu.x.temperature and
> > > > >> fans run high (not able to get rpms).
> > > > >>
> > > > >> On a 4.6.3 Linux distro I get an average of 48C and fans are quite low.
> > > > >>
> > > > >> Both tests were kept in tty. No Xorg running just a boot and user login in
> > > > >> console.
> > > > >>
> > > > >> Do you have any clue?
> > > > >
> > > > > Did you enable powerd? It can slow down the CPU when the system
> > > > > is idle, and increase the CPU speed when needed. This should have
> > > > > an effect on CPU temperature and fan speed.
> > >
> > > > Yes, I had powerd enabled, I tried -a adaptive, -a hiadaptive as
> > > > suggested by Erich but it seems that only -a min has some little
> > > > effect. I could get a temperature of 52C. I've tested back on Linux
> > > > and I got an average much lower (41C).
> > >
> > > We really need to see what speed the CPU is running at when idle.
> > >
> > > I think the fans running high - presumably from the sound and airflow? -
> > > rules out the sort of sensor errors Arthur reported (ie below ambient :)
> > > and I assume the box actually feels warmer .. 57C suggests a busy CPU or
> > > two .. but then it is summer there; what background ambient temp. range?
> > We have air-conditioning at work so ambient temperature is normal, somewhat
> > between 24-28.
> Ok, just checking. So the temperatures are indeed obviously excessive.
> > > > By the way the other sensors in hw.acpi.tz* are also much higher than
> > > > Linux (using lm_sensors). The highest value is my tz5 which is at 78C
> Isn't that tz2? tz5 is only 55C in your listing (and not active), but
> tz2 is indeed 78C and working the fan moderately hard.
> > > > almost 5 seconds after boot while the maximum tz value in Linux
> > > > sensors is 55.
> > > >
> > > > I have no idea what's wrong. :(
> > > % sysctl hw.acpi
> > > % sysctl debug.acpi
> > > % sysctl dev.est # assuming intel? if not, maybe dev.hwpstate?
> > > % sysctl dev.cpu # best while idle, maybe plus when busier
> > Thanks for your answer, here I posted the output of the sysctl variables you
> > asked for:
> > http://markand.fr/files/result.txt
> Thanks, that's a useful format. Well, a couple of things ..
> > I've ran them on a FreeBSD memstick, I needed to install a Linux distro until
> > I can find a solution because this drains my battery a lot.
> But did you start powerd after the memstick boot? From the CPU speed,
> assuming it was generally idle, I suspect not?
> The big surprise is that CPU frequency (at least when you asked) is at
> maximum (except for Turboboost mode) and that it's only using C1 state
> when halted. C2 and C3 states provide a huge win for power consumption
> - and so proportionally less heat. Alexander takes it to the limit at:
> hw.acpi.cpu.cx_lowest: C1 # the 'master setting' for cx_lowest
> # ==> sysctl dev.cpu
> dev.cpu.1.cx_usage: 100.00% 0.00% 0.00% last 14427us
> dev.cpu.1.cx_lowest: C1
> dev.cpu.1.cx_supported: C1/1/1 C2/2/1 C3/3/162
> dev.cpu.0.cx_usage: 100.00% 0.00% 0.00% last 13756us
> dev.cpu.0.cx_lowest: C1
> dev.cpu.0.cx_supported: C1/1/1 C2/2/1 C3/3/162
> dev.cpu.0.freq_levels: 2101/35000 2100/35000 1600/23888 1200/15000
> dev.cpu.0.freq: 2100
> Also noted that it's running on battery (which is good for this purpose)
> hw.acpi.acline: 0
> hw.acpi.battery.info_expire: 5
> hw.acpi.battery.units: 1
> hw.acpi.battery.state: 1
> hw.acpi.battery.time: 91 # can likely be doubled of better
> hw.acpi.battery.life: 79
> > The average temperature was 50C, a bit less than the installed version on
> > hard drive where I have seen higher values. This is interesting. I will
> > double check if something else makes the CPU more busy.
> Well you need to duplicate basic conditions when booted from memstick;
> after boot, as root you should be able to:
> # service powerd onestart
> # sysctl hw.acpi.cpu.cx_lowest=Cmax
> and then observe dev.cpu.0.freq and dev.cpu.0.cx_usage ..
