The Notebook Review forums were hosted by TechTarget, who shut down them down on January 31, 2022. This static read-only archive was pulled by NBR forum users between January 20 and January 31, 2022, in an effort to make sure that the valuable technical information that had been posted on the forums is preserved. For current discussions, many NBR forum users moved over to NotebookTalk.net after the shutdown.
Problems? See this thread at archive.org.

    Long Term Overvolting/Overclocking Stability

    Discussion in 'Gaming (Software and Graphics Cards)' started by Robbo99999, Oct 3, 2013.

  1. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,346
    Messages:
    6,824
    Likes Received:
    6,112
    Trophy Points:
    681
    I've been doing some extended long term testing in benchmarks & games with different overvolts & overclocks to work out a range of long term stable overvolt/overclock settings, and I thought you guys would be interested in the results because it might help you work out your own stable settings. (Testing done on my GTX 670MX in my sig).

    First thing I noticed is that passing through a 3DMark11 benchmark run sucessfully was not a measure of long term stability. In fact long term stability was usually gained by dialing back core clock by one notch (13Mhz) after achieving a successful 3DMark11 run. Second thing, there seems to be a direct linear relationship between volts & stable overclock frequency; if you take a look at the attached picture you will see that the graph is linear, so that could be useful for you to use to help dial in your own further overvolts/overclocks. (You won't be able to use the exact same settings, because everyone's GPU will behave a little different, even if it's the same model, but knowing that it's a linear relationship could help you work out further overvolts & overclocks).

    Hope anyone that reads this will find this useful or interesting!

    EDIT: I was kind of flabbergasted at how linear & exact this relationship was!
     

    Attached Files:

  2. sponge_gto

    sponge_gto Notebook Deity

    Reputations:
    885
    Messages:
    872
    Likes Received:
    307
    Trophy Points:
    76
    That linear variation is going to continue until you reach the region of diminishing returns, where bumping the voltage will have little effect on your max. clocks. The end of the linear region can be used as a rough "maximum" OC for most intents and purposes.
     
    Robbo99999 likes this.
  3. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,346
    Messages:
    6,824
    Likes Received:
    6,112
    Trophy Points:
    681
    Hi, thanks for the info, that's useful to know, if I feel brave I may push it further sometime, but I'd need to have the voltage & Mhz sliders increased in NVInspector to get any more. svl7 might be able to do that for me somehow, but I'll think I'll wait a while to do that - I want to get some more useful life out of my GPU before I risk upping the voltage beyond 1.05V.

    At what kind of voltage would you expect those diminishing returns to start kicking in?
     
  4. sponge_gto

    sponge_gto Notebook Deity

    Reputations:
    885
    Messages:
    872
    Likes Received:
    307
    Trophy Points:
    76
    Wish I knew better :p I was actually impressed that you got good frequency gains from 0.925V all the way to 1.05V. From how I understand it, the actual physical relationship between voltage and max frequency is most definitely non-linear however within a a range of "small" changes, the behavior can always be approximated as linear (recall linearization techniques used for stability analysis, if one's familiar with such stuff ;)). It is reasonable to expect predictable behavior within that range i.e. gradual increase in temperature/power usage with voltage.

    The huge OC/OV headroom one finds in a 670MX could be due to its artificially limited stock voltage which was intended to fit the thermal/power constraints of mobile devices. The same core used in desktop cards are designed to operate with much higher voltages and so the MX doesn't show any diminishing returns for the amount of OV allowed by NVInspector.

    Many other cards, for e.g. my dear old 7970M's, do not scale well with voltage. For the life of me I could not get improvement in OC with any amount of OV :eek:

    PS: the 670MX is 600MHz at stock? Oh boy that's some insane value for money with your OC :thumbsup:
     
  5. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    4,346
    Messages:
    6,824
    Likes Received:
    6,112
    Trophy Points:
    681
    Yes, I agree with you, it was my conclusion too that the 670MX had been artificially limited by NVidia, both to allow for good thermals, but also so that the card didn't encroach too far on the 680M and then the future 770M. There's a lot of untapped potential in those cards.

    Yes, the same core, when used in desktops has a limit of 1.175V as a boost voltage when the card is under 70degC. That makes me think that my card could be safe up to 1.175V if I keep it below 70 degC, provided I don't go over 100W current consumption for the card. (it doesn't go over 65 degC at the moment). Even with my maximum overvolt & overclock it only uses up to 85W (worked out from total consumption minus CPU & other overheads). But, I am reluctant to push it further for now, as I don't need to for gaming at 1600x900, so I might push it some more when games require it further down the line.

    Yep, the 670MX is stock clocks of 600Mhz, so my current max overclock of 1124Mhz is an 87% overcock! Not bad value for just over £200! I was lucky though, I nearly splashed out another £150 odd quid for a 7970M, but the company said it wasn't suitable for my laptop & recommended the 670MX at which point I had no idea it would overclock so well!
     
    sponge_gto likes this.