XFX HD7970 Temperature too hot?

edited February 2014 in Hardware support

Hello, I use GUIMiner (I know n00by and that) and have a K/hash rate of ~500. I have it overclock to GPU Clock 1125 MHz Memory Clock 6300 MHz 80-100 % Fan Speed and Power Target% 1. I get 74-77 Celsius. Is this too hot or good? Also how can I increase the K/HashRate and reduce Temperature  

Comments

  • Anything under 80C is good.

  • Thanks. I'll donate some LTC when I have 1 LTC.

  • @TheColdQuasar,



    While 80C won't hurt the electronics on the card,
    as edwardc said, running the fans that fast is likely to kill them
    within 6 months to a year.  I've had to send three cards back to the
    factory for just this reason.  You may wish to look at this thread.  I
    have a config file there which shows how I deal with heat.



    https://forum.give-me-coins.com/discussion/666/heat-death-how-to-kill-your-gpu-in-less-than-a-year#Item_9



    Separating cards with an air gap and allowing the fans to run slower is
    great.  However, if they're jammed next to each other in a conventional
    case, they cannot get enough air flow.  (PS A standard double width
    slot space is 1.6" based on my research.  Thinner cards are better.) 
    Since the fans are not designed for 24/7 100% operation, and probably
    have crummy sleeve bearings, they die quickly.  This essentially kills
    the card, unless you get a warranty repair done, or mount an after
    market cooler to it (which will void the warranty, if applicable, and
    generally increases the card's width to 3 slots rather than two).



    I have a
    theory, which I cannot prove, but which makes sense, that the fan life
    decreases exponentially with speed.  So, if the card is designed for
    normal gaming use a few hours a day, and even then, unless you're in
    crossfire mode with dual cards, the fans may not run too hard; then you
    can assume the card will last at least what it's warranted for, which in my case,
    is 3 years.  I personally killed some other cards in 6 months or so
    running at 100%.



    But, I don't think the line between 3 years and 6
    months is linear.  I have some corroboration for my theory in the
    readme for gpu mining in the cgminer directory.



    So, making an
    educated guess about the issue, I took a piece of graph paper and used a
    protractor to draw a circular line in the top right quadrant of the
    graph area, from top left going to the right and arcing down to the
    bottom right.  I'm assuming the card life curve is somewhat of this
    shape, an exponentially decreasing function.  It may be parabolic, rather than circular, but that's not the point.



    I label the vertical
    axis on the left starting at the bottom for card life starting at 6 months, going
    up to the top at 3 years.  I label the horizontal axis at the bottom starting at the left
    for fan speed starting at 20%, which is the minimum it would probably
    ever run normally, all the way to 100% at the right.



    As I said,
    this is an educated guess, but it is probably representative of
    reality.  So, tracing from various fan speeds on the bottom, up to the
    curve, and over to the left, you can estimate the card's lifetime. 
    Using this method, I estimate the card's lifetime at 97% fan speed to be
    12 months.  This number, in particular, could be WAY too optimistic.  I
    estimate the card's life at 82% fan speed to be 24 months.  I estimate
    the card's life at 67% fan speed to be about 30 months.  And, finally, I
    estimate the card's life at 20% fan speed to be it's full warranted
    life of 3 years or 36 months.  As you can see, increasing fan speed toward it's upper limits quickly kills the card.



    In the past, I was limiting the card's fan speed to 85%.  I am now
    limiting it to 65%.  This should only deprive minimal life from the
    unit, maybe 17% of it's life.  Remember, we're running 24/7.  If the
    card fails in the warranty period, I'll get it refurbed from the
    factory, then sell it, as I don't want to sell someone a used up piece
    of garbage.  And, I don't THINK you can make two warranty claims unless
    they missed something on the first refurb.



    So, doesn't that make the heat worse limiting fan speed to 65%?  Well,
    it would if I ran the card at the same level.  But I don't.  I
    sacrifice some clock speed, and a little bit of production, to allow
    the card to live longer before being replaced.



    For a typical ATI / AMD 7850 card, I set the gpu clock engine parameter
    to 300-1050 and turn on auto-gpu and auto-fan.  This will overclock the
    card IF IT'S RUNNING COOL ENOUGH.  However, if the card gets too hot,
    it will increase the fan speed, only up to 65%.  Once it's at 65%, it
    will start REDUCING the clock speed to lower the temperature while
    still saving the fans from excessively premature death.  Never limit
    the fan speed in this way without allowing the clock speed to reduce.



    As I look at my status display, I have 4 cards in a conventional
    computer case with a generous amount of case fans, including two
    CoolerMaster JetFlo fans.



    The clock speeds on the cards are: 1050, 1050, 920, 765.  This is
    working from the bottom of the case up toward the cpu.  All the cards
    are limited to 65% fan speed.  The first 2 (bottom) cards are happy,
    and are running in an overclock mode.  However, the 3rd card is
    overclocking only slightly, but not as much as the others.  The 4th
    card is not very happy, and is substantially underclocking.  But,
    they're all in production.  And, they're all in a SAFE zone for both
    temperature (maintaining about 82 - 86 degrees) and a SAFE zone for
    prolonging the fan life.  I think this is a good compromise.  IF I'm
    able to move all the cards to an open frame rack, and allow a couple of
    inches of air gap between them, I would probably be able to run them
    all in overclock mode and STILL keep the fan speed low.



    Regarding guiminer, I used that for a while.  It starts a separate
    cgminer thread for each card.  I found out that this creates conflicts
    in controlling the clock speed and fan speed of all gpu's separately
    while still being coordinated.  Now, I've shifted to using one instance
    of cgminer per computer, started from a batch file, and controlled by
    the aforementioned config file.  That cgminer talks to one
    give-me-coins worker for the whole pc.  This works like a charm most of
    the time.  Every so often, things get wonky and I have to reboot.  But,
    that is rare.  I keep an eye on clock speeds, loads, temperatures, and
    fan speeds with 4 instances of GPUZ on Windows.  If I'm on Linux, I can
    monitor most of that with the aticonfig command.



    Hope this info helps.



    Sincerely,



    Ron


Sign In or Register to comment.