OnDemand Speed Governor crashes Core2 on Abit AB9Pro - Hardware

This is a discussion on OnDemand Speed Governor crashes Core2 on Abit AB9Pro - Hardware ; I just tried the OnDemand speed governor on my Abit AB9 Pro with a Core2 E6700. When I built this system last fall I noticed that it would crash if the OnDemand governor was used so I switched to the ...

+ Reply to Thread
Results 1 to 9 of 9

Thread: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

  1. OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    I just tried the OnDemand speed governor on my Abit AB9 Pro with a Core2
    E6700. When I built this system last fall I noticed that it would crash
    if the OnDemand governor was used so I switched to the Performance
    Governor which has been stable (this system has been up 24/7 for 5
    months). I just tried the OnDemand governor again hoping that the
    2.6.20.x kernel had fixed it, it didn't. My system crashed within two
    hours. It appears that this is an IO system problem, unfortunately there
    is nothing in /var/log/messages and I didn't take a picture of the error
    messages on the console.

    Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    problem.

  2. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    General Schvantzkoph writes:

    > I just tried the OnDemand speed governor on my Abit AB9 Pro with a Core2
    > E6700. When I built this system last fall I noticed that it would crash
    > if the OnDemand governor was used so I switched to the Performance
    > Governor which has been stable (this system has been up 24/7 for 5
    > months). I just tried the OnDemand governor again hoping that the
    > 2.6.20.x kernel had fixed it, it didn't. My system crashed within two
    > hours. It appears that this is an IO system problem, unfortunately there
    > is nothing in /var/log/messages and I didn't take a picture of the error
    > messages on the console.
    >
    > Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    > Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    > problem.


    I'm getting total lockups with the ondemand governor with Core2 on an
    nvidia 680i board. I haven't been able to determine any exact cause
    of the problem as it will suddenly freeze the machine completely
    without leaving any trace of what happened.

    I've disabled speedstep entirely instead since it didn't make much of
    a difference to the CPU temperature anyway.

    --
    Måns Rullgård
    mans@mansr.com

  3. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    General Schvantzkoph writes:
    >Is anyone using OnDemand with a Core2 Duo?


    Yes, a friend of mine is using it on a machine with an E6600 and a 945
    chipset. It runs stable, as far as we have noticed (the machine is
    not used much under Linux).

    - anton
    --
    M. Anton Ertl Some things have to be seen to be believed
    anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
    http://www.complang.tuwien.ac.at/anton/home.html

  4. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    On Wed, 02 May 2007 09:37:39 +0100, Måns Rullgård wrote:

    > General Schvantzkoph writes:
    >
    >> I just tried the OnDemand speed governor on my Abit AB9 Pro with a
    >> Core2 E6700. When I built this system last fall I noticed that it would
    >> crash if the OnDemand governor was used so I switched to the
    >> Performance Governor which has been stable (this system has been up
    >> 24/7 for 5 months). I just tried the OnDemand governor again hoping
    >> that the 2.6.20.x kernel had fixed it, it didn't. My system crashed
    >> within two hours. It appears that this is an IO system problem,
    >> unfortunately there is nothing in /var/log/messages and I didn't take a
    >> picture of the error messages on the console.
    >>
    >> Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    >> Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    >> problem.

    >
    > I'm getting total lockups with the ondemand governor with Core2 on an
    > nvidia 680i board. I haven't been able to determine any exact cause of
    > the problem as it will suddenly freeze the machine completely without
    > leaving any trace of what happened.
    >
    > I've disabled speedstep entirely instead since it didn't make much of a
    > difference to the CPU temperature anyway.


    That's interesting, you have a completely different chipset on that board
    so it points to a problem with the Core2 rather than the chipset. I'm
    running my system as a server so I'm logged in remotely rather than
    locally. When no one is logged in locally the local display acts as a
    console. When my machine locks up I see messages on the console about
    lost interrupts.

  5. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    General Schvantzkoph writes:

    > On Wed, 02 May 2007 09:37:39 +0100, Måns Rullgård wrote:
    >
    >> General Schvantzkoph writes:
    >>
    >>> I just tried the OnDemand speed governor on my Abit AB9 Pro with a
    >>> Core2 E6700. When I built this system last fall I noticed that it would
    >>> crash if the OnDemand governor was used so I switched to the
    >>> Performance Governor which has been stable (this system has been up
    >>> 24/7 for 5 months). I just tried the OnDemand governor again hoping
    >>> that the 2.6.20.x kernel had fixed it, it didn't. My system crashed
    >>> within two hours. It appears that this is an IO system problem,
    >>> unfortunately there is nothing in /var/log/messages and I didn't take a
    >>> picture of the error messages on the console.
    >>>
    >>> Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    >>> Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    >>> problem.