> Though whenever you plug it in, or unplug it, you'll have to set the
> sysctl again, unless you update the settings in /etc/rc.conf (possible?)
> Which should a) have it drop back to 1200 MHz and b) allow it to use C2
> and probably C3 .. you might check dmesg for any mention of 'C2' or 'C3'
> as certain combinations of chosen timecounter can limit C3 or even C2
> use, and will say so (usually to do with use of the TSC as timecounter)
> On 9.3 one still had to explicily set these:
> !grep cx /etc/defaults/rc.conf
> performance_cx_lowest="HIGH" # Online CPU idle state
> economy_cx_lowest="HIGH" # Offline CPU idle state
> !grep cx /etc/rc.conf
> But on head sources from a couple of months ago:
> !grep cx /usr/head/src/etc/defaults/rc.conf
> performance_cx_lowest="C2" # Online CPU idle state
> economy_cx_lowest="Cmax" # Offline CPU idle state
> So you might want to check what is there for 10.3? Setting both 'Cmax'
> (or at least to C2) should be safe, the head defaults above are likely
> more conservative for a few boxes that aren't happy with C3 and higher.
> If that works, with powerd running CPU at 1200 MHz, it should save lots
> of power and run plenty cooler. Good luck!
> Warner Losh has suggested disabling Turbo mode on these if having heat
> problems, by adding (in your case) '-M 2100' to powerd_flags, but that
> might be something for later. Meanwhile go for C2 and C3+ if possible.
> As for fans, tz2 and tz0 are the only ones with 'active' cooling, though
> tz2 is the only one also with passive cooling, so is most probably the
> CPU/s - see acpi_thermal(4).
> This one doesn't turn on passive cooling (ie slowing the CPU) until only
> 3C below critical shutdown temperature, which seems a bit close to me.
> At that moment it was above 45C and below 62C, so running at level 3.
> hw.acpi.thermal.tz2._TSP: 300
> hw.acpi.thermal.tz2._TC2: 2
> hw.acpi.thermal.tz2._TC1: 1
> hw.acpi.thermal.tz2._ACx: 84.0C 74.0C 62.0C 45.0C 38.0C 30.0C -1 -1 -1 -1
> hw.acpi.thermal.tz2._CRT: 108.0C
> hw.acpi.thermal.tz2._HOT: -1
> hw.acpi.thermal.tz2._PSV: 105.0C
> hw.acpi.thermal.tz2.thermal_flags: 0
> hw.acpi.thermal.tz2.passive_cooling: 1
> hw.acpi.thermal.tz2.active: 3
> hw.acpi.thermal.tz2.temperature: 53.0C
> This fan is also active at level 3 (0-based, so above 72C, with 3 higher
> speeds to go). It can be hard to tell what TZ matches what 'device'.
> hw.acpi.thermal.tz0._TSP: -1
> hw.acpi.thermal.tz0._TC2: -1
> hw.acpi.thermal.tz0._TC1: -1
> hw.acpi.thermal.tz0._ACx: 105.0C 96.0C 87.0C 72.0C 64.0C 57.0C 30.0C -1 -1 -1
> hw.acpi.thermal.tz0._CRT: 108.0C
> hw.acpi.thermal.tz0._HOT: -1
> hw.acpi.thermal.tz0._PSV: -1
> hw.acpi.thermal.tz0.thermal_flags: 0
> hw.acpi.thermal.tz0.passive_cooling: 0
> hw.acpi.thermal.tz0.active: 3
> hw.acpi.thermal.tz0.temperature: 78.0C
> That one seems pretty high. I'd suspect the GPU(/s?) but you're not
> running X, so hmmm .. definitely not CPU though.
> cheers, Ian
Sorry for this late response, I was in vacation.
I've spent some time into this problem again, and I think I have found
the culprit. I first disabled completely the new console driver vt(4) to
use the old syscons and the CPU coretemp temperature immediately fell.
Running with vt(4) and radeonkms loaded: average of 47°, no X, just a
plain tty with nothing running.
Running with syscons: average of 38°.
That's why I've CC'ed to freebsd-x11@,
Is there power issue in the new vt(4) driver? I've also tried with
textmode with the same results.
More information about the freebsd-x11