[GRLUG] random loss of local keyboard

[Michael Mol]

On Thu, Jul 5, 2012 at 5:07 PM, John-Thomas Richards <jtr at jrichards.org> wrote:
> On Thu, Jul 05, 2012 at 04:34:05PM -0400, Michael Mol wrote:
>> On Thu, Jul 5, 2012 at 4:25 PM, John-Thomas Richards <jtr at jrichards.org> wrote:
>> > On Thu, Jul 05, 2012 at 04:05:32PM -0400, Michael Mol wrote:
>> >> On Thu, Jul 5, 2012 at 4:03 PM, John-Thomas Richards <jtr at jrichards.org> wrote:
>> >> > On Thu, Jul 05, 2012 at 03:53:06PM -0400, Michael Mol wrote:
>> >> >> On Thu, Jul 5, 2012 at 3:45 PM, John-Thomas Richards <jtr at jrichards.org> wrote:
>> >> >> > On Thu, Jul 05, 2012 at 03:37:35PM -0400, John-Thomas Richards wrote:
>>
>> [snip]
>>
>> >> >> >> Jul  5 14:32:19 auerbach kernel: [19029.452136] [Hardware Error]: Machine check events logged
>> >> >>
>> >> >> This is the line you should really care about. You have failing
>> >> >> hardware somewhere, and the kernel logged it.
>> >> >>
>> >> >> I think this page would set you in the right direction to identifying
>> >> >> what's going on:
>> >> >>
>> >> >> https://en.wikipedia.org/wiki/Machine_Check_Exception#Decoding_MCEs
>> >> >
>> >> > Thanks!  I'm checking it now.
>> >> >
>> >> >> ...but a random stab. The weather's been crazy hot lately. Is that
>> >> >> machine in a well-cooled room?
>> >> >
>> >> > It's in my office...so no.  It's not.  AC is running but I'm still
>> >> > sweating.
>> >>
>> >> Semi-related. Do you have power management configured? I.e. CPU
>> >> underclocked when not in use, GPU underclocked when not in use, that
>> >> kind of thing? If not, that would probably be a good idea for now.
>> >
>> > Yes; I just checked the logs again and saw this:
>> >
>> > Jul  5 15:59:52 auerbach mcelog: Processor 7 heated above trip temperature. Throttling enabled.
>> > Jul  5 15:59:52 auerbach mcelog: Please check your system cooling. Performance will be impacted
> [snip]
>> > Jul  5 15:59:52 auerbach mcelog: Processor 5 below trip temperature. Throttling disabled
> [snip]
>> > (I just installed mcelog in response to your suggestion.)
>>
>> Set your CPU governer to "conservative" or even "powersave" until the
>> heat wave passes. You've got eight cores, you'll be fine. :)
>
> It's currently on "ondemand"; I almost never see it running at full
> speed (2.2GHz) but most often at either 800MHz or 1.3GHz.  Now to find
> out how to change from ondemand to conservative...

There's an entry somewhere under /proc, I expect. Though I usually
just use a widget from XFCE, GNOME or whatever.

>
>> Similarly, find out if your GPU has power settings you can tweak downward.
>
> That's running at Adaptive mode.  Curiously, I'm not running any video
> or graphically demandings apps (an xterm, Chromium, and the nVidia app)
> and it keeps jumped up to the highest performance level.  Not sure
> why.

A lot of apps are starting to tap into the OpenCL backend (so, using
the GPU for more general calculations), so that'll have an impact.
Otherwise, yeah, there are any number of reasons Chromium might tax
your GPU.

Are you using a compositing window manager? Compositing window
managers give apps the continual perception of their being visible.
Consequentially, they draw updates far more often than they do when
invisible, and the compositing window manager directs these draw
operations into a texture. Then the compositing window manager will
have the GPU continually blend each app's "texture" into its region of
the screen. So, app redraws (even when minimized) may also cause load
spikes for your GPU.

I'd probably see about pinning the GPU to a lower performance mode.
Your GPU is going to consume more energy (and thus generate more heat)
than any other component in your system.

You might also turn down the brightness controls on your LCD or LED
display. (So, via the controls on the display, not via your computer's
software.) My 22" LCD screen's PSU label indicates a 200W draw. I get
huge gains from, e.g. my cell phone by turning down the backlight
brightness. The same would likely apply to a desktop monitor.

One last place you might be able to get some gains...configure your
system to spin down your hard drives when they're idle. The gain from
this will only be about 1-5W per drive, though, so YMMV.

-- 
:wq