    >>
    >> I'm getting total lockups with the ondemand governor with Core2 on an
    >> nvidia 680i board. I haven't been able to determine any exact cause of
    >> the problem as it will suddenly freeze the machine completely without
    >> leaving any trace of what happened.
    >>
    >> I've disabled speedstep entirely instead since it didn't make much of a
    >> difference to the CPU temperature anyway.

    >
    > That's interesting, you have a completely different chipset on that board
    > so it points to a problem with the Core2 rather than the chipset. I'm
    > running my system as a server so I'm logged in remotely rather than
    > locally. When no one is logged in locally the local display acts as a
    > console. When my machine locks up I see messages on the console about
    > lost interrupts.


    I've tried running with a serial console but it gave nothing.
    Unfortunately I need the machine for real stuff so I can't run a whole
    lot of tests.

    --
    Måns Rullgård
    mans@mansr.com

  6. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    anton@mips.complang.tuwien.ac.at (Anton Ertl) writes:

    > General Schvantzkoph writes:
    >>Is anyone using OnDemand with a Core2 Duo?

    >
    > Yes, a friend of mine is using it on a machine with an E6600 and a 945
    > chipset. It runs stable, as far as we have noticed (the machine is
    > not used much under Linux).


    In my case, the crashes invariably occur when the machine is somewhat
    busy. It's happened several times when compiling things and quite a
    few times while playing videos, possibly connected with simultaneous
    light disk I/O. Never has it crashed while only idling.

    --
    Måns Rullgård
    mans@mansr.com

  7. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    General Schvantzkoph writes:
    >Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    >Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    >problem.


    I have just played around a bit with a Xeon 3070 (which is a different
    name for the Core 2 Duo E6700). I have not used the ondemand
    governor, but I have measured the power consumption at different
    speeds:

    clock idle load 1 load 2
    1600MHz 103W 112W 122W
    2133MHz 104W 120W 136W
    2666MHz 104W 133W 156W

    As you can see, the idle power is the same at all speeds, so there is
    little point in using the ondemand governor, just use performance
    (well, maybe ondemand runs a little cooler if you run xine or
    something else that produces a noticable CPU load, but not enough for
    ondemand to raise the speed to the maximum).

    The other thing that is probably interesting to you is:

    # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
    1000000 800000 600000
    # cat /proc/cpuinfo
    processor : 0
    vendor_id : GenuineIntel
    cpu family : 6
    model : 15
    model name : Intel(R) Xeon(R) CPU 3070 @ 2.66GHz
    stepping : 6
    cpu MHz : 1000.000
    ....
    bogomips : 5323.99
    ....

    So, having the wrong speeds in scaling_available_frequencies is not
    just an Abit problem. And getting the speed wrong in the "cpu MHz"
    was pretty shocking to me; even my first Linux box (a 486-66) got that
    right.

    The kernel is a Debian Etch 2.6.18 kernel with the perfctr patch
    applied.

    The hardware is: Xeon 3070 (=Core2 Duo E6700, Socket 775, 2.66GHz, 4MB
    L2), Supermicro PDSME+ (Intel E7230 chipset), 8GB DDR2 RAM, 2 320GB
    SATA hard disks, 1 DVD-ROM, 1 floppy drive, Supermicro case with
    Ablecom SP645-PS 645W power supply.

    - anton
    --
    M. Anton Ertl Some things have to be seen to be believed
    anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
    http://www.complang.tuwien.ac.at/anton/home.html

  8. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    On Fri, 04 May 2007 16:49:51 +0000, Anton Ertl wrote:

    > General Schvantzkoph writes:
    >>Is anyone using OnDemand with a Core2 Duo? How about with an Abit AB9
    >>Pro? I'm wondering if this is an Abit problem or a Core2/965 chipset
    >>problem.

