Abit temperature readings on P4 motherboards

Discussion in 'Abit' started by Bill Drake, Oct 3, 2003.

  1. Bill Drake

    Bill Drake Guest

    The following is a copy of a post I made to the Abit forum
    on October 2, 2003:

    Topic: "Intel Chipset Socket 478 Motherboards"

    Thread: "Official Update for the P4 motherboards temperature issue"

    PostID: http://forum.abit-usa.com/showthread.php?s=&postid=171232


    Text as follows:


    Abit have used the W83627HF temperature monitoring chip for
    the entire P4 motherboard range. The same chip and the same
    design is used all the way from the BE7 to the IC7. The design
    and specification for this item is actually a carryover from the
    BE6-2.

    The entire P4 range of Abit motherboard models using the above
    hardware monitor chip has exhibited the approx 10degC offset.
    Other motherboard manufacturers using the *same* chipset do
    not exhibit the offset. This confirms the problem is not the chip
    or the hardware design.


    People who have installed outboard thermal sensors near the
    CPU-HSF junction to monitor that temperature (including myself)
    have all shown similar temperatures when monitoring the heatsink
    base. These temperatures have consistently shown a 10-15 degC
    offset in comparison to the on-die reading.


    Reading this thread, the following observations are unavoidable:

    1. Users with high-grade heatsinks (such as the Zalman 7000Cu
    or its equivalent) are solving the shutdown-on-overheat problem
    by reducing their CPU *peak* temperatures below the stock
    overheat-shutdown trigger-point through the use of high-grade
    cooling.

    2. Users with stock heatsinks installed in marginal cases are running
    into the shutdown-on-overheat problem because the on-die
    *reported* temperature is exceeding their shutdown-on-overheat
    setting in their BIOS -- even though the true heat-output of the
    CPU is still in the safe-to-operate range.



    The above forces the following conclusions:

    1. It is *possible* that because the P4 heatsink has an integrated
    heat-spreader, there is a higher delta-T between the on-die
    thermal diode inside the CPU silicon itself and the centre of the
    IHS heat-transfer surface where Intel mandate the temperature
    of the CPU should be measured for the overheat-shutdown
    trigger point.

    Note: The delta-T on P3 CPUs between the on-die thermal diode
    and the heatsink itself was 2degC to 4degC on a properly
    applied heatsink. This has been confirmed by extensive
    measurement by myself and others.

    Many of us are *expecting* the P4 to have a similar
    delta-T between the on-die thermal-diode and the
    heatsink -- matching that found on the P3. This is to be
    expected from long-standing Intel practice.
    However, THIS MAY NOT BE THE CASE.

    If the new heat-spreader design creates a higher delta-T
    between the on-die thermal diode and the centre of the
    IHS, then the temperatures we are reading are correct
    FOR A P3-STYLE FLIP-CHIP MOUNT. Intel's own
    documentation does not make a distinction here -- nor
    is there any clarification in the Intel documentation to
    confirm whether there is a difference in the thermal path
    resistance between P3 and P4 CPU mounts. We need Intel
    to confirm this one way or the other.

    2. It is a known fact that the current BIOS algorithm Abit are using
    for the on-die thermal diode is the same one used for all their P3
    motherboards.

    If there is a known delta-T for the
    on-die-thermal-diode-to-IHS-centerpoint -- then there is no
    doubt that a new thermal calculation algorithm is *required* to
    account for the thermal resistance in the
    on-die-thermal-diode-to-IHS-centerpoint pathway. Modifying
    the algorithm is the *only* valid way to get a temperature
    reading in sync with Intel's published specifications.


    Final comments:

    I suspect that both Intel and the other motherboard manufacturers
    who are showing on-die CPU temperatures in sync with Intel's
    published recommendations have implemented a new algorithm to
    accommodate the above. It would behoove Abit to follow suit.




    Best I can do for now. <tm>


    Bill
     
    Bill Drake, Oct 3, 2003
    #1
    1. Advertisements

  2. Bill Drake

    David P. Guest

    It seems to me that people who want/need a highly accurate thermometer can
    find one better, cheaper and easier to use than a PC motherboard.

    I'm satisfied that Abit motherboards read high. Since my reported highest
    temps are well below the 75 degree maximum operating temperature for my
    mildly overclocked Intel 2.8C, I'm confident I'm not going to hurt the CPU
    running [email protected] 24/7. What would I gain be having the temperatures
    reported more accurately?

    I think the more interesting obsessions are those of the extreme
    overclockers. :)

    David P.
     