    >
    > I have just played around a bit with a Xeon 3070 (which is a different
    > name for the Core 2 Duo E6700). I have not used the ondemand governor,
    > but I have measured the power consumption at different speeds:
    >
    > clock idle load 1 load 2
    > 1600MHz 103W 112W 122W
    > 2133MHz 104W 120W 136W
    > 2666MHz 104W 133W 156W
    >
    > As you can see, the idle power is the same at all speeds, so there is
    > little point in using the ondemand governor, just use performance (well,
    > maybe ondemand runs a little cooler if you run xine or something else
    > that produces a noticable CPU load, but not enough for ondemand to raise
    > the speed to the maximum).
    >
    > The other thing that is probably interesting to you is:
    >
    > # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
    > 1000000 800000 600000
    > # cat /proc/cpuinfo
    > processor : 0
    > vendor_id : GenuineIntel
    > cpu family : 6
    > model : 15
    > model name : Intel(R) Xeon(R) CPU 3070 @ 2.66GHz
    > stepping : 6
    > cpu MHz : 1000.000
    > ...
    > bogomips : 5323.99
    > ...
    >
    > So, having the wrong speeds in scaling_available_frequencies is not just
    > an Abit problem. And getting the speed wrong in the "cpu MHz" was
    > pretty shocking to me; even my first Linux box (a 486-66) got that
    > right.
    >
    > The kernel is a Debian Etch 2.6.18 kernel with the perfctr patch
    > applied.
    >
    > The hardware is: Xeon 3070 (=Core2 Duo E6700, Socket 775, 2.66GHz, 4MB
    > L2), Supermicro PDSME+ (Intel E7230 chipset), 8GB DDR2 RAM, 2 320GB SATA
    > hard disks, 1 DVD-ROM, 1 floppy drive, Supermicro case with Ablecom
    > SP645-PS 645W power supply.
    >
    > - anton


    I'm really surprised that a Supermicro motherboard exhibits the CPU
    frequency reporting problem. If the clock speed is set correctly on the
    Abit board then the clock speeds are reported accurately, if the board is
    overclocked they aren't. Abit overclocks their boards be default, I find
    it very hard to believe that Supermicro would engage in a practice like
    that. Supermicro is in the server market where the top concern is
    reliability, running the CPUs out of spec is something that I would
    damage their reputation.


    Your figures on power consumption are interesting. What they are saying
    is that Intel has done a good job of shutting off the unused portion of
    the core. In an idle loop the CPU is using such a small percentage of the
    circuitry that changing the clock rate doesn't produce a measurable
    effect. So in a system with adequate cooling there is no reason to use
    anything but the Performance governor. In systems without adequate
    cooling, I'm thinking laptops, you would want to use User Mode to control
    power not OnDemand anyway.


  9. Re: OnDemand Speed Governor crashes Core2 on Abit AB9Pro

    General Schvantzkoph writes:
    >On Fri, 04 May 2007 16:49:51 +0000, Anton Ertl wrote:
    >> # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
    >> 1000000 800000 600000
    >> # cat /proc/cpuinfo

    ....
    >> cpu MHz : 1000.000
    >> ...
    >> bogomips : 5323.99
    >> ...

    ....
    >I'm really surprised that a Supermicro motherboard exhibits the CPU
    >frequency reporting problem. If the clock speed is set correctly on the
    >Abit board then the clock speeds are reported accurately, if the board is
    >overclocked they aren't. Abit overclocks their boards be default, I find
    >it very hard to believe that Supermicro would engage in a practice like
    >that. Supermicro is in the server market where the top concern is
    >reliability, running the CPUs out of spec is something that I would
    >damage their reputation.


    Looking at the bogomips, I doubt that the CPU is overclocked.

    >Your figures on power consumption are interesting. What they are saying
    >is that Intel has done a good job of shutting off the unused portion of
    >the core. In an idle loop the CPU is using such a small percentage of the
    >circuitry that changing the clock rate doesn't produce a measurable
    >effect.


    Yes, and either they are not reducing voltage, or they switch off so
    much in idle state that the voltage reduction also has little effect
    on power in idle state.

    > So in a system with adequate cooling there is no reason to use
    >anything but the Performance governor.


    And then I see no reason to use any of the cpufreq stuff.

    - anton
    --
    M. Anton Ertl Some things have to be seen to be believed
    anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
    http://www.complang.tuwien.ac.at/anton/home.html

+ Reply to Thread