    David P., Oct 4, 2003
    #2
    1. Advertisements

  3. Bill Drake

    Spajky Guest

    nothing ... :)
    mostly true ... :)

    -- Regards, SPAJKY
    & visit - http://www.spajky.iscyber.com
    Celly-III OC-ed,"Tualatin on BX-Slot1-MoBo!"
    E-mail AntiSpam: remove ##
     
    Spajky, Oct 4, 2003
    #3
  4. Bill Drake

    Spajky Guest

    right!
    no, it is a thru Bios programming a HWmon chip .. !
    that is normal !!!!
    thats normal, worse AirFlow (for more silent operation) rises temps !
    Yes ...
    Yes it is possible, same is with Tuallies ... there is a thin thermal
    pad between an unprotected core & IHS & depents how everything is
    tighted together (thats why some people took off that IHS more or less
    successfuly ... :) to gain some °C at edge OC ...
    .... is more, is more at least double! Depends how you have done
    measurements; See mine experiments with temperatures @
    http://freeweb.siol.net/jerman55/HP/Spajky.htm
    it should be quite matching (I tried with a Celeron Cu-Mine) as p3 has
    its core protected with that blue stuff if you do not "lap" it to gain
    something ...
    like I have understood in the past (incl.Tualatins, P4 should not be
    exeption), Intel gives the temperatures NOT from a die, (like also AMD
    & others) but from the bottom of the HSF closest to core (practically
    measured IHS temperature) & that temps (max.) are long term full load
    recommended not to exceed! (temps from on die are approx. 10°C higher
    & short terms w/o problem peaks even more) .
    the main algorithm is the same, but with some changing with different
    generations of CPUs, the on die diode with same chip & sensing
    algorithm WILL show different temps for a 0,25u Cpu & different for
    0,13u Cpu ....
    yes ... also ...
    temps for server boards are accurate, but mostly for consumer market
    is that not so ....

    -- Regards, SPAJKY
    & visit - http://www.spajky.iscyber.com
    Celly-III OC-ed,"Tualatin on BX-Slot1-MoBo!"
    E-mail AntiSpam: remove ##
     
    Spajky, Oct 4, 2003
    #4
  5. Bill Drake

    Skid Guest

    <big snip>

    Bill,

    Since you posted there first, I'm sure that you read the lengthy thread in
    the Abit forum about temp readings on Abit's Springdale and Canterwood
    boards.

    Abit's response could be paraphrased as:

    Our engineers use the algorithm provided by Winbond, the company that
    provides the temperature monitoring hardware and software. It's all correct.
    Turn the alarm sensor up to 85C so it won't go off. Don't worry, be happy.

    Most of us took that for what it was worth, an acknowledgement that Abit
    boards were going to read 10C higher than others, but that didn't mean the
    reported temps were dangerous.

    A better explanation would be nice, a fix would be better. I doubt we're
    going to see either, but if it makes you feel better, blast away.
     
    Skid, Oct 4, 2003
    #5
  6. Bill Drake

    Bill Drake Guest

    Hi, Skid. If you look back through that thread, you will find an older
    message from me detailing my findings at that time. The thread
    was dissed by a bunch of fanboys who tried (unsuccessfully) to
    counter the reality of the data.

    Since then, a *large* number of posts have come in confirming
    typical Load temps of 60degC -- even with high-grade cooling. In
    each case, an on-die reading of around 60degC correlated with a
    HSF temp of around 48 degC.

    The other observable fact was that some users put the very same
    CPU/HSF combo in other motherboards and got temps in line with
    Intel recommendations and Industry Standard observations.


    With the above in hand, it is inescapable that the algorithm is
    incorrect.


    (By the way, I was the one who *developed* the algorithm
    info used in MBM for the Abit BE7/BH7. This info was
    used as the basis for the IC7. If you read the motherboard
    monitor Voltage.ini file, you will see that the info for the IC7
    is an *exact copy* of my info for the BE7. That is correct,
    as the monitoring chipset implementation is identical between
    the two boards -- all that changes is the sensor locations.)


    Conclusion:

    It's broken, and I'm not about to let Abit off the hook on this issue.



    Best I can do for now. <tm>


    Bill
     
    Bill Drake, Oct 5, 2003
    #6
  7. Bill Drake

    Skid Guest

    I'm no expert, but I could tell from your post you knew what you were
    talking about.

    I can, however, read and think clearly enough to figure out that my OC 2.4 @
    3.3g is just fine idling at 45-47 and topping out at 60C under torture
    testing on my IC7. That's with a Vantec Aeroflow. It was 10C higher under
    load with the stock hsf, which is why I replaced it.

    For my purposes, it doesn't matter what the number is, all I need to know is
    that it's where it ought to be. IMHO, temp readings, like benchmarks, are
    really only good for comparing changes inside the same box. There are too
    many variables for rational comparisons between one system and another --
    which is why most "is my cpu too hot?" threads dissolve into chaos and bad
    advice.

    Fight the good fight, and let us all know if you wake anybody up in Taiwan
    ;>)
     
    Skid, Oct 5, 2003
    #7
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.